Book of Preprints

September 2008

INES Savoie Technolac, 50 avenue du Lac Lman, BP 332 73377 LE BOURGET DU LAC - FRANCE

INES Book Preprint - September 2008

INES Book of Preprints

INES is the main solar energy institute in France. It is located in Chambry, in the Rhne-Alpes region. More than 150 researchers from CEA, CNRS and Universit de Savoie are now working at INES, and others will join as soon as new buildings will be operational. Their mission is to carry out research activities and to develop innovative solutions to increase the use of solar energy. In addition to these R&D activities, almost 10 people are involved in training.

This book gathers the papers presented within this European photovoltaic conference in which INES researchers are authors or co-authors.

For more information : INES Tel : +33 4 79 44 45 46 Fax : +33 4 79 68 80 49 E-mail: catherine.maniglier@cea.fr

INES Book Preprint - September 2008

Table of Contents

SILICON SOLAR CELL TECHNOLOGY

17.8 % EFFICIENCY OBTAINED ON C-SI SOLAR CELLS USING A SELECTIVE EMITTER INDUSTRIAL TYPE PROCESS
Rmi Monna, Nicolas Enjalbert, Yannick Veschetti, Mickael Lozach and Marc Pirot

LIGHT-INDUCED DEGRADATION AND REGENERATION IN COMPENSATED UPGRADED METALLURGICAL SILICON

Sbastien Dubois1, Nicolas Enjalbert1, Jean Paul Garandet1, Rmi Monna1, Jed Kraiem2

BENEFICIAL EFFECTS OF DOPANT COMPENSATION ON CARRIER LIFETIME IN UPGRADED METALLURGICAL SILICON

Sbastien Dubois1, Nicolas Enjalbert1, Florence Servant1, Jean Paul Garandet1, Rmi Monna1, Jed. Kraiem2 MINORITY CARRIER LIFETIME AND DIFFUSION LENGTH IN VOLONTARILY CHROMIUM CONTAMINATED N-TYPE mc-Si WAFERS 1 Santo Martinuzzi, 2Sbastien Dubois, 2Nicolas Enjalbert,

SILICON HETEROJUNCTION SOLAR CELLS : SURFACE PASSIVATION QUALITY ON LARGE AREA N TYPE AND P TYPE MONOCRYSTALLINE SILICON
Pierre Jean Ribeyron1, Thibault Desrues, Aurlie Vandeneynde1, Florent Souche1, Anne Sophie Ozanne1, Martin Labrune2, Pere Roca i Cabarrocas2, Rmi Chouffot3, Jean Paul Kleider3 NEW PROCESS INTEGRATION FOR INTERDIGITATED BACK CONTACT (IBC) a-Si:H/c-Si HETEROJUNCTION SOLAR CELLS Thibault Desrues1,5, Pierre-Jean Ribeyron1, Aurlie Vandeneynde1, Anne Sophie Ozanne1, Florent Souche1, Yannick Veschetti1, Armand Bettinelli1, Pere Roca i Cabarrocas2, Martin Labrune2, Dicknoum Diouf3, Jean Paul Kleider3, Mustafa. Lemiti4

DEVELOPMENT OF ZNO:AL TO REPLACE INDIUM TIN OXYDE FOR HETEROJUNCTION SOLAR CELLS : EFFECT ON SURFACE PASSIVATION QUALITY AND SHORT CIRCUIT CURRENT
Pierre Jean Ribeyron2, Nathalie Baclet1, HenriBoucher1, Thibault Desrues2, Aurlie Vandeneynde2, Florent Souche2, Anne Sophie Ozanne2

FUNCTIONAL ALL-SILICON NANOWIRE SOLAR CELLS

Philippe Thony, Rgis Delsol, Claude Jaussaud, Nathalie Rondel, Pascal Gentile, Emmanuelle Rouvire, Sverine Poncet, Simon Perraud, Cline Mouchet, Pascal Gentile, Thierry Baron

MODULE CHARACTERIZATION
OUTDOOR EVALUATION OF THE ENERGY PRODUCTION OF DIFFERENT MODULE TECHNOLOGIES
Jens Merten, Lionel Sicot, Yves Delesse, Antoine Gurin de Montgareuil

APPLICATION OF THE MOTHERPV METHOD TO THE ACCURATE MONITORING OF THE GRID-CONNECTED PHOTOVOLTAIC SYSTEMS
Antoine Gurin de Montgareuil *, Frdric Mezzasalma *, Jens Merten **

INES Book Preprint - September 2008

PV SYSTEMS AND DISTRIBUTED GENERATION OF ELECTRICITY

SHADOWING EFFECTS ON GRID CONNECTED PV SYSTEMS
N. Chaintreuil, F. Barruel, Xavier Le Pivert, Herv Buttin, Jens Merten

DESIGN OF A MANAGEMENT AND SIMULATIONS TOOL FOR SOLAR CAR PARK

Thang Vu, Frank Barruel, Jens Merten

PRISMES: THE INES MICROGRID PLATFORM

Frank Barruel, Herv Buttin, Olivier Wiss, Sbastien Grhant, Jens Merten

SYSTEM MODELLING AND ENERGY MANAGEMENT FOR GRID CONNECTED PV SYSTEMS WITH STORAGE
Yann Riffonneau*, Arnaud Delaille*, Franck Barruel*, S.Bacha

STATISTICAL APPROACH FOR THE GUARANTEE OF PHOTOVOLTAIC ENERGY OUTPUT

A. Gurin de Montgareuil

STORAGE SYSTEMS
CHARACTERIZATION OF ADVANCED NICKEL-CADMIUM BATTERIES DEVELOPED FOR PHOTOVOLTAIC APPLICATIONS
Arnaud Delaillea, Florence Matteraa, ke Nystrmb, Charles Strebb

ASSESSMENT OF STORAGE AGEING IN DIFFERENT TYPES OF PV SYSTEMS: TECHNICAL AND ECONOMICAL ASPECTS
Elisabeth Lemaire-Potteau, Florence Mattera, Arnaud Delaille, Philippe Malbranche

ORGANIC SOLAR CELLS

HIGH-EFFICIENCY LARGE AREA FLEXIBLE ORGANIC SOLAR CELLS
Nolla Lemaitre, Rmi de Bettignies*, Sverine Bailly, Pascal Maisse, Stphane Cros, Stphane Guillerez

RELATIONSHIP BETWEEN ENCAPSULATION BARRIER PERFORMANCE AND ORGANIC SOLAR CELL LIFETIME
Stphane Cros, Stphane Guillerez, Rmi de Bettignies, Nolla Lematre, Severine Bailly, Pascal Maisse

17.8 % EFFICIENCY OBTAINED ON C-SI SOLAR CELLS USING A SELECTIVE EMITTER INDUSTRIAL TYPE PROCESS

Rmi Monna, Nicolas Enjalbert, Yannick Veschetti, Mickael Lozach, Marc Pirot INES CEA/LITEN/Laboratory of Solar Components (LCS) BP 332 Savoie Technolac, 50 avenue du Lac Lman 73377 Le Bourget-du-Lac, France Phone: +33 4 79 44 46 74 E-mail: remi.monna@cea.fr

ABSTRACT The formation of selective emitters is a key issue to achieve higher efficiencies on industrial screen printed silicon solar cells. A study on the development of selective emitter concepts is presented using POCl3 diffusion for the development of the high and low doped emitters. The work is focused on the application of the developed technology to an industrial level, considering the alignment requirements between the highly doped emitter pattern and the front silver grid. Our selective emitter process shows a great efficiency improvement of 0.5% absolute which is related to an improved spectral response in the short wavelength range. Efficiencies of 16.6% are obtained on 150 mm x 150 mm mc-Si cells and 17.8% on 125 mm x 125 mm c-Si cells. Keywords: selective emitter, screen printing, surface passivation, emitter current saturation

INTRODUCTION

Screen-printing is widely used in the photovoltaic industry to elaborate the metallization, having the advantage of a relative low cost compared to vacuum technologies. One of the disadvantages of this technique is the need of a high phosphorous surface concentration to obtain low metal-semiconductor contact resistance and to limit the shunts in the emitter [1]. The fabrication of selective emitters is one possible way to combine the advantages of thin and thick emitters [2]. It offers good blue response by enhanced front surface passivation due to low surface phosphorous concentration [3], high open circuit voltage due to lower emitter saturation current density [4] and reduced silicon/metal contact resistance due to a localized high doped region. In this study selective emitters are created by two high temperature steps to have a better emitter control. We evaluate different doping techniques giving controlled phosphorous profiles to obtain a passivated emitter surface with low saturation currents. Moreover, the industrial development of selective emitters depends on the availability of an easy and reproducible alignment process. To this end, studies were done to develop a technique able to achieve a reproducible alignment between the highly doped emitter pattern and the front silver grid pattern.

alignment marks are made by laser for the future alignment between the metallization and the heavily doped areas. Saw damage is removed and surface is textured by an alkaline wet etching using a low concentrated KOH solution. Alignment marks formation Saw damage removal and KOH texturing Selective emitter formation SiNx S-ARC deposition Front and back metallization screen printing Co-firing in a belt furnace Laser edge isolation Figure 1: Cell process description

Silicon

1. Diffusion barrier deposition 2. Local removal of the diffusion barrier by screen printed etching paste 3. Highly doped emitter formation by POCl3 diffusion 4. Diffusion barrier HF etch 5. Lightly doped emitter formation

Silicon

EXPERIMENTS

Silicon

2.1 Solar Cell Processing In our experiment, The starting silicon materials are p-type 150 mm x 150 mm mc-Si and p-type 125 mm x 125 mm c-Si with a nominal resistivity of = 0.5 to 2 .cm. The investigated process described below in figure 1 and figure 2 requires five additional steps related to the standard homogeneous emitter process. Two symmetric

Silicon

Silicon

Figure 2: Selective emitter steps description The PECVD deposited diffusion barrier is locally

etched by the use of a screen printing etching paste so that the heavily doped emitter regions can be created by POCl3 diffusion according to the front silver pattern. A lightly diffused region is created by POCl3 diffusion after the diffusion barrier has been totally removed. After a HF dip, a silicon nitride layer is deposited as a single antireflecting coating. Silver metallization is screen printed on the front side and full aluminium on the backside of the cells. For the cells with selective emitter, the front contact screen is aligned with the heavy diffused areas before printing. The distance between fingers on the cells with selective emitter was reduced from 2.5 mm to 2 mm to avoid additional series resistance. After co-firing in an IR belt furnace, laser edge isolation is performed. The reference cells receive a standard industrial type emitter with a sheet resistance of 50 /.

Table 1: Average cell efficiencies on homogeneous and selective emitters 125 mm x 125 mm mono-crystalline silicon solar cells. Average of 20 cells per lot. * Independently confirmed (FhG-ISE CalLab) Voc Jsc FF Eff Emitter (mV) (mA/cm) (%) (%) Homogeneous 620 35 78 16.9 emitter Selective emitter / process 1 Selective emitter / process 2 Selective emitter / best cell 627 35,8 76.8 17.3

632.6

35,6

78.1

17.6

630

36,3

77.6

17.8*

2.2 Alignment for industrial selective emitter solar cells production. An accurate alignment between the heavily doped areas and the front metallization lines is key for the fabrication of efficient selective emitter screen-printed solar cells. Figure 3 shows the alignment between the high doped regions and the metallization. A reproducible alignment is obtained between the 300 m width diffused region and the 120 m width metallization lines. Further studies are underway to test the stability of this alignment after thousand prints in industrial conditions.

Table 2: Average cell efficiencies on homogeneous and selective emitters 150 mm x 150 mm multi-crystalline silicon solar cells. Average of 20 cells per lot. * Independently confirmed (FhG-ISE CalLab) Voc Jsc FF Eff Emitter (mV) (mA/cm) (%) (%) Homogeneous 610 33.2 77.3 15.7 emitter Selective 620 33.5 78.4 16.3 emitter Selective emitter / best 624.4 34.2 77.6 16.6* cell With the selective emitter structure we observe an increase of 0.5 % of the average efficiency compared to the homogeneous emitter structure. On the best cells, an efficiency of 16.6 % is obtained on 150 mm x 150 mm mc-Si cells and 17.8 % on 125 mm x 125 mm c-Si cells.

Figure 3 : Alignment between the high doped region and the metallization on one finger extremity

100

RESULTS AND DISCUSSION

IQE (%)

90 80 70 60 50 40 30 20 10 0 200 400 600 800 1000 1200

3.1 Cells electrical results The results obtained on the 125 mm x 125 mm monocrystalline silicon solar cells are presented in Table 1 and in Table 2 for the results obtained on the 150 mm x 150 mm multi-crystalline silicon solar cells. An improvement of about 10 mV on the open circuit voltage (Voc) and of 0.3 to 0.6 mA/cm on the short circuit current (Jsc) is obtained with the selective emitter structure compared to the homogenous emitter. The increase of Voc is due to the decrease of the emitter current saturation (Joe) due to lower Auger recombination. The increase of the current in due to a better blue response as shown in the Internal Quantum Efficiency graphic represented in figure 4.

Homogeneous emitter Selective emitter

Wavelength (nm)

Figure 4: Comparison of the spectral response of homogeneous and selective emitter solar cells. mc-Si 150 mm x 150 mm cells

4 LIGHTLY DOPED EMITTER BY SPIN ON DOPED (SOD) DIFFUSION Phosphorous doping pastes present an interesting alternative to conventional POCl3 doping source. They offer the possibility to control the phosphorous profile in order to reduce the emitter saturation current with adequate surface passivation. Selective emitter cells were done using SOD for the lightly diffused emitter instead of POCl3. Results of LBIC measurements made at 406, 862, 946 and 976 nm are show in table 3. We observe an improvement of the photo-generated current at short wavelength on the solar cells with SOD selective emitters due to an optimized emitter profile which give less recombination. Table 3: LBIC measurement on two kinds of selective emitter solar cells compared to the POCl3 homogeneous emitter solar cell POCl3 POCl3/SOD Standard Current selective selective process emitter emitter I406nm 67 75 83.6 (A) I862nm 82 80.5 81 (A) I946nm 73.7 72 72 (A) I976nm 77 75.3 75 (A) LBIC mappings at 406 nm represented in figures 5, 6, and 7 are showing a great improvement of the photocurrent generated at short wavelength on the selective emitter solar cells. Heavily doped areas are clearly underlined around bus and fingers by a reduction of the signal intensity.

Figure 7 : LBIC measurement mapping at 406nm of a POCl3/SOD selective emitter solar cell. Scale is from 31A (red) to 101A (dark blue) The solar cells using SOD films to realize the lightly doped emitter present lower efficiencies due to poor fill factor values (see table 4 below).

Table 4: Comparison of the lightly doped emitter process (POCl3 and SOD). Average cell efficiencies on selective emitters 150 mm x 150 mm multi-crystalline silicon solar cells. Average of 20 cells per lot. Lightly doped Voc Jsc FF Eff emitter (mV) (mA/cm) (%) (%) POCl3 SOD 619 616 33.4 33.2 77.9 70.2 16.1 14.4

This is probably due to the negative effects of impurities initially present within the SOD film [5]. More investigations are currently done to fully understand this effect and to expect further electrical improvements. Nevertheless, the use of SOD for the emitter formation remains interesting to increase the gain on this technology, due to the possibility to develop a more optimised doping profile and to obtain a better surface passivation. 5 ENCAPSULATION OF SELECTIVE EMITTER SOLAR CELLS The efficiency enhancement on the selective emitter solar cells is mainly due to current gain in the blue region (300-500 nm). One may ask whether this gain could be lost after cell encapsulation due to the absorption of the EVA-glass couple. To evaluate the gain after encapsulation, Glass-EVA mini-modules were made by Photowatt Technologies using solar cells with homogeneous and selective emitter structures. The gain of short circuit current after encapsulation in shown in table 5 for both kind of emitter structures

Figure 5 : LBIC measurement mapping at 406nm of homogeneous 50 / emitter solar cell. Scale is from 31A (red) to 101A (dark blue)

Figure 6 : LBIC measurement mapping at 406nm of a POCl3 selective emitter solar cell. Scale is from 31A (red) to 101A (dark blue)

Table 5: Average short circuit current of selective and homogeneous emitter solar cells before and after glassEVA encapsulation. Homogeneous Selective emitter emitter Jsc Cell before 32.5 mA/cm 33.5 mA/cm encapsulation Jsc Cell after 34 mA/cm 35 mA/cm encapsulation Gain after 4.5 % 4.3 % encapsulation The short circuit current gain observed after encapsulation on the selective emitter solar cells is similar to the gain observed on the homogeneous emitter cells. With regards to the obtained results, we can conclude that the selective emitter cell technology could be applied on standard encapsulation line without module technology modification and without efficiency loss.

ACKNOWLEDGEMENTS

Part of this work was funded by the French Agency for Environment and energy management (Ademe) in the REDUCOP project. The authors want to acknowledge Photowatt Technologies for the cell encapsulation and for helpful discussions.

REFERENCES

COST ANALYSIS

Yield loss on solar cell fabrication lines plays an important role, dramatically increasing costs if an additional step in included in a production process. It was clearly demonstrated that for an investment of 2 M and annual costs of 500.000 with a process step that causes a yield loss of 2%, the additional technology has to deliver an efficiency gain of more than 0.6 % [6]. For the moment, the additional cost of our selective emitter process is too high to make profitable a transfer into a standard production line. A reduction of the number of process steps has clearly to be done. The development of partial diffusion barrier in order to perform selective emitter with a single diffusion step and the replacement of the etching paste by a Laser ablation process step are currently under investigation.

[1] V.Lowe and A.Day. High-temperature-stable contact metallization for silicon solar cells. IEEE Trans. Electron Device, vol.ED-31, p. 626 (May 1984) [2] K.Misiakos and F.Lindholm. Toward a systematic design theory for silicon solar cells using optimization techniques. Solar Cells, vol. 17, p.29 (1986). [3] J.Klaer and al. Improving blue response in RTP silicon solar cells by two-depth emitter. 13th EUPVSEC, Nice (1995). [4] M.Bhr and al. Surface passivation and contact resistance on various emitters of screen-printed crystalline solar cells. 19th EUPVSEC (2004). [5] Y.Veschetti and al. Influence of the emitter formation technique on the performance of silicon solar cells, 20th EUPVSEC, Dresden, 2006. [6] PJ.Rostan and al. New generation production technology for crystalline silicon solar cells, 22th EUPVSEC, Milan, 2007, p.1539

CONCLUSION

A selective emitter process was developed which can be used with industrial screen-printing technology. The process shows a great absolute efficiency improvement of 0.5%. This efficiency gain is related to a more suitable emitter profile which leads to an improved spectral response in the short wavelength range and to a better front surface passivation. Further optimizations of the front silver pattern to reduce shadowing loss could lead to more efficiency improvement. Best cell efficiencies of 16.6% are obtained on 150 mm x 150 mm mc-Si cells and 17.8% on 125 mm x 125 mm c-Si cells. However five additional steps are needed for the realization of the selective emitter structure. To decrease the number of steps, the development of partial diffusion barrier in order to perform selective emitter with a single diffusion step and the replacement of the etching paste by a Laser ablation process step are currently under investigation.

LIGHT-INDUCED DEGRADATION AND REGENERATION IN COMPENSATED UPGRADED METALLURGICAL SILICON Sbastien Dubois1, Nicolas Enjalbert1, Jean Paul Garandet1, Rmi Monna1, Jed Kraiem2 1 INES CEA/LITEN/Laboratory of Solar Components (LCS) BP 332 Savoie Technolac, 50 avenue du Lac Lman 73377 Le Bourget-du-Lac, France Phone: +33 4 79 44 46 48 E-mail: sebastien.dubois@cea.fr 2 Apollon Solar, 69002 Lyon, France

ABSTRACT This study is devoted to the effects of the Light-Induced-Degradation (LID) and Light-Induced-Regeneration (LIR) in strongly compensated Solar-Grade (SoG) multicrystalline (mc) silicon (Si) solar cells. We show that p-type compensated SoG Si is particularly sensitive to LID due to high amounts of boron and oxygen concentrations. This is less the case for compensated n-type Si because the recombination strength of these complexes is lower in n-Si, and furthermore it is possible that for low resistivity values, less complexes are formed in n-Si. Then we studied the LID kinetics in compensated SoG mc-Si solar cells and show that the LID kinetics are much slower in compensated Si than in usual uncompensated Si and cannot be described by the existing models. We discuss how the formation of Boron-Oxygen complexes may be altered by the effects of compensation. We show then that the process of LIR can be successfully transposed to compensated SoG mc-Si, and study the stability of the regenerated state. 1 INTRODUCTION lasting several hours [2]. It has been proposed that the slow process is characterized by the diffusion of a mobile positively charge O dimer (Oi2) towards a negatively charged substitutional B atom [2]. The Oi2 diffusion takes place via a Bourgoin-Corbett mechanism, which involves changes in the structure of the dimers by alternately capturing an electron and a hole [3]. For uncompensated EG Cz-Si, Palmer et al. proposed a model to describe the reaction kinetics of this slow process which is in good agreement with experimental data [4].The defect can be de-activated by annealing above 200 C [2]. The formation as well as the dissolution of this complex is thermally driven reactions described by an Arrhenius equation. Recently, Herguth et al. [5] have shown that under the conjugated action of the temperature and the illumination, most of the BO related defects can be converted to a new state called regenerated state which is weakly recombination active and stable under typical working condition. Up to now, this regeneration process has been investigated only in EG Cz-Si wafers. 3 AMPLITUDE OF THE LID FOR THE SOG-SI CELLS For the coming experiments wafers from two different mc-Si ingots were used. First ingot was grown from EG-Si which was intentionally B doped to obtain ptype Si with a resistivity of about 0.5 ohm.cm. Second ingot was grown from SoG Si purified throughout the PHOTOSIL project. B and P feedstock concentrations, estimated from Glow-Discharge-Mass-Spectroscopy (GDMS) measurements, were respectively equal to 3.11017 cm-3 and 3.41017 cm-3. The SoG ingot has been cut into wafers and electrically characterised in terms of conduction type and resistivity. With such a feedstock material, the difference in partition coefficients (k) between P (k = 0.35) and B (k = 0.8) induced a change in the type of conductivity (p- to n-type Si) from the bottom to the top of the SoG ingot [1]. For both ingots 125125 cm2 solar cells have been processed using an industrial standard process at the CEA-INES pilot line [6]. Cells results are given in Ref. 1. Notice that cells from the ntype region of the SoG ingot were rear junction cells, the junction being formed by means of Al-Si alloying [7]. All the SoG cells were annealed at 230C and then

New Solar Grade (SoG) silicon (Si) production techniques, by metallurgical route, are generally less effective in removing doping species especially boron (B) and phosphorus (P) than the conventional Siemens process. Consequently there is often in such materials high and similar amount of both species. These high concentrations of dopants can affect the carrier mobility () and the carrier lifetime (). Dopants can also form complexes with other impurities, especially light elements or metals, these complexes being often strong recombination centres. For all these reasons, it is necessary to study the influence of the dopants and especially their interactions with other impurities, on the solar cell performances. This study is devoted to the effects of Light-InducedDegradation (LID) and Light-Induced-Regeneration (LIR) in compensated SoG multicrystalline Si (mc-Si). For this work we used SoG Si directly purified from metallurgical grade Si via the PHOTOSIL project [1]. We studied on these cells the effects of both LID and LIR and compared them with those observed in ElectronicGrade (EG) Cz and mc-Si cells. This study is particularly devoted to the amplitude of the LID, to its kinetics, and to its possibility to be cancelled by the LIR. This study is of paramount importance to understand the behaviours of compensated SoG-Si cells under illumination. 2 LID AND LIR IN CRYSTALLINE SILICON

LID is due to the formation under illumination of complexes associating B and oxygen (O). This complex introduces a deep energy level within the Si band gap which decreases and consequently the conversion efficiency of the solar cell. In EG Cz-Si, the concentration of formed complexes exhibits a linear dependence on the substitutional boron (Bs) and a quadratic dependence on the interstitial oxygen (Oi) concentration suggesting the formation of a BsOi2 [2]. In EG Cz grown single-crystals, it has been shown that LID occurs in two distinct and independent steps: a fast process reaching completion on a time scale of tens of seconds followed by an independent slow process,

placed under illumination at 120C during 4 minutes, this step being sufficient to place the samples in the degraded state (as experimentally verified). We measured for all cells the Voc before and after the illumination. The evolution in Voc losses due to the LID are reported along the ingot height in Fig. 1. Despite the presence of many other recombination centres (metal, doping species), the Voc losses in the p-type region of the ingot were very high, ranging from 10 mV in the middle part of the ingot to 15 mV in the bottom part of the ingot. The increase in the Voc losses in the bottom part may be due to higher O concentration ([O]) than in the remainder of the ingot. Indeed the partition coefficient of O is generally believed to be larger than 1, consequently [O] in unsolidified Si melt tends to decrease during solidification [8]. Furthermore in the bottom part of the ingot O can be transferred in the Si bulk from the silica crucible.

formed in n- Si. In our case, as it will be mentioned later, the resistivity varied strongly on the surface of the wafer. Consequently it is possible that on some locations of the n-type wafers, less BO complex are formed, leading to weaker LID for these cells. Both evoked reasons confirm that the use of compensated n-type Si rather than compensated p-type Si could limit the effect of the LID on . 4 LID KINETICS FOR SOG SI CELLS

Figure 1: Resistivity and Voc losses (due to LID) profiles along the SoG-Si ingot height.

The strong LID in the SoG ingot may be caused by large [O] introduced during the SoG purification process or during the ingot growth, but also by the fact that the resistivity of this ingot does not reflect the total [B]. Indeed this ingot is highly compensated. Consequently there is a large amount of B- and P+. A large fraction of B- does not contribute to the conductivity, but all the Bcan contribute in the BO complexes formation and thus to carrier recombination. Consequently compensated materials can be particularly sensitive to LID. It is particularly interesting to focus on the LID occurring in the n-type region of the SoG-Si ingot. Indeed, the Voc losses are very low compared to the ptype region; they are only about 2 mV. This behaviour can be explained by two different reasons. First, the BOi2 complex acts as an attractive Coulomb center which has an electron/hole capture cross section ratio equal to 9.3 [9]. Thus the recombination strength of this complex is much higher in p-Si than in n-Si. Second reason is more based on speculations. An important issue to be discussed is that the Oi++-B- complexes are formed only if Coulomb attraction is present. If the Fermi level (EF) is shifted above the energy level introduced by the O dimer, it is in neutral state and the complex cannot be formed [10]. Unfortunately, the location in the Si band gap of the energy level introduced by the Oi2 has not been yet reported in the literature. Nevertheless in Ref. 11, LID has been observed in n-type Si samples with a resistivity equal to 6 .cm. This indicates first that this energy level is located in the upper half of the Si band gap. However, the exact location is not well defined, and it is possible that for lower resistivities, the BO complex cannot be

p-type solar cells from both mc-Si ingots were used. The B-content of the wafer from the EG ingot was [B]=4.01016 cm-3. As for the wafer from the SoG ingot, the dopant concentrations, determined by GDMS, were [B]=1.951017 cm-3 and [P]=9.01016 cm-3 on the center and [B] = 2.21017 cm-3 and [P]=1.351017 cm-3 on the edges. These differences in dopants content can be related to segregation mechanisms, namely interface curvature and convective mass transport, during crystal growth. Both cells were placed on a Temperature (T) controlled chuck and illuminated by a halogen lamp with an intensity equivalent of 1 Sun. Under illumination, the Voc was measured every second at a given prescribed T ranging from 25C to 65C. Time variations of the actual T were also monitored and T-fluctuations corrected. For both cells, the evolution of the Voc losses at 60C is reported in Fig. 1 and Fig. 2. After a degradation stage, Voc is seen to reach a plateau where it remains constant. Notice that the initial Voc were not recovered by stopping the illumination. Furthermore, the observed degradations were found to be fully reversible by a low T-anneal at 230C, which is a characteristic of B-O complexes induced LID [2]. The degradation/recovery cycle was repeated many times without notable changes. We also never observed noticeable changes in the fill factor of the cells, which means that the observed Voc losses are mainly due to bulk mechanisms. For all these reasons we can legitimately say that the observed degradations are related to BOi2 formation. From the measured Voc values, for minority carrier diffusion length (Ln) much lower than the bases thickness (W), a normalized concentration (N*) of the illumination-induced defects can be defined as [4, 12]:
N * (t ) =

(1) Previous studies of the LID have shown that in uncompensated Cz-Si, N*(T,t) experimental results can be represented by an expression of the form: N * (t ) N * (T , t ) (2) N (t ) = = exp[ R (T )t ]
rel

1 1 exp[q 2Voc (t ) / k B T ] exp[q 2Voc (t = 0) / k B T ]

N * (t )

gen

Rgen (T ) = h h p( [ B] / N Si ]exp( EmA / kBT ) =4(rc)2[NOi2]2/3 (3) p is the hole concentration, h the hole thermal velocity, NSi the Si atomic density, h the hole-capture cross section of Oi2 and the number of sites around Bs which, when reached by Oi2 causes the automatic formation of the BOi2 complex. The value of can be derived through a capture radius rc, rc being defined as the radial distance from a B- ion where the Coulomb energy between the Oi2++ - B- pair is equal to the Boltzmann thermal energy kBT. Within the sphere of radius rc, diffusion proceeds by

where Rgen is the generation rate of the complex. The validity of the form given by eq. (2) has been theoretically confirmed by Palmer et al. [4] who also proposed for Rgen the following expression:

correlated jumps towards B-. The estimated rc was thus about 9.4 nm. In the model, the concentration of Oi2 sites [NOi2] is taken equal to the density of substitutional sites in the Si matrix. Concerning h, Palmer et al. deduced that h should be of the order of 10-17 cm2. The functions Nrel(t) are respectively plotted in Fig. 2 (b) and Fig. 3 (b) for the EG and SoG mc-Si cells from experimental data measured at 60C using Eq. 1 for the analysis of the data. Indeed, the Ln are of the order of 90 m for the SoG mcSi cell, and thus much smaller than the base thickness (W = 220 m). For the EG mc-Si cell the situation is more complex since Ln is of the order of 150 m, and consequently much closer from W. Nevertheless these values being again significantly lower than W, Eq. 1 was again used to compute N* from Voc.

(b)). Nevertheless, Edeg values for both EG and SoG Si remain consistent with existing experimental data concerning the slow process of the LID due to B-O complexes formation [2]. Regarding 2, the activation energies are slightly smaller, namely 0.43 0.18 eV for the SoG-Si cell, and 0.27 0.13 eV for the EG-Si cell. In the EG material, it may appear surprising that the 2 kinetics do not follow the same Arrhenius law as 1, but it should be kept in mind that the way the exponential fit is carried out on the data of Fig. 1 (b) gives only a small weight to the data coming from times where N*(t) is close to N*(), therefore a change in complex formation mechanism cannot be easily detected on global fits. However, the most striking result is that, according to equation 3, both [B] and p values being higher for the SoG cell, 1 and 2 values for the SoG cell should be lower than those for the EG cell. But as shown in Fig. 3, both 1 and 2 values are much higher for the SoG cell than for the EG-cell. Thus the LID presents a kinetic much slower in compensated SoG Si than in uncompensated EG Si, as can also be seen in Fig. 2 (a). Furthermore at 60C the ratio between 1 for the SoG cell and 1 for the EG cell is equal to 1.9 and the ratio between 2 for the SoG cell and 2 for the EG cell is equal to 8.2, indicating a pronounced slowing down of the long time kinetics for the SoG-cell.

Figure 2: (a) Voc reduction at 60C as a function of time for the EG-Si cell. Comparison between the experimental data and data computed from Palmers model. (b) Evolution of Nrel(t) at 60C for the EG-Si cell, computed from Voc measurements.

For the EG cell, the function Nrel(t) is perfectly fitted by a single exponential (Fig. 3 (b)). From the fit, we extracted an experimental Rgen of 3.910-3 s-1 at 60C, in fair agreement with the theoretical value 6.110-3 s-1 obtained from Palmers model (cf. eq. (3)), as can also be seen in Fig. 2 (a) where both experimental and theoretical degradation curves are shown. Notice that for this computation we made the assumption that h and h were the same in mc-Si than in Cz-Si, which may be questionable due to different hole mobility kinetics (see discussion below). The temperature variations of Rgen were found to follow an Arrhenius law with an activation energy of 0.39 0.05 eV, again in agreement with the value 0.475 0.035 eV reported in Ref. 2. For the SoG mc-Si cell, this function cannot be fitted by a single exponential. Thus Palmers model is unable to describe the LID kinetics in this type of Si. Notice also that these kinetics cannot also be fitted by the sum of two or three exponentials. Nevertheless, for purposes of comparison, we reported in Fig. 4 the times 1 and 2 corresponding to Nrel values equal respectively to 1/e and (1/e)2 for both EG and SoG Si. Fig. 4 indicates that in all cases the data can be fitted by an Arrhenius expression. Regarding 1, the thermal-activation energy Edeg was found equal to 0.52 0.12 eV for the SoG-Si cell, and to 0.36 0.11 eV for the EG-Si cell. Notice that in such a situation where only one point is selected to follow the degradation kinetics, the spread for the EG material is much larger than the one derived previously where all the points of the relaxation curve were taken into account (see Fig 1

Figure 3: (a)Voc reduction at 60C as a function of time for the SoG cell. Comparison between the experimental data and data computed from Palmers model. (b) Evolution of Nrel(t) at 60C for the SoG-Si cell, computed from Voc measurements.

To explain the slower LID in compensated Si and the impossibility to fit the experimental data by a single exponential, we consider three mechanisms working in parallel. First, in compensated Si, the concentration of ionized positively charged P is of the same order of magnitude as the concentration of ionized negatively charged B. Thus a proportion of B- will have a P+ in their neighbourhood meaning that the Coulomb field of the Bions will be screened by the presence of the P+, resulting in a smaller capture radius, or equivalently in a longer random walk for the dimers. Depending on the relative positions of the B- and P+ ions, the diffusion kinetics of the Oi2 can vary from quite similar to much slower compared with the uncompensated case. A second issue concerning the slower kinetics in SoG Si is related to the presence of higher concentrations of impurities (metals, dopants, light elements) which affect the hole mobility

( h). Indeed, even in compensated Si, the initial steps of the LID (i.e. when the increase in N* is due to Oi2 not affected by the screening effect discussed above) should be in good agreement with Palmers model. However Rgen value, computed from equation 3 from measured [B] and p=[B]-[P] values is at 60C about 7.010-2 s-1, that is 35 times higher than the measured 1/1 value for the SoG cell, namely 2.010-3 s-1. In equation 3, the only parameter which could be sensitive to the presence of impurities is h, computed in Palmers model under the assumption that the collisions between holes are the dominant scattering mechanism. However, in the SoG Si, h estimated from hole concentration and resistivity () measurements by the formula h ([B]-[P]) |e| = 1, is of the order of 150 cm2.V-1.s-1. Compared to Cz Si, such values are very low, about a factor 3 smaller, meaning that there are in the SoG Si many additional scattering centers, not only dopants but also metals and crystal defects. Therefore it is likely that the reduced h will lead to slower kinetics, even though other factors may be necessary to account for the order of magnitude discrepancy between experimental results and model predictions. Finally, as for the impossibility to fit the degradation kinetics in the SoG Si by a finite sum of exponentials, we already mentioned that distance distribution between B- and P+ ions is such that Coulomb attraction for the Oi2 will vary within the wafer. Another issue is that as mentioned above, [P], and to a lesser extent [B], are not uniform across the wafer, resulting in a variation of p = [B]-[P] and thus of the diffusivity of Oi2. As the measured Voc is a parameter which integrates information from all the regions of the wafers, the LID kinetics cannot thus be expected to be fitted by a unique exponential.

of regeneration can be successfully transposed to compensated SoG mc-Si. Indeed after a duration dependent of the temperature, all the illuminated cells recovered their initial electrical properties. The regenerated cells were then placed during 100 hours under typical working conditions (AM 1.5G, 0.1 W.cm-2) and showed no significant instability (Fig. 5(b)). Notice that for a given T the recovery kinetic is much slower for compensated SoG-Si than for uncompensated Cz-Si. Indeed, at 160 C the time for a complete recovery of the cell performances for the SoG-Si is about 80 minutes, and from Herguth et al. [5] this time is about 20 minutes for Cz-Si (resistivity equal to 1.4 ohm.cm).

Figure 5: (a) Regeneration of the Voc of a p-type compensated SoG-Si solar cell induced by illumination at different T. Solid lines are only guides to the eyes. (b) Stability test of the regenerated state under standard working conditions (1 Sun, T=25 C).

CONCLUSION

We showed that p-type (unlike n-type) compensated SoG Si could be particularly sensitive to LID. We showed that the LID kinetics in compensated SoG Si are considerably slower than in uncompensated materials and proposed reasons to explain such discrepancies. Concerning the process of LIR, we showed that it can be successfully transposed to this kind of material and checked the stability of the regenerated state under standard working conditions. REFERENCES
[1] R. Einhaus, J. Kraiem, F. Lissalde, S. Dubois, N. Enjalbert, R. Monna, To be published in the Proceedings of the 33rd IEEE Photovoltaic Specialists Conference, San Diego (IEEE, New York, 2008). [2] K. Bothe and J. Schmidt, J. Appl. Phys. 99, 013701 (2006). [3] J. C. Bourgoin, J. W. Corbett, Physics Letters A. 38, 135137 (1972). [4] D. W. Palmer, K. Bothe, J. Schmidt, Physical Review B. 76, 035210 (2007). [5] A. Herguth, G. Schubert, M. Kaes, G. Hahn, Progress in Photovoltaics: Research and Applications. 16, 135-140 (2008). [6] R. Monna, J. F. Lelivre, M. Pirot, S. Noel, N. Rondel, M. Lemiti and C. Jaussaud, Proceedings of the 20th European Photovoltaic Solar Energy Conference, Barcelona, Spain (WIP, Munich, 2005), p. 1152. [7] S. Dubois, S. Martinuzzi, F. Ferrazza, Proceedings of the 21st European Photovoltaic Solar Energy Conference, Dresden, 4-8 September 2006 (Munich, WIP), p. 984. [8] Z. Xi, J. Tang, H. Deng, D. Yang, D. Que, Solar Energy Materials and Solar Cells. 91, 1688-1691 (2007).

Figure 4: Arrhenius plot of the times 1 and 2 corresponding to Nrel values equal respectively to 1/e and (1/e)2 for both EG-Si and SoG-Si solar cells. The solid lines represent single exponential fits to the experimental data.

REGENERATION OF THE SoG-Si CELL

The cells used for this study were first annealed at 230 C, to ensure that all the harmful BO complexes were dissociated. Afterwards, we placed the cells under illumination at elevated T (from 120 to 160 C) and followed the evolution in the course of time of the Voc, as shown in Fig. 5. Notice that between each experiment, at different temperature, the cells were again annealed at 230 C. The main result of this study is that the process

[9] J. Schmidt and A. Cuevas, J. Appl. Phys. 86, 3175 (1999). [10] E. R. Weber, Appl. Phys. A: Solids Surf. 30, 1 (1983). [11] K. Bothe, J. Schmidt, R. Hezel, Proceedings of the 29th IEEE Photovoltaic Specialists Conference, New Orleans, 19-24 May 2002 (New York: IEEE), p. 194. [12] M. A. Green, Solar Cells (UNSW, Sydney, 1992).

BENEFICIAL EFFECTS OF DOPANT COMPENSATION ON CARRIER LIFETIME IN UPGRADED METALLURGICAL SILICON Sbastien Dubois1, Nicolas Enjalbert1, Florence Servant1, Jean Paul Garandet1, Rmi Monna1, Jed. Kraiem2 INES CEA/LITEN/Laboratory of Solar Components (LCS) BP 332 Savoie Technolac, 50 avenue du Lac Lman 73377 Le Bourget-du-Lac, France Phone: +33 4 79 44 46 48 E-mail: sebastien.dubois@cea.fr 2 Apollon Solar, 69002 Lyon, France

ABSTRACT This study is devoted to the variations of the carrier lifetime and minority carrier diffusion length with the compensation level in solar-grade crystalline silicon. Especially we show, by using the Shockley-Read-Hall statistics, that an increase in the compensation level reduces the recombination strength of doping species and of some metal impurities. These theoretical results are confirmed by the chemical and electrical characterizations of strongly compensated multicrystalline silicon wafers and solar cells, from silicon purified by the metallurgical route. These results are of paramount importance since an accurate control of the compensation level can lead to strong improvements in silicon solar cells efficiencies. Nevertheless, possible limits of too high compensation levels are also evoked.

INTRODUCTION

New Solar Grade (SoG) silicon production techniques, by metallurgical route, are generally less effective in removing doping species, especially boron (B) and phosphorus (P), than the conventional Siemens process. Furthermore the use of compensated silicon for solar cell production could enlarge the choice of available feedstock and thus support the strong growth of the solar cell production capacities Hence, studies concerned with the influence of dopant compensation on the conversion efficiency of the solar cells have to be undertaken. In the last decades, few studies were devoted to the effects of compensation on the performances of silicon solar cells. Compensation, by increasing the number of scattering centers could directly affect the mobility () of both majority and minority carriers in silicon [1]. In addition compensation could be a significant source of recombination in silicon [2]. Consequently low minority carrier diffusion lengths (L) are expected to be found in compensated silicon. In contrast with what it has been previously suggested, we will show, from simulations and experimental results, that increases in the compensation level can also result in strong increases in the carrier lifetime (), counterbalancing the decreases in and leading to improvements in L and hence higher cell efficiencies. This study is undertaken by combining numerical simulations and experimental data obtained from the electrical characterizations of SoG-Si, purified by the metallurgical route, wafers and solar cells 2 EFFECTS OF COMPENSATION ON CARRIER LIFETIME 2.1 Case of doping species Doping species, such as boron (B), phosphorus (P), aluminium (Al), introduce energy levels in the silicon band gap in the vicinity of the valence band (Ev) for acceptor species (B, Al) or in the vicinity of the conduction band (Ec) for donor species (P). These energy levels are known to enhance the trapping of the free

carriers, but present at high levels they can also contribute significantly to carrier recombination. By applying Shockley-Read-Hall (SRH) statistic [3], it is possible to compute the effects of doping species on . For instance, B introduces in the silicon band gap an energy level located at 0.045 eV above Ev. Due to the proximity of this energy level with Ev, the carrier recombination rate is limited by the capture of electrons by the neutral dopant. The electron capture cross section (n) of neutral B has not been determined accurately and values from the literature concerning capture cross sections of various neutral dopants cover several orders of magnitude [2]. In our simulations, n was thus allowed to vary from 10-17 cm2 to 10-19 cm2, 10-17 cm2 being, as proposed by Geerligs et al. [2], a likely upper bound for a neutral dopant capture cross section.

Figure 1: Evolution with Cl of limited by recombination throughout B atoms in p-type silicon, computed from SRH statistic. Comparison between different values for the electron capture cross-section of neutral B. Fig. 1 presents the evolution of , due to the effects of B atoms, with the compensation level (Cl), computed from the SRH statistic. Cl is defined as ND/(NA-ND), NA and ND being respectively the total concentrations of acceptor and donor species. Notice that for this simulation, we considered only p-type silicon (NA > ND),

fixed [B] equal to 31017 cm-3 and took the injection level (n) equal to 1012 cm-3, which corresponds to lowinjection conditions. We checked that the simulation results were only weakly dependent on n, at least as long as n remains much smaller than NA. Only B is present as acceptor element, consequently NA=[B]. Fig. 1 shows a drastic increase of with Cl. Indeed when Cl varies from 0.1 to 10, is multiplied by 10. This means that increase in Cl reduces considerably the recombination strength of B atoms and could lead to strong improvements in the efficiency of highly B doped solar cells. To account for this a priori surprising result, let us first consider the comparison with standard uncompensated p-type silicon. When the doping level is increased, the recombination strength of shallow levels is expected to increase. Indeed, the Fermi level (EF) gets closer to the acceptor energy level, and by modifying its charge state, enhances its ability for the capture of minority carriers. Thus, in such a strongly doped p-type material, the reduction of the free carrier concentration through the increase of the concentration of donors results in a shift of EF towards the center of the gap, meaning that the attractivity of neutral B atoms for electrons is reduced, and resulting in an increase of . In this first set of simulations, we neglected possible interactions between B and donor or metallic species. We also used the simplified SRH statistic, instead of the more general solution of the continuity equations that takes into account carrier trapping. The recombination centers density being very high (31017 cm-3), we used a model by Macdonald and Cuevas [4] to check whether this simplified SRH statistic could be applied. The Macdonald and Cuevas model states that the simplified SRH statistic can only be applied if the recombination center density is at least one order of magnitude lower than a critical level Ncrit which depends on the location of the energy level in the silicon band gap, on the ratio () between the hole and electron capture cross sections, and on the hole concentration (p0). In our simulations with low-injection conditions, the lowest value of Ncrit is obtained for the lowest p0 value (1015 cm-3) and the lowest value. The hole capture cross section (p) of ionized B- is generally assumed, due to a strong Coulombic attraction, to be very large. For instance a value for p of 7.6510-12 cm-2 measured at 4.2 K is reported in Ref. 5. Thus a value equal to 10 can be seen as a very safe lower bound for . However, even with =10, we found that the minimum value for Ncrit (in lowinjection condition) was equal to 5.21018 cm-3. Such a large value can be traced to the proximity of the B energy level to Ev [4]. In any case, the conclusion is that the use of the simplified SRH model in our simulations was fully justified. 2.2 Case of metal impurities SoG Si contains high concentrations of doping species, but also large concentrations of metal-impurities (especially iron Fe and chromium Cr). In a second set of simulations, we showed that an increase in Cl reduces also the recombination strength of metal-impurities. Interstitial metal impurities (Fe, Cr) are in thermal equilibrium generally paired with substitutional B atoms. These donor acceptor pairs introduce, like doping species, shallow energy levels in the silicon band gap.

For instance the iron-boron (FeB) pair introduces an energy level located at 0.26 eV below Ec and the chromium-boron (CrB) pair introduces an energy level located at 0.28 eV above Ev [6,7] n and p of both pairs have been determined at 300 K [8,9]. We computed, from SRH statistic, the variation of limited by these complexes with Cl, as shown in Fig. 2. In these simulations, [B] and n are respectively set to 1016 and 1012 cm-3. As shown in Fig. 2, due again to the shift of EF towards mid-gap, increases strongly with Cl (for Fe, when Cl varies from 0.1 to 10, is multiplied by about 3). In other words, an increase in Cl reduces the recombination strength of both FeB and CrB pairs and limits their effect on solar cell performance. An important issue to be discussed is that donor-acceptor pairs are formed only if Coulomb attraction is present. If EF is shifted above the energy level introduced by the donor state of the metallic impurity, the interstitial atom is in its neutral state and the pair cannot be formed [10]. Regarding interstitial Cr, its donor level is located at Ec0.22 eV [7].

Figure 2: Evolution with Cl of limited by recombination throughout FeB and CrB pairs in p-type silicon, computed from SRH statistic. Since our focus is only on p-type Si, EF remains always in the lower half of the Si band gap and therefore the CrB pair is formed even at high Cl. As for interstitial Fe, its donor level is located at Ev+0.38 eV [6], also in the lower half of the gap, therefore the situation is a priori more complex. However, even at the highest Cl considered in our simulations, p0 remains above 1014 cm3 . The EF corresponding to this value is located at Ev+0.3 eV, i.e. below the level introduced by Fei. As a consequence, it can also be safely assumed that FeB pairs remain associated even at high Cl. 2.3 Auger and radiative recombinations By reducing p0, increases in Cl reduce also the Auger and radiative recombination rates, both of then, especially for low-injection conditions, increase strongly with the free carrier concentration [11]. 2.4. Application for highly-doped silicon solar cells In Fig. 3 we computed via the simulation software PC1D [12] the variation of the conversion efficiency () of a highly B-doped solar cell with the P concentration. For this simulation, n for the B atoms is equal to 710-18 cm-3. We neglected recombination throughout P atoms.

Nevertheless, Auger recombination is taken into account. Effects of both doping species on is taken into account by using the empiric expression linking for electrons and holes with the sum [P]+[B] given in Ref. 13. These simulation results confirm that increases in Cl, reduce the recombination strength of the B atoms and can lead to strong improvements in . For instance, for [B]=61017 cm-3, when [P] varies from 1016 cm-3 to 41017 cm-3, varies from 10.5% to 12.2% (1.7 absolute %). This reveals the high potential of an accurate control of Cl to increase the efficiency of highly doped silicon solar cells.
15 14 13

and on the center equal to 50 cm2.V-1.s-1. Such values call for two comments : first, they are quite low on an absolute scale, meaning that they are in our SoG material many scattering centers, not only dopants but also metallic impurities. As for the difference in p between center and edges of the wafer, it should be kept in mind that these metallic impurities, dissolved and/or precipitated will like P segregate around the center, and their effect on mobility is thus expected to be stronger there. A point that should be stressed is that the crystallographic characteristics (dislocation density, grain size) were very homogeneous on the surface of the wafer.

Efficiency (%)

12 11 10 9 8 1,00E+16 1,00E+17 1,00E+18

-3

[B]=8E+17 cm-3 [B]=4E+17 cm-3 [B]=2E+17 cm-3

[B]=6E+17 cm-3 [B]=3E+17 cm-3 [B]=1E+17 cm-3

[P] (cm )

Figure 3: Computed variation of the conversion efficiency of a highly B-doped solar cell with the P concentration. EXPERIMENTAL VERIFICATION To experimentally confirm this theoretical study, a multicrystalline silicon (mc-Si) ingot was grown from a highly compensated Si feedstock purified by the metallurgical route [14]. B and P feedstock concentrations, estimated from Glow-Discharge-MassSpectroscopy (GDMS) measurements, were respectively equal to 3.11017 cm-3 and 3.41017 cm-3. With such a feedstock material, the difference in partition coefficients (k) between P (k = 0.35) and B (k = 0.8) induced a change in the type of conductivity (p- to n-type Si) from the bottom to the top of the ingot [15]. A wafer from the bottom part of this ingot was electrically characterized in terms of resistivity (), as shown in Fig. 3. was measured by the eddy current method. The map reveals a strong increase of from the edges to the center of the wafer. We measured by GDMS [B] and [P] on the edges and at the centre of the wafer. On the edges, [B] was equal to 2.61017 cm-3 and [P] equal to 1.41017 cm-3. Around the center, [B] remained equal to 2.61017 cm-3 but [P] was up to 2.11017 cm-3. For a dopant such as P with a relatively small partition coefficient, convectodiffusive solute transport in molten silicon, along with the curvature of the crystallization front, induce a significant increase of [P] from the edges to the center of the wafers. On the other hand, the partition coefficient of B being much closer to unity, these factors have only a very limited influence on the repartition of B across the wafer. Therefore, Cl increases from the edges to the centre of the wafer (from 1.2 to 4.2). Knowing the hole concentration ([B]-[P]) and values, we estimated the hole mobility ( p) on the edges and on the center of the wafer by the formula p ([B][P]) |e| = 1. On the edges p was equal to 110 cm2.V-1.s-1 3

Figure 3: Resistivity map of a mc-Si wafer from an ingot grown from a strongly compensated silicon feedstock purified via a metallurgical process. From this wafer, a solar cell was made by using a standard industrial process at the RESTAURE platform at CEA-INES [16]. Then, a light beam induced current (LBIC) map was made at various near infrared wavelengths, which enabled us to compute a L map, as shown in Fig. 4. The formula L=(D)0.5, where D stands for the electron diffusion coefficient, allows to identify two competing mechanisms: similarly to what has been observed for holes, we can legitimately suppose that the electron mobility ( e) is also affected by increases in the total amount of impurities. As D is directly proportional to e, a negative contribution can be expected. Nevertheless, we can observe a strong increase of L, from 75 m to 147 m from the edges to the center of the wafer. This increase is perfectly correlated with the increase of Cl, and confirms that the strong increase in with Cl counterbalances the effect of the increase of the total amount of impurities on , and thus leads to increase in L. Even without considering the effect of the variation of mobilities, this means that is at least multiplied by 3.8 when Cl varies from 1.2 to 4.2. Table 1: Compensation level, hole mobility and minority carrier diffusion length of the SoG-Si solar cell. Values determined at the center and on the edges of the cell.

Cl Edge Center 1.2 4.2

p (cm2.V-1.s-1) 110 50

Ln (m) 75 147

the ingot length and in addition changes in the type of conductivity may occur. Variation of Cl on the surface of the wafer, especially if the front of crystallisation is disturbed, may also occur.

By comparing this result with the previous simulations, this very strong dependence of with Cl means probably that in the finished cell is rather limited by doping species than metal impurities, which have been probably removed by the efficient gettering effect developed by the P-diffusion (to form the emitter) [17].

CONCLUSION

Our purpose in this letter was to demonstrate that compensation can induce strong improvements in values. Chemical and electrical characterizations of strongly compensated mc-Si wafers and solar cells, which show a strong correlation between increases in Cl and L, support the theoretical findings. These results are of paramount importance, since an accurate control of Cl throughout the feedstock selection, the ingot crystallisation, or in the as-cut wafer, can lead to strong improvements in both and cell efficiencies. AKNOWLEGMENTS The authors would like to thank FerroPem company for the supply of metallurgically purified Si feedstock, and the Apollon Solar and Cyberstar companies for the crystallization of the feedstock material.

REFERENCES [1] M. Cuevas, Physical Review. 164, 1021 (1967). [2] L. J. Geerligs, D. Macdonald and G. Coletti, 17th NREL Workshop on Crystalline Silicon Solar Cells & Modules: Materials and Processes, Vail, Colorado, 5-8 August, 2007. [3] W. Shockley and W. T. Read, Phys. Rev. 87, 835 (1952). [4] D. Macdonald and A. Cuevas, Physical Review B. 67, 075203 (2003). [5] J. M. Brown and A. J. Jordan, J. Appl. Phys. 37, 337 (1966). [6] A. A. Istratov, H. Hieslmair, E. R. Weber, Appl. Phys. A 69, 13-44 (1999). [7] H. Conzelmann, K. Graff and E. R. Weber, Appl. Phys. A 30, 169 (1983). [8] D. Macdonald, T. Roth, P. N. K. Deenapanray, T. Trupke and R. A. Bardos, Appl. Phys. Lett. 83, 142107 (2006). [9] J. Schmidt, R. Krain, K. Bothe, G. Pensl and S. Beljakova, J. Appl. Phys. 102, 123701 (2007). [10] E. R. Weber, Appl. Phys. A: Solids Surf. 30, 1 (1983). [11] D. K. Schroder, IEEE Transactions on Electron Devices. 44, 160 (1997). [12] P. A Basore, Proceedings of the 22nd IEEE Photovoltaic Specialists Conference, 299 (1991). [13] N. D. Arora, J. R. Hauser and D. J. Roulston, IEEE Trans. Electron Devices 29. 292 (1982). [14] R. Einhaus, Y. Caratini, D. Bernou, F. Servant, R. Monna, C. Trassy, J. Kraiem, J. Degoulange, S. Martinuzzi, I. Prichaud, Proc. of the 21st EPVSEC, Dresden, Germany (2006). [15] R. Einhaus, J. Kraiem, F. Lissalde, S. Dubois, N. Enjalbert, R. Monna, To be published in the Proceedings of the 33rd IEEE Photovoltaic Specialists Conference, San Diego (IEEE, New York, 2008).

Figure 4: Figure 4: Minority carrier diffusion length map of a mc-Si solar cell from an ingot grown from a strongly compensated silicon feedstock purified via a metallurgical process.

4 LIMITS OF COMPENSATION ON SOLAR CELL PERFORMANCES Increases in Cl improve . Consequently, we can legitimately suppose that increases in Cl will also improve the efficiency of Si solar cells containing a large amount of impurities. Nevertheless we will show that increases in Cl have also their limits concerning the cells performances. First, increases in Cl affect both the electron and hole mobility. Thus very high Cl may affect both the Jcc (by affecting Ln) and the Fill Factor (by increasing the series resistances) of the cell. In addition, when Cl is very high, due to the location of EF towards the center of the Si band gap, the band curvature is lower, and consequently low Open Circuit Voc are expected to be found. Third, high Cl being associated with large amounts of doping species, numerous interactions between doping species themselves, or between doping species and other elements (metal impurities, carbon, oxygen, as shown in Ref. 18) may also affect and counterbalance the beneficial effects of the compensation. Finally, its particularly difficult to accurately control Cl. One possibility is to add to the feedstock a given concentration of B or P. Nevertheless, due to the stronger segregation of P compared to B, Cl will strongly vary on

[16] R. Monna, J. F. Lelivre, M. Pirot, S. Noel, N. Rondel, M. Lemiti and C. Jaussaud, Proceedings of the 20th European Photovoltaic Solar Energy Conference, Barcelona, Spain (WIP, Munich, 2005), p. 1152. [17] S. Dubois, O. Palais, P. J. Ribeyron, N. Enjalbert, M. Pasquinelli and S. Martinuzzi, J. Appl. Phys. 102, 083525 (2007). [18] S. Dubois, N. Enjalbert, J. P. Garandet, R. Monna, J. Kraiem, in the proceedings of this conference.

MINORITY CARRIER LIFETIME AND DIFFUSION LENGTH IN VOLONTARILY CHROMIUM CONTAMINATED N-TYPE mc-Si WAFERS
1 Santo Martinuzzi, 2Sbastien Dubois, 2Nicolas Enjalbert, University Paul Czanne Marseille - France - 2 INES CEA/LITEN/Laboratory of Solar Components (LCS) BP 332 Savoie Technolac, 50 avenue du Lac Lman 73377 Le Bourget-du-Lac, France

ABSTRACT In n-type silicon the recombination strength of most metallic impurities is neatly smaller than in p-type, due to asymmetric electron and hole capture cross sections for minority carriers. That is true for several impurities (Fe, Ni, Co, Al) but not for Cri. Cri is a fast diffuser which degrades the minority carrier diffusion length L and lifetime in p-type like in n-type Si. Ntype mc-Si wafers were voluntarily contaminated by ion implantation and then annealed at 850C for 90 min. In contaminated samples p and Lp were around 20 s and 120 m, respectively, while the initial values were 100 s and 400 m, respectively. After phosphorus diffusion at 850C for 30 min, the initial values are practically recovered, due to a marked gettering effect. It appears that Cr is easily removed by heterogeneous precipitation and external gettering. Additional thermal treatments might release Cr atoms segregated at extended defects. However, as Cr atoms are volatilsed or are segregated at the top of multicrystalline ingots during the directional solidification, the Cr concentration is sufficiently low in the non voluntary contaminated wafers to be not detrimental for the solar cells. Keywords: C-Si; impurities; minority carrier diffusion length and lifetime.

1. INTRODUCTION N-type material is of interest because it possesses some remarkable advantages: the absence of boron-oxygen complexes, a low variation of minority carrier lifetimes with the injection level and, most importantly, the fact that carrier capture cross sections of several metallic impurities like Fe, Mo, Ni, Ti, Al, Co, are markedly smaller for holes than for electrons. The low values of result from the formation of donor trap levels in the gap, positively charged or neutral, that are thus strongly attractive for electrons and poorly attractive for holes [1]. Although the diffusion coefficient of electrons is 3 times higher than that of holes, it is expected that minority carrier diffusion lengths will be higher in n-type than in p-type silicon [2; 3]. Fig. 1 and 2 show the computed variations of the SRH lifetime in p and n-type wafers as a function of the injection level and dopant densities. It is clear that n-type wafers have a neat advantage at low injection level and also under one sun illumination. This is of paramount importance for multicrystalline silicon (mc-Si) prepared directly from metallurgical feedstock, because the density and activity of the recombination centres are related to the interaction between crystallographic defects and metallic impurities. It is expected that the consequences of impurity- defect interaction are reduced in n-type silicon. However, for some impurities like interstitial chromium atoms Cri, substitutional Au and Zn atoms, the values of in n and p-type Si are close. Cri atoms are fast diffusers, frequently detected in mc-Si wafers, which the capture cross sections were recently revaluated [4] and are given by table 1. Table 1: minority carrier capture cross sections of Cri atoms and of Cri-Bs complexes for holes and for electrons

Cri CriBs

Et Ec - 0,23 eV Ev + 0,27 eV

2 p (cm ) 4,00E-15 1,00E-14

2 e (cm ) 2,00E-14 5,00E-15

A Cri concentration of ~ 1012 cm-3 suffices to reduce the lifetime of holes and consequently the conversion efficiency of solar cells [5], as shown by the computed results given in Fig. 3 and Fig.4. The reduction is more marked for rear junction solar cells. Note that for front side junction cells the conversion efficiency is higher for n-type wafers than for p-type, whatever is the Cri concentration and the doping level of the wafers. In the present work we try to demonstrate that Cr is less detrimental than expected, for the cell conversion efficiency, because it is easily removed during the processing steps needed to make the cells and that its concentration in the multicrystalline silicon ingots cannot be too high because it is volatilized during the crystal growth.
1,00E-04

p-Si
SRH Carrier Lifetime (s)

[B]=1E14 cm-3 -3 [B]=1E15 cm-3 [B]=1E16 cm-3

[CrB]=1E13 cm

1,00E-05

1,00E-06
1,0E+10 1,0E+11 1,0E+12 1,0E+13 1,0E+14 1,0E+15 1,0E+16 1,0E+17 1,0E+18

Excess Carrier Density (cm )

-3

Fig. 1 Variation of SRH lifetime vs injection level in ptype mc-Si wafers for various dopant densities.

2. EXPERIMENTAL N-type mc-Si wafers, resistivity 1.6 cm were Cr implanted at a dose of 1014 cm-2 and an energy of 100 keV, then annealed at 850 C for 90 min in argon flux. On other samples 500 nm of Cr were deposited by vacuum evaporation and annealed under the same
1,00E-04

[P]=1E14 cm-3

SRH Carrier Lifetime (s)

n-Si [Cri]=1E13 cm
1,00E-05

[P]=1E15 cm-3 -3 [P]=1E16 cm-3

1,00E-06
1,0E+10 1,0E+11 1,0E+12 1,0E+13 1,0E+14 1,0E+15 1,0E+16 1,0E+17 1,0E+18

Excess Carrier Density (cm )

-3

conditions. Then the samples are chemically etched in order to remove 15 m from each side. As the diffusion coefficient is D ~ 3x10-7 cm2/s at T ~ 850C, the intragrain concentration [Cri] reaches the solubility limit at 850C, i.e. ~1.5x1012cm-3. Lifetime p and diffusion length Lp of minority carriers were determined by Inductively-Coupled Quasi-Steady-State Photoconductance Decay (IC-QssPC) measurements (surfaces passivated by hydrogen rich SiN layer deposited by plasma enhanced chemical deposition) and LBIC scan maps (obtained thanks to Au-Si diodes for which the thickness of the Au layer was 8 nm). Wafers were investigated before and after they experienced gettering treatments by phosphorus diffusion at 750C for 30 min, 850C for 30 min and at 900C for 1h. Wafers were also submitted to phosphorus diffusion at 850C for 30 min and to hydrogenation after deposition of a hydrogen rich silicon nitride layer: L(Pdif +H) was measured. Other wafers experimented phophorus diffusion and aluminium gettering by formation of a AlSi alloy at 850C for 30 min. L(P+AlSi) was measured. 3. RESULTS AND DISCUSSION In the raw wafers, p and Lp are close to 100 s and 400 m, respectively. After contamination and annealing, there is no difference when Cr ions are implanted or when Cr is deposited by vacuum evaporation. As shown by Fig. 5, it is found after contamination: tau(Cr) = 20 s and Lp(Cr) = 120 m in both Cr implanted and Cr evaporated samples. Note that lifetime of carriers does not change substantially with the injection level in our ntype silicon wafers, as expected. Using the SRH statistics with p ~ 4x10-15 cm2 the concentration of Cri is evaluated to ~ 1.25x1012 cm-3, that is always close to the solubility limit at 850 C and suggests that Cr atoms in excess have precipitated at the surface and at extended crystallographic defects, or are included in intra-grain
1000

Fig. 2 Variation of SRH lifetime vs injection level in ntype mc Si for various dopant densities.
1 0,9 0,8

relative efficiency (%)

0,7 0,6 0,5 0,4 p-Si cell 0,3 n-Si cell (front junction) 0,2 n-Si cell (rear junction d=300 m) 0,1 0 1,00E+10 n-Si cell (rear junction d=150 m) 1,00E+11 1,00E+12 1,00E+13 1,00E+14

Dissolved Cr concentration (cm-3)

carrier lifetime (us)

Fig. 3 Computed variations of normalized conversion efficiency in p and n-type solar cells as a function of Cr concentration, for front side and backside junction cells. The wafer doping level is 1015 cm-3.

Cr implanted P01 Cr implanted P02 Cr evaporated reference

100

1 0,9 0,8

relative efficiency (%)

0,7 0,6 0,5 0,4 0,3 0,2 0,1 0

10 1E+13

1E+14

1E+15

1E+16

1E+17

injection level (cm-3)

p-Si cell n-Si cell (front junction) n-Si cell (rear junction d=300 m) n-Si cell (rear junction d=150 m)

Fig. 5. Variation of carrier lifetime in Cr contaminated wafers and in a reference wafer, as a function of injection level.
1,0E+13 1,0E+14

1,0E+10

1,0E+11

1,0E+12

Dissolved Cr concentration (cm-3)

Fig. 4 Computed variations of normalized conversion efficiency of p and n type solar cells as a function of Cr concentration. The wafer doping level is 1016 cm-3.

micro-precipitates during the annealing and the cooling of the samples. After P diffusion at 750 C for 30 min, Lp reaches 120 m. After P diffusion at 850C for 30 min Lp and p reach 380 m and 100 s, respectively. After P diffusion at 900C for 1 h, Lp is about 400 m, which is the initial value in the non contaminated sample. That suggests that

Cri atoms (like other metallic atoms) can be easily gettered during P diffusion. This is very useful as P diffusion may be used, in n-type silicon, to passivate the front surface of rear junction solar cells. Fig. 5 illustrates these evolutions.

Diffusion length
450 400 350 300 250 200 150 100 50 0 1 raw contaminated P diff. 750 C 30min Pdiff. 850 30 C min P diff. 900 1h C

never so high, like in our voluntary contaminated samples, for two reasons. Fist, a large part of Cr atoms evaporates from melted silicon during the crystal growth, because the evaporation temperature of Cr is rather low: 1205 C [6], and then the partition coefficient of Cr between liquid and solid phases of silicon is very low: ~ 10-6. So, a large part of chromium out-diffuses from the melt and is accumulated in the very top of the ingots, which is generally discarded.

Fig. 6 Evolution of the minority carrier diffusion length in phosphorus diffused uncontaminated and Cr contaminated wafers. Fig. 8 LBIC contrast Idef/Io scan map of a contaminated and phosphorus diffused sample after annealing: GBs are enlarged and grains are degraded. This is why very large values of Lp and of p are found in n-type mc-Si wafers, although the feedstock used to growth the crystal might be contaminated by chromium. 4. CONCLUSIONS From these results we can conclude that the presence of Cr in mc-Si is less harmful than expected because during the growth of an ingot by directional solidification, a large part of Cr is volatilised and certainly removed by segregation. Moreover, phosphorus diffusion - an indispensable processing step - is able to remove rapidly interstitial Cr toms. Cr atoms trapped at extended crystallographic defects and included in microprecipitates might be released during subsequent thermal treatments, but the concentration of interstitials atoms is sufficiently low, to be not detrimental. AKNOWLEDGEMENTS The authors would like to thank the French Agency for Environment and Energy Management (ADEME), and the French National Agency for Research (ANR) for their financial support through the PHARE research contract. REFERENCES Fig. 7 LBIC contrast Idef/Io scan map of a Cr contaminated sample after phosphorus diffusion at 850C for 30 min. [1] D. Mac Donald, J.L. Geerligs, Proc. of 19th European Photovoltaic Solar Energy Conf., Paris France (GraphicaLito 2004 Rome) p.492-495 [2] A. Cuevas, M.J. Kerr, C. Samundsett, F. Ferrazza, G. Coletti, Appl. Phys. Lett. 81 (2002) 4952-4955. [3] S. Martinuzzi, O. Palais, M. Pasquinelli, F. Ferrazza,. EPJ AP, 32 (2005) 187-192. [4] J. Schmidt, R. Krain, K. Bothe, G. Pensl, S. Beljakowa, J. Appl. Phys. 102 (2007) 123701.

When the gettered wafers are submitted to the annealing needed for the metallization, in a RTA or in a conventional furnace, it is found that Lp ant p are divided by 2 and 4 respectively. The comparison of the LBIC scan maps of Fig. 7 (before annealing) and Fig. 8 (after annealing) shows that the influence of grain boundaries (GBs) are enlarged and that the grains are degraded after the annealing. That is explained by the dissolution, during the annealing, of impurities included in intra-grain micro precipitates or segregated at GBs, remaining after the P diffusion. The deposition of a PECVD SiN-H layer increases a little bit the mean value of Lp: L(Pdif + H) ~ 120 m, only.

However, in the industrial production the problem is not so hard because the Cr atom concentration in the wafers cut from ingots produced by directional solidification is

Lp (m)

[5] J. Schmidt, K. Bothe, R. Bock, C. Schmiga, R. Krain, R. Brendel, G. Pensl, S. Beljakowa, Proc. of 22nd European Photovoltaic Solar Energy Conf., Milano, Italy (WIP Munich 2007) p. 998-1001. [6] K. Graff, in: Metal Impurities in Silicon Device Fabrication (Springer- Berlin 1995) p. 116-117.

SILICON HETEROJUNCTION SOLAR CELLS : SURFACE PASSIVATION QUALITY ON LARGE AREA N TYPE AND P TYPE MONOCRYSTALLINE SILICON
Pierre Jean Ribeyron1, Thibault Desrues, Aurlie Vandeneynde1, Florent Souche1, Anne Sophie Ozanne1, Martin Labrune2, Pere Roca i Cabarrocas2, Rmi Chouffot3, Jean Paul Kleider3 1 INES CEA/LITEN/Laboratory of Solar Components (LCS) BP 332 Savoie Technolac, 50 avenue du Lac Lman - 73377 Le Bourget-du-Lac, France 2 LPICM (UMR 7647, CNRS) Ecole Polytechnique - 91128 Palaiseau Cedex, France, 3 LGEP, CNRS UMR 8507, CNRS ; SUPELEC ; Univ. Paris-Sud; UPMC Univ Paris 6; 11 rue Joliot-Curie, Plateau de Moulon - 91192 Gif-sur-Yvette Cedex, France

ABSTRACT The aim of this work is to present our progress on heterojunction solar cells, based on a low temperature route using hydrogenated amorphous silicon (a-Si:H) ultrathin films deposited by PECVD on bulk crystalline silicon. We focus our work on both n-type and p-type monocrystalline silicon to achieve large area solar cells for industrial application. The analysis of the a-Si:H/c-Si interface quality is analyzed by the QSSPC technique that allows evaluating the passivation properties as well as qualifying the different steps. First, a new cleaning process is evaluated with respect to passivation properties, demonstrating that it is possible to obtain nearly as good passivation on textured as on polished samples with appropriate cleaning. An intrinsic layer is added to an emitter or BSF double side structure in order to show the impact of such a layer on passivation properties. Finally, full heterojunction solar cells have been characterised thanks to I-V measurements. Record Voc values up to 668 mV have been obtained on 5*5 cm cells. Efficiencies of 16,2% and 15,8 have been obtained on P type and N type wafers, for cells of 5*5 cm . Keywords: heterojunction, c-Si, interfaces

INTRODUCTION 2 EXPERIMENTAL The solar cell structures are described in Figure 1. Two different types of monocrystalline silicon substrates were used for this study: p type c-Si wafers (<100> oriented, W=300m, =14-22 .cm) and n type c-Si wafers (<100> oriented, W=300m, =2-5 .cm). The emitter and the back surface field were deposited on polished and textured samples. A multistep cleaning recipe was applied to the wafers just before processing of the front and back side.(see also [3]) A doped a-Si:H layer was deposited either directly onto the cleaned substrate or onto a thin intrinsic a-Si:H layer. Both doped a-Si:H(n or p) and intrinsic a-Si:H layers were deposited by means of a capacitively-coupled PECVD system at an RF frequency of 13.56 MHz under various conditions. The thicknesses of these layers, as measured by spectroscopic ellipsometry, are in the range of a few nanometers. After deposition of the front layer, the wafers were flipped for deposition of the back surface layer. Then, a Transparent Conductive Oxide (typically ITO) was deposited by DC Magnetron sputtering on the front and back of the wafer in the case of N type silicon and only at the front in the case of P type silicon. Evaporated or sputtered aluminium was deposited at the back. Finally, a low-temperature silver paste was screen-printed onto the TCO layer, and annealed for front contact formation. The whole process was performed at very low temperature (T<200C).(fig.1) Passivation properties were evaluated by QSSPC measurements [4] (Quasy Steady State PhotoConductance Technique) on wafers covered on both side in order to correlate the deduced effective lifetimes with the solar cell performance. From QSSPC

The rapid growth of photovoltaic industry leads to develop new structures with high efficiency potential on p-type and more recently n-type crystalline silicon (c-Si). The amorphous/crystalline heterojunction solar cells (a-Si:H/c-Si) concept is very promising and is attracting more and more attention around the world. Sanyo is up to now, the only company using this structure for the production level and has demonstrated record efficiencies on n-type monocrystalline silicon at the R&D level [1]. This structure has many advantages: high open-circuit voltage values, low saturation current, and low processing temperatures [2]. Furthermore, a-Si:H/c-Si structures are well adapted for the reduction of wafer thickness due to the low mechanical stresses induced by the low temperature process. The active part of the device is basically produced by a low temperature growth (150-250C) of ultra-thin layers of amorphous silicon (a-Si:H) onto both sides of a thin crystalline silicon wafer-base, that form both the emitter and back surface field (BSF), while acting at the same time as a surface passivation. Each step of the heterojunction cell fabrication process is really critical to achieve high passivation properties of the device and excellent solar cells characteristics. This work focuses on critical steps of heterojunction solar cells process integration on large area (4 inch wafers) solar cells which can impact the passivation properties. We present the role of an optimized cleaning after texturation on the passivation. Regarding the PECVD films, we study the introduction of an amorphous intrinsic layer between the doped a-Si:H and the c-Si substrate.

we can deduce the value of the emitter saturation current density, Joe, and the so-called implied open-circuit voltage (Voc) calculated from eff value at 1 sun illumination. Each processed cell was analyzed by I(V) measurements.

Figure.2: SEM pictures of Si pyramids surface obtained by KOH treatment before (a) and after (b) chemical cleaning On polished and textured samples, the same type of intrinsic amorphous layer was applied on both sides. In figure 3, carrier lifetime curves measured on polished and textured samples are presented.
10000
DSP w afer effective carrier lifetime (us)

Transparent conductive oxyde (TCO) Amorphous silicon a-Si:H (p) or (i/p)

1000
Textured w afer

Crystalline Si c-Si (n)

Amorphous silicon a-Si:H (n) or (i/n) Transparent Conductive Oxyde (TCO)

100

Al

10 1E+13

1E+14

1E+15

1E+16

Excess carrier concentration(cm -3)

Transparent conductive oxyde (TCO) Amorphous silicon a-Si:H (n) or (i/n)

Figure.3: QSSPC effective lifetime with respect to injection level on 14-22 ohm.cm p-type silicon passivated on both sides by a 20 nm intrinsic a-Si:H layer. One can see that effective carrier lifetime for polished sample is slightly higher compared to that of textured wafer for approximately the same passivating layer thickness. Nevertheless, the level of surface passivation is quite good as the effective lifetime reaches up to 1 ms, even on the textured sample. The slight difference of level passivation can be attributed to the larger surface area in the case of textured samples. A second parameter which can slightly decrease the surface passivation for textured samples is the <111> crystallographic orientation of textured samples which can increase the number of interface defects due to the amount of silicon atoms at the surface which is higher than in <100> plans [2]. Values of surface recombination velocities achieved on polished and textured wafers are summarized in Table 1. The use of appropriate cleaning conditions allows to obtain surface recombination velocities as low as 18 cm/s (at an injection level of 1015 cm-3) on textured samples, values that are comparable to those obtained on polished wafers.

Crystalline Si c-Si (p)

Amorphous silicon a-Si:H (p) or (i/p)

Al

Figure .1 : Schematic view of our double heterojunction solar cell on n-type and p-type monocrystalline silicon

RESULTS AND DISCUSSION 3.1 Assessment of the cleaning efficiency after texturation

Textured heterojunction solar cells are made from the same type of wafers as polished ones. The wafers are preliminary textured by a classical KOH-IPA wet process. This kind of texturation generates chemical pollutions at the interface such as metallic ions, particles and organics. So, it is mandatory to make an efficient cleaning of both surfaces. In this study, we have optimised the classical RCA process followed by HF dip prior to the different layer depositions. The figure 2 shows SEM pictures obtained on textured wafer before and after chemical cleaning. After applying the appropriate cleaning, the surface contamination is highly reduced: particles are removed and the pyramids side walls are smoother, enabling a better conformation of aSi:H nanometric layers. Furthermore, metallic contamination is removed which is mandatory for high quality passivation purposes.

eff at 1E15cm-3 SRV (cm/s) Textured wafer Polished wafer 880 1700 18 8

Table.1: Comparison of the effective lifetimes measured at an excess carrier density of 1015cm-3 and surface recombination velocities obtained on polished and textured p-type wafers.

(a)

(b)

3.2 Impact of an intrinsic layer on passivation properties of doped N+ amorphous layers on N type silicon (improvement of the BSF)

It is well known that the interface state density caused by the doping material can damage the junction properties and therefore the introduction of a very thin intrinsic aSi:H layer between the doped a-Si:H and the c-Si substrate can improve drastically the passivation properties by decreasing the Dit level. In order to investigate the influence of an intrinsic layer on passivation properties of doped N+ amorphous layers on N type, we performed passivation tests with different process conditions of amorphous silicon layer. The ratio SiH4/H2 was maintained constant and the flows were changed. The stack a-Si:H / a-Si:H (N+) was deposited in the same run or in two separated runs with a high vacuum step between.

3.3 Impact of an intrinsic layer on passivation properties of doped P+ amorphous layers on N type silicon (improvement of the emitter) The Boron doped amorphous layer is well known to contain less active doping atoms than the Phosphorous doped layers. So, it is needed to have a strongly doped layer in order to obtain the right conductivity. Thus, an intrinsic layer at the interface between the crystalline silicon and the amorphous doped layer is mandatory to obtain good passivation level. This hypothesis is confirmed by the experimental results on figure 5.
1000
DDV (s)

10000

100
Effective Carrier Lifetime (s)
a-Si:H(i ) (process 2) / Vacuum step / a-Si:H(n)

i 5nm (1)/P+ 10 nm
1000
a-Si:H(i )(process 1) / a-Si:H(n) (same run)

i 5nm(2)/P+ 10 nm
a-Si:H(i) (Process 1) / Vacuum step / a-Si:H(n )

P+ 10 nm
10
1E+13 1E+14 1E+15
Ninj (cm -3)

100

a-Si:H(n)

1E+16

1E+17

10 1E+13 1E+14 1E+15 1E+16 1E+17 Excess Carrier Concentration (cm^-3)

Figure.4: QSSPC effective lifetime with respect to injection level on 2-5 ohm.cm n-type silicon for a double side deposition of a BSF (phosphorous doped amorphous layer) with or without intrinsic layers. eff at 1
a-Si:H(n) a-Si:H(i)(process1)/vacuum/ a-Si:H (n) a-Si:H(i)(process1)/ a-Si:H (n) a-Si:H(i)(process2)/vacuum/ a-Si:H (n) sun (s) 400 429 750 4000 Voc
(mV)

Joe
(fA.cm-2)

660 662 690 720

200 170 60 17

Figure.5 : Lifetime measurements (s) on N type silicon 2-5 ohm.cm with respect to injection level (QSSPC) for a double side deposition of an emitter ( boron doped amorphous layer) with or without intrinsic layers. Nevertheless, the way, the intrinsic layer is obtained has a tremendous impact on the passivation properties of the stack (i/p). Implied Voc up to 657 mV are obtained with an appropriate intrinsic layer. However, this is not yet sufficient to obtain very high Voc at the solar cell level. Thus, the intrinsic layer and the doped layer still have to be improved. 3.4 Solar cell results on N and P type monocrystalline solar cells, 5*5 cm. Finally, we have processed solar cells, n-type (FZ, 2-5 ohm.cm) and p-type (14-22 ohm.cm, CZ) without intrinsic layers on textured samples in order to obtain reference results, using our new cleaning process. The front metallization was made by screen printing process of low temperature pastes with a maximum temperature of 200C. In the case of P type solar cells, the emitter has no intrinsic layer, as a really good passivation level can already be achieved with phosphorous doped layers. The thickness of the emitter was 8 nm. At the back, a simple P+ doped layer was applied, 30 nm thick. On P type silicon, Aluminum is directly applied on the P+ amorphous layer. The results are presented on table 3. Record Voc are obtained, up to 668 mV without any intrinsic layer. Nevertheless, the short circuit current (Isc) and fill factor value (FF) are quite low. This is mainly due to the back surface field which is not sufficiently passivating at low injection level. Nevertheless, we can assume that adding intrinsic layers will enable to obtain significantly better results.

Table.2: Results deduced from QSSPC measurements for different (i) layers, and processes. As shown in fig. 4 and table 2, the introduction of an intrinsic amorphous silicon layer between c-Si and aSi:H(n) allows a significant improvement in terms of surface passivation. We obtained Effective lifetime greater than 1 ms, Voc implied of 720 mV and Joe lower than 20 fA.cm-2. So, even if the phosphorus doped layer enables to obtain Voc implied up to 660 mV, which is quite good, adding an intrinsic layer enables to obtain outstanding passivation with Voc up to 720 mV and Surface recombination velocities less than 2,5 cm/s. These results were strongly dependent on the type of intrinsic silicon layer. It is therefore needed to optimize the process conditions in order to have a high quality film. Another critical parameter to emphasize is the process sequence of the two layers which has to be studied into more details in the coming months. It has to be noted that this kind of conclusions can be generalized to P type silicon. But, in this case, it is the emitter which is strongly improved.

Voc (mV) p-type n- type

Jsc(mA/cm)

FF(%)

668 620

32.9 33.5

73,7 76.2

(%) 16.2

[2] M. Taguchi, K. Kawamoto, S. Tsuge, T. Baba, H. Sakata, M. Morizane, K. Uchihashi, N. Nakamura, O. Oota, Progr. Photovolt. Res. Appl. 8 (2000) 503. [3] P.-J. Ribeyron, A. Vandeneynde, J. Damon-Lacoste, P. Roca i Cabarrocas, A. S. Gudovskikh, R. Chouffot, J. P. Kleider, Proc of the 21th European Photovoltaic Solar Energy Conference, Dresden, Germany ( 2006), p 926 [4] A.Cuevas, Solar Energy Materials and Solar Cells, 57 (1999), p 277-290.

15.8

Table 3: I-V results for the best textured solar cells. For the p-type c-Si cell, the thickness of front and back layers was 8 nm and 30 nm, respectively. These values were 10 nm and 25 nm for the n-type c-Si cell. In the case of n-type c-Si, a simple emitter layer was deposited at the front, 10 nm thick. At the back, the new BSF with intrinsic layer was applied with outstanding passivation properties. The results are presented on table 3. Results are poorer than on p-type c-Si (see Table 3). This is mainly due to the front surface which is not enough passivated. So, we strongly believe that adding the appropriate intrinsic layer will enable to drastically improve these results

CONCLUSION and PERSPECTIVES

In this paper we have discussed about the progress achieved on some steps of heterojunction solar cell fabrication, focusing our approach on the surface passivation properties. An appropriate cleaning step before deposition is crucial for textured wafers and allows to achieve excellent surface passivation, comparable to that obtained on polished wafers. A surface recombination velocity as low as 18 cm/s has been reached. We have demonstrated that higher surface passivation can be achieved by the introduction of an optimized thin intrinsic a-Si:H layer between the doped a-Si:H and the Si-c substrate on N type, but this demonstration could also be done on P type wafers. Finally, we have reached Voc up to 668 mV without any intrinsic layer, showing that a great potential of improvement is possible at the solar cell level.

ACKNOWLEDGEMENTS This work was partly supported by ANR (Agence Nationale de la Recherche) in the framework of the French national photovoltaic PHARE project, and by European Community's Seventh Framework Programme (FP7/2007-20013) under grant agreement n211821 (HETSI project). REFERENCES [1] T. Sawada, N. Terada, S. Tsuge, T. Baba, T. Takahama, K. Wakisaka, S. Tsuda and S. Nakano, Proc. of 1st World Conference on Photovoltaic Energy Conversion, Hawaii, USA, (1994), p. 1219.

NEW PROCESS INTEGRATION FOR INTERDIGITATED BACK CONTACT (IBC) a-Si:H/c-Si HETEROJUNCTION SOLAR CELLS Thibault Desrues1,5, Pierre-Jean Ribeyron1, Aurlie Vandeneynde1, Anne Sophie Ozanne1, Florent Souche1, Yannick Veschetti1, Armand Bettinelli1, Pere Roca i Cabarrocas2, Martin Labrune2, Dicknoum Diouf3, Jean Paul Kleider3, Mustafa. Lemiti4 INES CEA/LITEN/Laboratory of Solar Components (LCS) BP 332 Savoie Technolac, 50 avenue du Lac Lman 73377 Le Bourget-du-Lac, France (33)479444603 2 CNRS/LPICM, Route de Saclay F-91192 Gif-sur-Yvette; (33)169334314 3 LGEP, UMR 8507, CNRS; Suplec; Univ. Paris-Sud; Univ. Pierre et Marie Curie Paris 6; 11 rue Joliot-Curie, Plateau de Moulon, F-91192 Gif-sur-Yvette Cedex; (33)169851645 4 INL (INSA de Lyon) 20 Av. A. Einstein F-69100 Villeurbanne Cedex; (33)472438731 5 Corresponding author: thibaut.desrues@cea.fr ABSTRACT This study deals with a way to improve the efficiency of conventional a-Si:H/c-Si heterojunctions (HJ) solar cells using the interdigitated back contact (IBC) structure. First we optimize doped and intrinsic amorphous hydrogenated silicon passivation layers using lifetime measurements and mapping. As a reference we realize conventional a-Si:H/cSi HJ solar cells on large area (25 cm2) monocrystalline n-type c-Si. We show that conventional HJ cells efficiency is mainly limited by the front side cell structure. One way to overcome the optical and resistive losses of conventional aSi:H/c-Si HJ cells is the implementation of an interdigitated back contact (IBC) design. We propose a comparison between two fabrication processes for large area IBC-Heterojunction solar cells based either on screen printing technology or on the use of metallic masks. Then, we also develop a passivating front surface and anti-reflective layer. Finally, we fabricate IBC a-Si:H/c-Si HJ solar cells (28 cm2) with efficiency up to 8.2% (Voc 639mV; Jsc 25.7mA.cm-2; FF 49.8%), mainly limited by a high series resistance. Keywords: c-Si, heterojunction, back contact

INTRODUCTION

Two technologies currently allow high efficiency (>20%) large area solar cell manufacturing at an industrial scale. First, a-Si:H/c-Si heterojunctions (HJ) use thin hydrogenated amorphous silicon layers deposited on c-Si wafers to both passivate the c-Si surface and create either emitter or base contact. This technology shows several advantages: low processing temperatures (below 300C), excellent surface passivation (high VOCs) and high throughput (PECVD equipment). The second technology is based on the Interdigitated Back Contacted (IBC) cell structure where both emitter and base contacts are located at the rear side of the cell. This type of cell has the advantage of avoiding optical and recombination losses (neither contact grid nor high doping at the front surface) and resistive losses (no trade off between grid shadowing and series resistance). Sanyo demonstrated 100cm2 HJ cells efficiencies of 22.3% with VOC values of 725mV [1] due to the insertion of an undoped a-Si:H layer between the n-type c-Si substrate and the doped a-Si:H layers (HITtm: Heterojunction with Intrinsic Thin layer). SunPower Corp. developed 148 cm2 IBC cells also on high quality n-type c-Si using homojunction technology i.e. high temperature doping steps and demonstrated efficiencies above 23%[2]. Combining both technologies in the same structure should allow an excellent surface passivation by reducing at the same time optical and resistive losses. Such a new device has already been introduced: the RECASH (Rear Contact amorphous / Crystalline Silicon Heterojunction)[3], the BEHIND (Back Enhanced Heterostructure with INterDigitated contacts)[4] and the IBC-SHJ [5] cells. Impressive VOC (up to 691mV [6]) and the best efficiency (13.9% [3]) demonstrate the

potential of the structure which remains limited by interface defects, high contact resistance [3] or poor aSi:H(i) conductivity[4][6]. However all these a-Si:H/c-Si HJ-IBC cells have been fabricated on small area (<6.25 cm2) c-Si using photolithography steps [3][6] or deposition through metallic masks followed by dry etching[4]. In this work, we propose two alternative fabrication processes adapted to larger cell area (28 cm2) focusing on the use of screen printing technology and metallic masks. N-type c-Si show very high diffusion length (required for rear emitter structures) and allow higher a-Si:H(p)/c-Si(n) heterojunction potential compared to the a-Si:H(n)/c-Si(p) one [7]. Therefore, this material was used in our experiments.

DEVELOPMENT OF a-Si:H LAYERS

After native oxide removal in buffered HF we deposit Boron- and Phosphorus-doped a-Si:H layers on 4 2-5 .cm n-type FZ polished wafers using capacitivelycoupled PECVD system at an RF frequency of 13.56 MHz. Their passivation properties as well as the influence of an intrinsic a-Si:H film between the substrate and the doped layer are investigated by lifetime measurement (W-PCD and QSSPC) Figure 1.

1000

a-Si:H (i) 5nm a-Si:H (p) 10nm

100

a-Si:H (n) 20 nm a-Si:H (i) 5nm + (p) 10nm a-Si:H (i) 5 nm + (n) 20 nm 1E+14 1E+15 1E+16 Excess Carrier Density (cm^-3) 1E+17

10 1E+13

surface texturing, absorption in the front TCO and a-Si:H (p) emitter. Increasing the emitter thickness induces more absorption and series resistance due to the poor quality and conductivity of the a-Si:H (p) layer. The slight improvement of the VOC is due to a better surface passivation for thicker (p) layer (enhancement of the field effect). The screen-printed metallization also limits the cell results due to the grid shadowing effect and resistive losses in the metal fingers (poor FF). The Interdigitated Back Contact (IBC) structure can help to overcome these limitations found on standard front emitter HJ cells by avoiding front optical losses while decreasing the series resistance.

Figure 1. Lifetime measurements of doped and stacked a-Si:H layers deposited on both side of the wafer. The insertion of a thin intrinsic a-Si:H layer has a detrimental effect under the p-layer whereas a slight improvement is observed for the n-type one. Other studies suggest that different a-Si:H(i) layers are probably required to optimize passivation and contact properties of a-Si:H(n) and (p) [8]. In a second step for optimisation, we fabricate solar cells using a single a-Si:H (p) emitter on the front side and the stack of a-Si:H (i)/ a-Si:H (n) on the rear surface. Figure 2 shows the lifetime uniformity on 4 inches c-Si substrates for the best passivation stacks. The lifetime of the a-Si:H(p)/c-Si(n)/ a-Si:H(i+n) structure is limited by the passivation properties of the p-type emitter layer.

Lifetime (s)

IBC PROCESS INTEGRATION: A COMPARISON A way to obtain the most efficient HJ-IBC cell is to deposit at the rear side three different layers and two contact electrodes (Figure 3). Without simplifying this geometry, five localization steps are then needed for the fabrication of the structure.

Figure 3. HJ-IBC cell geometry (not to scale) Two technologies that could be used at an industrial scale allow sufficient alignments between the different layers at the rear side. The first one is based on screen printing of polymeric paste (SP) to pattern dielectric masks (Lift off layers) and allows a perfect optical alignment between the different depositions [9]. The second one uses thin patterned metallic masks (MM) mechanically aligned to limit the number of fabrication steps [3][10]. Both processes are compared focusing on surface passivation and deposition conformity. 3.1 Influence of a-Si:H localization on rear surface passivation A test structure is fabricated on 4, 2-5 .cm n-type, FZ polished wafers to compare the passivation loss due to localization process. This consists in the deposition at the rear side of an a-Si:H (p)30nm emitter finger (1400m width) localized with one process, followed by a cleaning step (30s BOE) and deposition of an a-Si:H (n)20nm layer on the whole rear surface. A passivation layer (described in section 4) is deposited at the front surface. W-PCD lifetime mappings (Figure 4) are performed on the test structures to compare both processes.

Figure 2. W-PCD lifetime mapping for different aSi:H stacks (not at same color scale). Left: a-Si:H (p)10nm (both sides); Middle: a-Si:H(i)5nm + (n)20nm (both sides); Right (cell structure): a-Si:H(p) 10nm (front) and a-Si:H(i)5nm + (n)20nm (rear) The fabrication process is finished by depositing a TCO layer (80 nm ITO) on both sides. Then 1,2m Al is sputtered at the rear side followed by the screen printing of a front Ag grid (low-T paste). 25 cm2 solar cells are finally obtained by using LASER scribing. Electrical parameters of the cells depending for two front emitter thicknesses are shown in Table 1. Table 1. Electrical parameters of 25 cm2 n-type HJ solar cells from I(V) measurements under AM1.5 illumination on polished substrates. Jsc (mA/ cm2) 29.7 29.0 Voc (mV) 637 641 FF (%) 73.1 71.0 (%) 13.9 13.2

Emitter a-Si:H (p) 10nm 15nm

Cells performances are mainly limited by the poor short-circuit current due to optical losses: no front

Figure 4. Detail of W-PCD lifetime mapping at low injection level of localized a-Si:H (p)30nm (Emitter finger) and a-Si:H (n)20nm . Left: MM process; Right: SP Process The 1mm2-size of the excitation spot does not result in the absolute value of the local lifetime but gives only an average value. However the clear difference in the lifetime mapping shows that standard cleaning procedures (dip in buffered HF) are not adapted to the SP-process localization in contrary to MM-process that allows better surface passivation. 3.2 Deposition conformity by SSRM measurement Another parameter to be tested is the conformity of aSi:H deposition after localization. We performed SSRM (Scanning Spreading Resistance Microscopy) and topography measurements to obtain an accurate mapping of 20nm a-Si:H (p) layers localized on n-type c-Si substrates. The poor conformity of the layers deposited through metallic masks had to be measured by SSRM (Figure 5) to have a wide enough line scan whereas topography measurements give more information about the mask edge after lift off (SP process Figure 6)

Figure 6. Topography of the a-Si:H (n) layer edge localized with SP-process Lift off layers for the SP process are designed to make a-Si:H films removal easy to do in standard acidic solution. Figure 6 indicates the mask edge effect on the aSi:H film topography after lift off: a part of the film follows the shape of the lift off layer and could cause problems of cleaning or in the plasma for further deposition steps. This probably explains the poor surface passivation obtained with this process (Figure 4). Nevertheless the conformity of the SP-process localization is about a few nm what is much better than for the MM-process. Another important parameter for the comparison of both processes is the number of steps necessary to make one localized deposition. Six steps are needed with SPprocess against only three steps for metallic masks process. Considering this aspect and the better passivation obtained we chose the MM-process to fabricate complete solar cells. 4 FRONT SURFACE LAYER OPTIMISATION

Both modelling and experimental results show the impact of front surface passivation on rear contacted cells performance [5][6]. For such a structure it is necessary to deposit on the front surface a passivation layer and antireflective coating at low temperature (200C) with an effective front surface recombination velocity SRV below 50cm.s-1. As [4] we first focus on a double layer aSi:H/SiNx but using a n-type amorphous layer instead of an intrinsic one. A trade off between the passivating properties and the thickness of the a-Si:H(n) layer has to be found to reduce its thickness as much as possible. Figure 7 shows the influence of a-Si:H(n) layer thickness on the effective SRV assuming a constant bulk lifetime value of 5 ms.
1000

Figure 5. SSRM mapping of a-Si:H (n) layer locally deposited with MM process on n-type c-Si wafer. The transition from low voltages (c-Si) to high voltage (a-Si:H) is not well defined but occurs on at least 50 m. Patterned metallic masks are simply laid on the wafer so unconformed PECVD deposition could explain a slight overtaking (25 m) of the amorphous layer under the mask edge.

100 Seff (cm/s)

20s a-Si:H (n) + 120s SiN 40s a-Si:H (n) + 120s SiN

10

60s a-Si:H (n) + 120s SiN 80s a-Si:H (n) + 120s SiN

1 1E+13

1E+14

1E+15 ECD (cm^-3)

1E+16

1E+17

Figure 7. Effective SRV for different a-Si:H(n)/SiNx layers deposited on both sides of n-type c-Si wafers At AM1.5 illumination Seff of 35cm/s can be reached with 6 nm a-Si:H(n) layer (40s deposition time) covered

J (mA/cm^2)

by SiN. This stack is chosen to be used on complete cells and test structures and absorption measurements will be soon performed in order to evaluate the short circuit current losses due to this double layer.

30 25 20

Ag
TCO

a-Si:H(n) C-Si (n) a-Si:H(p)

15 10 5 0 0 100 200 300 400 500 600 V (mV)

Al

Rear emitter cell IBC cell init. IBC cell after Anneal

Passivating ARC C-Si (n) a-Si:H(n) a-Si:H(i) TCO Al a-Si:H(p) Al

IBC HJ CELLS

5.1 Geometry and a-Si:H layers passivation The rear geometry is chosen considering alignment requirements and possible shunts effects. An overlay of 200 m between the different a-Si:H layers allows the whole surface to be entirely covered and passivated. The properties of the different layers are described in Table 2. Table 2. Geometrical and material parameters used for the fabrication of IBC HJ cells with MM-process. Width Thickness Material (m) Emitter 1400 30nm a-Si:H (p) Em. contact 1000 3m Al Base 1050 20nm a-Si:H (n) Base contact 250 80nm/3m ITO/Al Gap layer 650 20nm a-Si:H (i) Effective lifetime measurements of the different layers and cell structures are presented on Figure 8. The layers test structures consist in an n-type c-Si wafer sandwiched between the front side double layer described in Section 4. and the passivation layer which is tested. Rear emitter cells (front 20 nm a-Si:H (n); rear 30 nm aSi:H (p)) were fabricated as a reference with TCO and screen-printed Ag-grid on their front surface.
1000

700

Figure 9.

I(V) curves under AM1.5 illumination

Rear a-Si:H (i) 20 nm Rear a-Si:H (p) 30 nm

The IBC cell short circuit current (25.7mA/cm2) is much lower than for the rear emitter cell (28.6mA/cm2) due to the low fraction of contacted emitter (40% of the cell surface). LBIC (Laser Beam Induced Current) measurements Figure 10 indicates that current collection losses take place over the base fingers. However both types of cell reach almost the same VOC values about 640mV after annealing, proving the potential of IBC HJ cells fabrication using MM-process. Considering the high PFF (82%) value measured on Sinton SunsVOC for our IBC cells, the measured FF (49.8%) is only limited by high series resistance which is not visible for rear emitter cell (FF 76.1%). Considering its doping and the structure pitch the base resistance cannot be the main limiting parameter (for 2-D modelling results see this conference 2DV-1-35).We tried to deposit thicker metallization (6m Al) but no improvement could be observed on the FF value, indicating that a high contact resistance is probably the most limiting parameters for our IBC HJ cells efficiency. Interestingly this parameter appears to be more detrimental for IBC cells (localized contacts) than for full contacted structure. Thus to achieve IBC cells efficiency (8.2%) as high as rear emitter cells (13.4%) and even front emitter cells (13.9%) a decrease of the contact resistance as well as an increase of the emitter fraction are necessary.

Lifetime (s)

Rear a-Si:H (n) 20 nm Rear emitter cell IBC cell

100 1E+14

1E+15

1E+16

ECD (cm^-3) 1E+17

Figure 8. Lifetime measurement for different layers and structures used in solar cell fabrication A 30nm thickness is necessary for the a-Si:H (p) layer to reach good surface passivation but also to avoid its degradation during Al sputtering. The a-Si:H(i) layer show poor passivating properties compared to the doped layers but allows a 645mV implied VOC for the IBC cell precursor which is as high as the rear emitter cell precursor. We assume that the good front surface passivation of IBC cell compensates the detrimental effect of the rear localized a-Si:H(i) layer. 5.2 IBC HJ cell characterization Figure 9 shows I(V) tests for complete IBC HJ cell and reference rear emitter cell under AM1.5 illumination at 25C.

Figure 10. LBIC measurements of IBC HJ cell @405nm using Semilab equipment

CONLUSION

The Interdigitated Back Contact (IBC) structure represents a very promising way to improve HJ cells by avoiding optical and recombination losses. We developed a complete fabrication process for large area (28cm2) IBC HJ solar cells on n-type c-Si substrates with efficiencies up to 8.2%. The high VOC potential is proved using localisation steps through metallic masks for both a-Si:H layers and metallizations. Further developments include an improvement of the a-Si:H(i) film (higher lifetime) and the implementation of front surface texturing coupled to new metallic masks with higher emitter fraction to enhance JSC. Contact resistance still remains the most

critical limit that could be avoided using a higher doping in a-Si:H layers.

ACKNOWLEDGMENTS This work has been supported by ADEME, the French Agency for Environment and Energy Management and by by the French National Research Agency (ANR).

REFERENCES

[1] S. Taira, Y. Yoshimine, Proc. 22nd EUPVSEC (2007) [2] D. De Ceuster et al, Proc. 22nd EUPVSEC (2007) [3] R. Stangl, M. Bivour, Proc. 22nd EUPVSEC (2007) [4] M. Tucci et al , Thin Solid Films 516 (2008) 6771 [5] M. Lu et al., Appl. Phys. Lett. 91, 063507 (2007) [6] M. Lu, S. Bowden, Proc. 22nd EUPVSEC (2007) [7] N. Jensen et al, Prog. Photovolt. Res. Appl. 10 (2002) 1. [8] M. Lu, S. Bowden, Proc. 33rd IEEE PVSC (2008) [9] J. Szlufcik et al, Proc. IEEE Vol. 85, 5 p. 711 (1997) [10] N. Matsuki et al, Thin Solid Films 486 (2005) 210

DEVELOPMENT OF ZNO:AL TO REPLACE INDIUM TIN OXYDE FOR HETEROJUNCTION SOLAR CELLS : EFFECT ON SURFACE PASSIVATION QUALITY AND SHORT CIRCUIT CURRENT
Pierre Jean Ribeyron2, Nathalie Baclet1, HenriBoucher1, Thibault Desrues2, Aurlie Vandeneynde2, Florent Souche2,Anne Sophie Ozanne2 CEA/LITEN/DTNM/Laboratory of Solar Technology (LTS) 17 rue des Martyrs - 38054 Grenoble Cedex 9, France Phone: +33 4 38 78 56 63 E-mail: nathalie.baclet@cea 2 INES CEA/LITEN/Laboratory of Solar Components (LCS) BP 332 Savoie Technolac, 50 avenue du Lac Lman 73377 Le Bourget-du-Lac, France Phone: +33 4 79 44 46 43 E-mail: pierre-jean.ribeyron@cea.fr
1

ABSTRACT The aim of the present work is to develop an alternative to ITO for silicon heterojunction solar cells. This study is motivated by cost and availability issues linked with indium material. It is proposed to replace ITO by ZnO doped with aluminum. A process for ZnO:Al films deposition is presented based on conventional RF magnetron sputtering. Experimental parameters are adapted in order to meet the specifications in terms of electrical resistivity and transmission. The microstructure of the films is also characterized by XRD. Optical simulation is carried out at the cell level in order to evaluate the optimal thickness of the Al-doped ZnO film regarding transmission. Finally, a comparison is made between a standard ITO and the newly developed ZnO:A film regarding the properties of the solar cell.. Keywords: heterojunction, c-Si, TCO Transparent conductive oxides

INTRODUCTION

The rapid growth of photovoltaic industry leads to develop new structures with high efficiency potential on p-type and more recently n-type crystalline silicon (c-Si). The amorphous/crystalline heterojunction solar cells (a-Si:H/c-Si) concept is very promising and focuses more and more interest around the world. Sanyo is up to now, the only company that uses this structure at the production level and has demonstrated record efficiencies on n-type monocrystalline silicon at the R&D level [1]. This structure has many advantages: high open-circuit voltage values, low saturation current, and low processing temperatures [2]. Furthermore, a-Si:H/c-Si structures are well adapted for the reduction of wafer thickness due to the low mechanical stresses induced by the low temperature process. The active part of the device is basically produced by a low temperature growth (150-250C) of ultra-thin layers of amorphous silicon (a-Si:H) onto both sides of a thin crystalline silicon wafer-base. a-Si:H forms both the emitter and back surface field (BSF), while acting at the same time as a surface passivation. Indium tin oxide (ITO) is usually the standard Transparent Conductive Oxide (TCO) for this kind of high efficiency solar cells. Indeed, ITO has a low electrical resistivity and optical properties (refractive index and absorption) that are well adapted to silicon heterojunction solar cells. Nevertheless, because of indium shortage, this TCO is very expensive and is expected to become more and more rare and expensive in the coming years. This problem could affect the competitivity of the whole process. Moreover, alternative TCO can offer wider range of electronic work function, able to obtain better contact between doped layers and TCO, leading to better FF at the solar cell level.

An alternative to ITO would be to use ZnO. Standard undoped ZnO shows a good transparency, but its electrical resistivity remains too high for silicon solar cells with largely higher currents than thin film cells. So, ZnO is then doped with different elements such as boron, aluminum or gallium in order to reduce its electrical resistivity [ref Minami n4]. . The experimental study presented in this paper focuses on the development of Al-doped ZnO films (so-called ZnO:Al or AZO films) regarding electrical and optical properties. A first comparison between ITO an ZnO:Al films is presented regarding the performances of the associated solar cell. 2 DEVELOPMENT OF THE ZnO COATING

2.1 Experimental The ZnO:Al films were deposited on silica glass (Herasil I from HERAUS) in a conventional r.f. magnetron sputtering system. A sintered oxide ceramic disk of ZnO (diameter 200 mm) mixed with 2.5 wt.% Al2O3 was used as target. The base pressure in the sputter chamber was about 10-8 mbar. Pre-sputtering of 10 minutes was carried out at an argon gas pressure of 10-2 mbar in order to clean the target surface. The sputtering power was varied between 300 to 500 W, and the argon gas pressure was controlled from 3.10-3 to 1.10-2 mbar. The deposition rate varied from 0.3nm.s-1 to 0.6nm.s-1 depending on the sputtering power. All the films were deposited at room temperature. The most important parameters required for the application of ZnO:Al film as a transparent conductor are its low electrical resistivity and high optical transmittance. The thickness of the films was measured using a Dektak 8 surface-profile measurement system. The crystal structure of the films was determined by X-ray diffraction (XRD) analysis using a Siemens D5000 system with Cu K radiation ( = 0.1540562 nm). The Hall measurements were made with the Van der Pauw

technique at room temperature. The optical transmittance and reflectivity measurements were performed with a Perkin-Elmer 9 spectrophotometer. The refractive index and extinction coefficient were calculated from spectrophotometry measurements using the Optilayer Thin Film software [3]. 2.2 Electrical properties on ZnO:Al thin films The experimental film thickness was fixed to 100 nm, which is close to the optimal thickness of transparent conductive oxide coating on the heterojunction solar cells studied. This consideration is crucial since the resistivity decreases with the film thickness [4] which was confirmed in this study (not detailed here). The electronic properties of magnetron sputtered ZnO:Al films strongly depends on the deposition conditions. Electrical resistivity has been measured after a thermal treatment of 40 minutes at 200C under air, which is a possible treatment performed in the heterojonction solar cells manufacturing process (following electrical contact deposition by screen-printing). Fig. 1 displays the pressure dependence of the resistivity of films under variation of sputtering powers. A minimum in resistivity is observed for a sputtering pressure of 7.10-3 mbar, whatever the sputtering power. The resistivity does not strongly depend on the sputtering power for the range studied, as indicated on Fig. 2. The lowest resistivity (6.9.10-4 .cm) is achieved for a 90 nm thick ZnO:Al film deposited at 450W and a sputtering pressure of 7.10-3 mbar.
1,00E-02

interest, the mean transmittance increases after annealing, going from 79.6% to 81.7% for the ZnO:Al film deposited at 450 W and 7.10-3 mbar. The average transmittance decreases when the sputtering pressure increases, as illustrated on Fig. 4 and detailed in Table 1. The increase in sputtering power leads to a decrease in average transmittance, as previously observed [5].
100 90 80 Transmittance (%) 70 60 50 40 30 20 10 0 300 400 500 600 700 800 900 1000 1100 1200 Wavelength (nm) 450W 7.10-3 mbar 450W 7.10-3 mbar annealed

Figure 3: Influence of annealing on optical transmittance of ZnO:Al films.

100 90 80 Transmittance (%) 70 60 50 40 30 20 5.10-3 mbar annealed 7.10-3 mbar annealed 1.10-2 mbar annealed

resistivity (ohm.cm)

10 0
300 400 500

600

700

800

900

1000 1100 1200

1,00E-03

Wavelength (nm)

400W 500W 1,00E-04 2,00E-03

450W 300W

Figure 4: Optical transmittance of ZnO:Al films for different sputtering pressures Table I: Average optical transmittance on the 3501150 nm range for the annealed ZnO:Al films deposited on silica substrates (the transmittance includes the noncoated backside of the substrate) Sputtering pressure 400W 450W 500W 5.10-3 mbar 84.2 83.5 82.6 7.10-3 mbar 83.0 81.7 80.3 1.10-2 mbar 82.4 81.3 75.9 2.4 Microstructure XRD patterns of the ZnO:Al films deposited at an argon pressure of 7.10-3 mbar and a different sputtering powers are shown in Fig. 5. All the films are found to have the hexagonal zincite structure. A prominent (002) peak at ~34.4 indicates that the cristallite structure of the films is oriented with their c-axis perpendicular to the substrate plane. The average crystallite size of the films was estimated, using Scherrers formula, from the full width at half-maximum (FWHM) of the (002) reflection corrected from the diffractometer resolution. For all the ZnO:Al films deposited at a sputtering power between 400W and 500W the grain size varies from 35 to 40 nm, whatever the sputtering pressure, in agreement with previous work [6].

7,00E-03 1,20E-02 1,70E-02 sputtering pressure (mbar)

2,20E-02

Figure 1: Electrical resistivity versus sputtering pressure

1,00E-02

resistivity (ohm.cm)

3.10-3 mbar 5.10-3 mbar 7.10-3 mbar 1.10-2 mbar

1,00E-03

1,00E-04 250

300

350 400 450 sputtering power (W)

500

550

Figure 2: Electrical resistivity versus sputering power

2.3 Optical properties Fig. 3 shows the effect of thermal treatment on the total transmittance of ZnO:Al film deposited on silica substrates. One might note that the measurement takes into account the 4% of reflectivity at the back-side of the substrate. On the 350-1150 nm wavelength range of

relative intensity (arb. units)

(002) (103) (101) (102) (112) 500W

450W

400W

30

35

40

45

50

55

60

65

70

75

80

2 (degrees)

these layers, as measured by spectroscopic ellipsometry, is in the range of a few nanometers. After deposition of the front layer, the wafers were flipped for deposition of the back surface layer. Then, a Transparent Conductive Oxide was deposited by Magnetron sputtering (DC magnetron for ITO, RF magnetron for ZnO:Al) on the front of the wafer. Sputtered aluminium was deposited at the back. Finally, a low-temperature silver paste was screen-printed onto the TCO layer, and annealed for front contact formation. The whole process was performed at very low temperature (T<200C). 3.2 Effective lifetime results on solar cell like stacks The effective lifetime of the wafers have been measured by QSSPC after PECVD and TCO deposition. The results are presented on figure 7.
10000

Figure 5: X-ray diffraction pattern for ZnO:Al films deposited at 7.10-3 mbar and different sputtering powers. 2.5 Optical performances of the heterojonction Based on the optical and electrical properties of the ZnO:Al films, the experimental conditions that have been selected for the optimization of the film properties are a sputtering power of 450W and a sputtering pressure of 7.10-3 mbar. The index of refraction n, and coefficient of extinction k of this ZnO:Al thin film has been determined from spectrophotometry. The optical performances of the standard heterojonction made of a silicon substrate, a 10 nm thick a-Si:H layer and a ZnO:Al film have then been calculated. The thickness of the ZnO:Al film has been fixed to 89 nm to minimize the reflectivity at 650 nm. Fig. 6 displays the optical performances of the heterojonction on the 350-1150 nm range together with the related average transmittance, reflectivity and absorption (without taking into account the substrate backside).
120 110 100 90 80 70 60 50 40 30 20 10 0 300 400

Effective lifetime (s)

AZO textu ITO textu

Voc implied : 657 mV Voc implied : 660 mV

AZO polished Voc implied : 674 mV ITO polished

1000 Voc implied : 678 mV

100

1E+14

1E+15

1E+16

Ninj (cm-3)

1E+17

Figure 7: Effective lifetime versus injection level for 4 different stacks with the same silicon amorphous layers : AZO textured and polished, ITO textured and polished Passivation level is very good for stack like solar cells with effective lifetime up to 1.8 ms for polished samples, and up to 1 ms for textured ones, even if there are no intrinsic layer at the interface between crystalline silicon and doped amorphous layers. Textured samples demonstrate very good level passivation, demonstrating that the cleaning process after texturation is efficient. Voc implied are in the same range for ITO and AZO, even if ITO layers seems to give slightly larger Voc (difference of about 4mV which may be not significant). Lifetime of AZO samples seem to be slightly higher than ITO ones. These passivation results will be checked through measurements on complete solar cells in the next section. 3.3 Solar cell results : AZO/ITO comparison on polished and textured samples p type, 5*5 cm , double heterojunction solar cells were manufactured. The solar cells were measured by I(V) under AM1.5 spectrum. The solar cells were either polished or textured. I(V) results are presented in table 2. Solar cells with AZO show better Voc on either polished or textured samples with Voc up to 668 mV on polished samples. The current for polished sample with AZO is lower than with ITO ones and indicates that some optimisation is still possible. Indeed, this non appropriate thickness tends to increase the effective reflectivity, obviously reducing the short circuit current. This effect is

transmittance

reflectivity

absorption

T(350-1150nm) = 80,4% R(350-1150nm) = 13,2% A(350-1150nm) = 6,4%

500

600 700 800 900 Wavelength (nm)

1000 1100 1200

Figure 6: Calculated optical properties of the heterojonction made of silicon substrate, 10 nm a-Si:H and 89 nm ZnO:Al

COMPARISON BETWEEN ITO AND ZNO:AL FILMS ON HETEROJUNCTION SOLAR CELLS

3.1 Experimental p type c-Si wafers (<100> oriented, W=300m, =14-22 .cm) were used for this study. The emitter and the back surface field were deposited on polished and textured samples. Textured samples (KOH wet chemical treatment) were cleaned after texturation.[7, this conference] Phosphorous doped a-Si:H layer was deposited at the front surface directly onto the cleaned substrate. Boron doped a-Si:H layer was deposited at the back. Both doped a-Si:H (n or p) layers were deposited by a capacitively-coupled PECVD system at an RF frequency of 13.56 MHz under various conditions. The thickness of

less important on textured samples because of the texturation which tends to decrease the difference in reflectivity between ITO and AZO-based solar cells. One may then expect a larger short circuit current for solar cells based on AZO with optimised thickness.. The fill factor is not so good in both cases. The fill factor is affected mostly by the series resistance. This is partly due to the back surface which is not sufficiently doped. AZO polished samples have the worst Fill factor but the situation is improved for textured ones. This means that a significant part of the series resistance is linked with contact resistance problem between the different interfaces. Fill factor is slightly higher in the case of ITO because the resistivity of ITO is better than AZO. This directly affects the series resistance, thus the fill factor. Finally, textured sample show equivalent efficiency for ITO and AZO, demonstrating that this transparent conductive oxide is really a potential candidate to replace ITO.

REFERENCES [1] T. Sawada, N. Terada, S. Tsuge, T. Baba, T. Takahama, K. Wakisaka, S. Tsuda and S. Nakano, Proc. of 1st World Conference on Photovoltaic Energy Conversion, Hawaii, USA, (1994), p. 1219. [2] M. Taguchi, K. Kawamoto, S. Tsuge, T. Baba, H. Sakata, M. Morizane, K. Uchihashi, N. Nakamura, O. Oota, Progr. Photovolt. Res. Appl. 8 (2000) 503. [3] http//www.optilayer.com [4] T. Minami, H. Sato, H. Nanto, S. Takata, Japanese Journal of Applied Physics 24(10) (1985) L781-L784. [5] J. Lee, D. Lee, D. Lim, K. Yang, Thin Solid Films 515 (2007) 6094-6098. [6] R. Cebulla, R. Wendt, K. Ellmer, Jounal of Applied Physics 83 (2) (1998) 1087-1095. [7] P-J.Ribeyron and al, 2CV4.35, this conference

Voc (mV) Jsc(mA/cm) FF(%) (%) AZO 29.1 69 13.3 668 polished AZO 660 32 73.8 15.6 textured ITO 662 30.5 73.8 14.9 polished ITO 653 15.55 32.2 74.3 textured Table II: I(V) measurements under AM1.5 spectrum of 5*5 cm double heterojunction solar cells, polished or textured with ITO or ZnO:Al at the front 4. CONCLUSION The present work was devoted to the optimization of a sputtering process for the deposition of ZnO:Al films that are integrated in heterojunction solar cell process. The influence of different experimental parameters has been demonstrated on both electrical resistivity and optical properties. AZO and ITO have been compared in a complete solar cell and preliminary results show that AZO is a potential candidate to replace ITO for heterojunction process. This conclusion will be re evaluated with higher efficiency solar cells. Further work will be performed to improve both the optical and electical properties of the transparent conductive oxide by using a multilayer coating that includes a very thin (a few nanometers) metallic layer such as silver in order to improve the resistivity without affecting the optical properties.

ACKNOWLEDGMENTS This work has been partly supported by the French national research agency (ANR). The research leading to these results has received funding from the European Community's Seventh Framework Programme [FP7/2007-2013] under grant agreement n211821 (HETSI project).

Functional All-Silicon Nanowire Solar Cells

Philippe Thony, Rgis Delsol, Claude Jaussaud, Nathalie Rondel, Pascal Gentile INES CEA/LITEN/Laboratory of Solar Components (LCS) BP 332 Savoie Technolac, 50 avenue du Lac Lman 73377 Le Bourget-du-Lac, France Phone: +33 4 79 44 45 97 E-mail: philippe.thony@cea.fr Emmanuelle Rouvire, Sverine Poncet, Simon Perraud, Cline Mouchet CEA/LITEN/Laboratory of Components Hybrides (LCH), 17 rue des martyrs - 38054 Grenoble cedex 9, France Pascal Gentile CEA/INAC, 17 rue des martyrs - 38054 Grenoble cedex 9, France Thierry Baron CNRS/Laboratory of Microlectronic Technology (LTM), 17 rue des martyrs - 38054 Grenoble cedex 9, France Phone: +33 4 79 44 45 97 E-mail: philippe.thony@cea.fr ABSTRACT A functional all-silicon nanowire solar cell was demonstrated. Silicon nanowires were grown by Chemical Vapour Deposition (CVD), using the vapour-liquid-solid technique with gold as a catalyst. N-type and p-type doping of the nanowires was achieved by introducing gaseous precursors inside the CVD growth chamber. Evidence of efficient doping has been shown through electrical transport measurements on single nanowires and Secondary Ion Mass Spectrometry (SIMS). A fully integrated process for fabricating nanowire-based solar cells was developed, including nanowires encapsulation and electrical contact formation. Several test devices were built with different electrical architectures. Test devices consisting of n-type silicon nanowires grown on a p-type silicon substrate were characterized. A rectifying behaviour was observed in the dark I-V characteristics due to the p-n junction formed at the nanowire layer / substrate interface. Under illumination, a photovoltaic effect was obtained with I(V) curves showing the typical behaviour of a photovoltaic cell. This result represents the first photovoltaic cell based on a homojonction structure build with silicon nanowires. Keywords: silicon solar cells, silicon nanowires, vapour-liquid-solid (VLS) growth.

INTRODUCTION Constant enhancement in solar cell efficiency is requested to lower the cost of photovoltaic energy. Recent progress in micro- and nanotechnology offers several advanced materials that show a great potential for raising the conversion efficiency above todays performance of silicon wafer cells. Different nanomaterials have been investigated, such as silicon quantum dots assembled in a superlattice structure [1], or silicon nanowires grown on silicon wafers or other types of substrates [2-4]. We believe nanostructures and nanomaterials will bring better performance in different ways: better antireflection properties, energy band gap engineering, multi-exciton generation These characteristics enlarge the possibility to develop siliconbased solar cells. In this paper, we focus on silicon nanowires technology. Different possibilities are offered to use nanowires in photovoltaic solar cells, depending on the diameter of the wires. Nanowires with very small diameters show quantum confinement that leads to a band gap larger than bulk silicon band gap [5]. This occurs with a diameter below 5 nm. A nanomaterial based on such nanowires will allow designing all-silicon multijunction solar cells. With larger diameter in the range of some tenth of nanometres, interesting architectures will be built by adjusting diameter to material quality [3]. With further increased diameter, specific optical effects could be achieved. The structure can be considered as a 2D photonic crystal, especially interesting for light guiding effects and optical confinement [6].

For different reasons, these one-dimension structures represent a very interesting alternative to quantum dot structures: nanowires show similar optical and spectral characteristics as quantum dots (quantum confinement exists and provides similar effect on optical properties); nanowires show a rather good conductivity in the longitudinal direction, there is no need to rely on tunnelling effects to collect photogenerated carriers; passivation of the nanostructured surface needs further investigation, recombination rate could indeed be significant because of the important developed surface and interface [6]; doping mechanism in quantum structures and in semiconductor nanomaterial is an ambitious question with issues like charge binding energy or requested concentration level. DEVICE DEFINITION Several device parameters have first-order impact on the solar cell performances: Nanowire diameter on energy band structure, driving mainly optical absorption when quantum confinement occurs; Nanowire diameter on series resistance, associated with silicon material quality (resistivity); Density and length of nanowires, impacting both absorption and electrical performances, mainly series resistance; Interface between the nanowires and the encapsulating material, impacting photogenerated carriers recombination at the nanowires surface.

Before developing a technology to control the nanowire geometry, we use a first process to assess the conductivity of silicon nanowires, the doping efficiency and the electrical contact design. This first process will be complemented with patterning techniques to achieve either dense nanowires with very small diameter, or larger nanowires placed on a regular pattern. NANOWIRE GROWTH VLS method The basic principle of the VLS growth mechanism, which is based on cracking of gaseous silane (SiH4), is the following [7]. A silicon substrate is covered with gold nanoparticles, which act as a catalyst for the growth of nanowires. The substrate is then heated at a temperature above the gold/silicon eutectic point (363C), in order to transform the gold nanoparticles into gold/silicon liquid droplets. The VLS growth of a silicon nanowire consist of three steps: (i) silane molecules from the vapour phase are adsorbed and cracked on the surface of a liquid droplet, leading to the incorporation of silicon atoms into the droplet (ii) silicon atoms diffuse through the liquid droplet (iii) silicon atoms condensate to form the crystalline silicon nanowire, which grows with the liquid droplet at its tip. The silicon column shows a crystalline structure equivalent to the structure of the substrate. We performed the VLS growth of silicon nanowires on silicon single-crystal substrates, in a CVD reactor at reduced pressure (10 Torr), at a temperature of 500C, with a flow of SiH4 diluted in H2. The crystalline orientation of substrates is mainly <100>, even if some growths have been performed with <111> substrates. Catalyst nanoparticles The gold nanoparticles were obtained by either dewetting of a gold thin film, or deposition of colloidal gold particles. In the first approach, a 2-nm-thick gold film was grown by evaporation on a deoxidized silicon substrate. For dewetting the gold thin film, the substrate was annealed during 10 min under a flow of H2. The annealing temperature for obtaining the highest particle density and the smallest particle diameters was found to be about 550C. The nanowire diameter distribution obtained by this dewetting technique is quite wide, and the most frequent diameter is about 40 nm (see figure 1).

the colloidal particles, the substrate was annealed during 10 min at 500C under a flow of O2, and then during 10 min at 550C under a flow of H2. The nanowire diameter distribution obtained with the colloidal particles, centred at about 40 nm, is much narrower than that obtained by the dewetting technique (see figure 2b).

(a)

(b) Fig. 2: SEM images of (a) deposited colloidal gold particles (b) silicon nanowires grown from the colloidal particles.

Fig. 1: Diameter distribution of silicon nanowires grown by the VLS technique, after dewetting of a 2-nm-thick gold thin film. Inset: scanning electron microscope (SEM) image of the silicon nanowires.

We also assess the possibility to grow silicon nanowires from smaller colloidal nanoparticles, with diameter 2, 5 and 10 nm. To realize nanowires small diameter, some conditions must be obtained: small diameter catalyst, low growth temperature to prevent the migration of the catalyst on the top of the nanowires and high pressure silane [8]. With small nanoparticles, we could grow nanowire that should present quantum confinement effects. These particles are deposited with an electrochemical method that allows controlling the amount and so the density of the deposited particles. For that purpose, particles are diluted in water and a low voltage is applied between the solution and the silicon substrate. The growth temperatures are from 400C to 550C and silane flow are from 50sccm to 200sccm with 3l/mn H2 flow and 20mbarr pressure. The figure 3 shows the ultimate Si nanowires growth obtained from 2nm gold colloid at 400C and 200sccm silane flow. Unfortunately, such small wires are very unstable. Mechanical vibrations have a very strong impact on the wires that are blown away. Optical characterization was also very difficult because of the reactivity of these small structures that are burnt with the illuminating probe beam even at very low intensity.

In the second approach, colloidal gold particles with a diameter of 20 nm were deposited on a silicon substrate. A density as high as 600 particles/m can be obtained (see figure 2a). In order to remove the organic coating of

needs to be tuned to account for interactions between doping species and catalyst. After the VLS growth, the gold catalyst is etched by using a KI/I2 solution, and the doping impurities were activated by a rapid thermal annealing process (750C during 5 minutes). NANOWIRES ENCAPSULATION IN A MONOLITHIC LAYER We developed an encapsulation process based on sol-gel techniques, to ensure mechanical stability of the nanowire field and to allow further processing such as front contact formation. Basically, silica sol-gel material provides a good capability for mechanical stabilisation. Its performance for electrical passivation needs to be checked. The combination of thermal oxidation of the wires and of silica encapsulation will provide a good lateral continuity in the structure. Attempts were made with several materials, such as spin-on glass (SOG, see figure 5) but also with organic polymers. After deposition, a thermal treatment is needed for solgel polymerisation and for gel densification. We then obtain a layer of silica if SOG is employed. This layer shows a rather low porosity. Nevertheless, the region close to the bottom of the layer show some voids.

Fig. 3: Silicon nanowires with small diameters, grown with deposited colloidal copper particles.

We have also used copper as catalyst, the Cu-Si phase diagram is more complicated with eutectic at 801C. With we must used a dewetted Cu film with 900C annealed temperature, and growth temperature from 750C to 850C.

1 m

20 m

Fig. 4: SEM Images of Si Nanowires growth from Cu catalyst.

We conclude that direct growth of quantum nanowires is probably a very difficult process. Obtaining a wide and regular field covered with 2 nm in diameter nanowires appears a difficult technique at this point. We suggest to grow nanowires with a wider diameter and then to reduce the silicon diameter by creating a core-shell structure with a controlled oxidation. Research is ongoing with the study of thermal processing of nanowire. DOPING OF NANOWIRES We tried several techniques to achieve active doping of the silicon nanowires: either by introducing impurity during the VLS growth or by thermal diffusion after growth. The first solution is preferred for process effectiveness. We can grow the wires at a rather low temperature (500C) and integrate in the same processing step the doping of the wires. Activation annealing is then performed. The second solution is an interesting alternative compatible with our encapsulation technique (see below). The doping impurity is added on the nanowires field as a component of the encapsulation stack. A thermal treatment let the impurity diffuse inside the silicon nanowires. In this process, we have used a silica sol-gel material containing a phosphorous or a boron precursor which is spin coated on the nanowires field. A high temperature annealing is then driving phosphorous or boron atoms in silicon, along with the stabilisation of the silica gel. In the following, we focus on in-situ doping during the VLS growth. Doping was achieved by using gaseous phosphine (PH3) and diborane (B2H6), for n-type phosphorous and p-type boron doping, respectively. Some trials have also been performed with Trimethyl aluminium (TMA) for aluminium doping. First attempts showed that there an interaction occurs between the Si/Au droplet and the gaseous precursor. Depending on the nature and the partial pressure of the precursor inside the CVD chamber, it seems that the volume of the catalyst droplet varies. So as to maintain a constant diameter of the grown wire, we need to carefully adjust gas flow and growth temperature. The growth process

Fig. 5: Spin-on-glass (SOG) encapsulation of silicon nanowires. Inset: perpendicular and slanted wires after polishing.

After wet deposition of the encapsulating material, we proceed to a chemical-mechanical polishing of the samples. We fabricated nanowires with a length of 2-4 micrometers in order to get an encapsulated layer of 1 to 2 micrometers in thickness. The corresponding material removal is enough to allow slanted wires to emerge at the top surface. We think that such a thickness is enough to show significant optical absorption. Optical measurement on a quartz substrate is pending. ELECTRICAL ASPECTS Electrical transport and doping level Resistance measurements on single silicon nanowires emerging from the SOG layer were performed by using a Scanning Spreading Resistance Microscopy (SSRM) technique. This technique is based on an atomic force microscope (AFM), the back side of the sample being polarized by a voltage source. A current is detected by the tip which scans the up side of the sample and this measurement provides spreading resistance information after calibration of the measurement chain (tip resistance). This set-up gives depth images and resistance images.

log( |J| ) (J in mA/cm)

The n-type silicon nanowires (phosphorous doping) emerging from the SOG layer appear in both images (see figure 6). The resistance of undoped nanowires was found to be about 10 M , whereas the resistance of ntype nanowires (PH3/SiH4 ratio of 210-3 during the VLS growth) was found to be about 10 k .

1 0 -1 -2 -3 -4 -5 -6 -1 -0,8 -0,6 -0,4 -0,2 0 U (V) 0,2 0,4 0,6 0,8 1 full p-type structure n-type w ires on p substrate

Fig. 7(a): Dark current-voltage characteristics of a test device, consisting of n-type silicon nanowires grown on a p-type silicon substrate. Fig. 6: Scanning spreading resistance microscopy (SSRM) measurements: topography (left) and resistance (right) images of the SOG surface.

Simple calculation shows that this result leads to a resistivity of some 10-3 .cm, depending on exact diameter of the wire. This value corresponds to a doping level close to 1020 at./cm3. This value is in coherence with gas ration we use for the growth of doped wires. Electrical contacts We used evaporated metallic layers for electrical contacts. The back contact was formed by evaporating and annealing a 100-nm-thick aluminium layer. The front contact consisted of a 9-nm-thick aluminium layer, thin enough to be transparent in the visible spectrum. This devidces are manly designed for electrical testing. The thickness of the front aluminium layer is a compromise for non-negligible optical transmission and acceptable electrical resistance. But the uncertainty of the deposited thickness is important and we can not estimate the optical transmission with enough precision to provide energy conversion efficiency. The front contact was deposited after chemical mechanical polishing of the SOG surface. Recent attempts to replace the aluminum front contact by a transparent conductive oxide show promising results. SOLAR CELL CHARACTERIZATION Test devices were fabricated and characterized. For the results presented here, we use gold dewetting technique to obtain the catalyst droplets. P-type <100> silicon substrates were used, with a resistivity in the range 1422 .cm. We designed a single junction solar cell, consisting of n-type silicon nanowires (PH3/SiH4 ratio of 210-3). Typical I-V characteristics in dark and under illumination (AM1.5G spectrum) are shown in figure 6. The current-voltage characteristics show a rectifying behaviour, ascribed to the p-n junction formed at the nanowire/substrate interface. Note that we checked that the current-voltage characteristics of devices consisting of p-type silicon nanowires grown on a p-type silicon substrate do not exhibit such a rectifying behaviour.

Under a 1 sun illumination, an open-circuit voltage and a short-circuit current of 125 mV and 0.53 mA/cm are obtained, respectively. The I(V) curve measured with illumination show a very important impact of series resistance. The emerged nanowires density appears to be rather low on figure 6. This impacts the resistance. On top of that, we can expect that most of the light is absorbed inside the bulk silicon. The current is forced through the doped nanowires and is especially affected by the series resistance of wires.
0,7 0,6 0,5 0,4 0,3 0,2 0,1 0 -0,1 -0,2 -0,3 -0,4 -0,5 -0,1 0,01 sun 0,1 sun 0,3 sun 0,5 sun 1 sun

J (mA/cm)

0,1

0,2

0,3

0,4

0,5

U (V)

Fig. 7(b): Current-voltage characteristics of same test devices in fig. 7(a), with standard AM1.5G spectrum illumination.

CONCLUSION We have fabricated functional photovoltaic cells based on a nanostructured layer. This layer is a field of silicon nanowires encapsulated in a dielectric sol-gel material. A full process has been developed, including encapsulation, doping, and full-sheet metal electrodes deposition. Samples have been electrically tested, in dark and under illumination. A photogenerated current has been obtained and typical diode characteristics are shown. This result are to our knowledge the first successful test devices to obtain a photovoltaic cell based on a homojonction structure based on grown silicon nanowires. In order to improve the performances of the nanowirebased solar cells, several device parameters have to be optimized. On top of them, emerged nanowires density has to be improved by adjusting the polishing depth and by increasing the density of catalyst droplets. Doping incorporation inside the nanowires seems to be low: additional measurements based on SIMS results will be analyzed. The quality of the nanowire/substrate interface

is also questionable, along with the front contact resistance. For our next test devices, we will use a transparent conductive oxide that will allow a better control of energy measurement. ACKNOWLEDGEMENTS The authors wish to thank the ANR (National Agency for Research) for supporting research project, D. Mariolle for SSRM measurements, and C. Nol for deposition of colloidal gold particles. REFERENCES [1] E.-C. Cho et al., Nanotechnology 19, 245201 (2008). [2] B. M. Kayes et al., J. Appl. Phys. 97, 114302 (2005). [3] L. Tsakalakos et al., Appl. Phys. Lett. 91, 233117 (2007). [4] Th Stelzner et al., Nanotechnology 19, 295203 (2008). [5] D. D. D. Ma et al., Science 299, 1874 (2003). [6] H. Fang et al., Nanotechnology 19, 255703 (2008). [7] R. S. Wagner and W. C. Ellis, Appl. Phys. Lett. 4, 89 (1964). [8] P. Gentile, T. David, F. Dhalluin, D. Buttard, N. Pauc, M. Den Hertog, P. Ferret, and T. Baron., Nanotechnology, 19(125608):125608, 2008.

OUTDOOR EVALUATION OF THE ENERGY PRODUCTION OF DIFFERENT MODULE TECHNOLOGIES

Jens Merten, Lionel Sicot, Yves Delesse, Antoine Gurin de Montgareuil INES CEA/LITEN- Laboratory for Solar Systems (L2S) BP 332 Savoie Technolac, 50 avenue du lac Lman, 73377 Le Bourget-du-lac, France Phone : +33 (0)4 79 44 45 60 E-mail: jens.merten@cea.fr

ABSTRACT PV-modules of identical nominal power do not perform equal. This paper examines the irradiance dependence of module performance for wafer based and thin film silicon modules. Technology factors limiting the performance are identified. For this purpose, data from variable illumination measurements (VIM) of the I/V curve are analysed using a modified two diode model. The VIM-method is a simple tool revealing rich information about PV modules. This allowed for a clear separation of five technological key issues determining the performance of the modules: the rectifying capacity, the second diode, the recombination in the i-layer for amorphous silicon modules, the series and parallel resistances.

Keywords: Module Ratio 1; PV-Modules - 2; VIM 3 Introduction The issues of interest in PV modules for investors in PV plants can be divided in three areas: Nominal power: What are the deviations of the real module power PSTC measured under standard conditions and their nominal power Pnom which has been paid for? Energy production: How much energy is produced by the PV modules of a certain technology on the specific site? Aging: How long do the modules last, what is the annual loss in production during the module lifetime? This paper focuses on the second aspect, the energy production, which is sometimes quite different for the existing module technologies, although the real module power PSTC measured under standard conditions is identical (a recent example is the productivity ranking published by Photon International [1]). Cause is the different sensivity of each module technology on temperature, solar spectrum, incident angle, and on the overall value of the solar irradiance. A simple method called MotherPV has been proposed in order to be able to preview the energy production of any PV module on any site taking into account all affecting parameters [3]. One key element of this method is the determination of the irradiance dependence of the overall module efficiency. This paper examines this dependence for different module technologies which is closely related with the manufacturing process of the device. A clear link of this dependence with the manufacturing process has been already established for thin film silicon modules in
Modules Resistances Code Techno Rs Roc 1 sun cm2 cm2 A c-Si 1.3 2.3 B c-Si 1.2 2.3 C p-Si 1.35 2.5 D c-Si 1.3 2.3 E1 p-Si 0.2 0.5 E2 p-Si 0.2 0.5 F p-Si 1.4 5.8 G1 a-Si 17 31 G2 a-Si 8 21 H a-Si 10 56

references [2,4]. A simple method to detect the issues of the manufacturing process which limit the module performance has been elaborated and demonstrated. It consists of variable illumination measurements (VIM) of the modules I/V curve and their interpretation using a modified two diode model [2]. These variable illumination measurements (VIM) can be performed outdoors [4]. Therefore, two novel test benches for the continuous outdoor scan of I/V curves have been set up at the headquarter of INES close to Chambry and in Cadarache in the South of France. VIM allows for a clear separation of the influence of technological ageing factors (such as deterioration of the contact layers) from degradation due to the StblerWronski effect in amorphous silicon modules [2]. In this paper, the same approach is also used for wafer based silicon modules. Experimental Procedure 1.1. The modules examined Six types of modules based on silicon wafer technology and two types of amorphous silicon modules have been exposed outdoors on the I/V test bench at INES in Chambry (Table I). In order to demonstrate the spread between modules of the same series, data from two modules of identical type are examined for the module types E and G. The thin film module G is a single junction module, whereas the module H is a multijunction module. Module F consists of one single cell only. 1.2. Outdoor test facilities at INES INES counts on two outdoor test facilities for PV modules:
n2 5.5 4.8 4.0 5.5 3.2 3.2 4.0 8.0 3.5 7.0 STC Characteristic Isc Voc FF Efficiency mA/cm2 V/cell per cell 31 0.60 0.76 14.2% 30 0.60 0.74 13.3% 31 0.59 0.72 13.2% 31 0.60 0.76 13.9% 30 0.57 0.74 12.5% 31 0.58 0.74 13.4% 32 0.61 0.68 13.2% 10 0.75 0.36 2.6% 11 0.78 0.43 3.5% 5.7 0.69 0.55 6.5%

Rp k cm2 6 13 22 9 >9 >15 >21 4 >100 116

Diode Parameters I01 n1 I02 nA/cm2 A/cm2 10 1.1 18 10 1.1 5 300 1.3 4 10 1.1 11 150 1.2 9 100 1.2 4 200 1.3 3 50 1.6 23 0.05 1.2 0.45 0.8 1.2 2

 At the INES Headquarter, close to Chambry, in the Rhne-Alpes region, with a soft alpine climate, At the solar outdoor facility in Cadarache, in the region of Provence, with a Mediterranean climate and frequent clear skies. Each I/V test bench has 24 channels for the continuous scanning of the I/V curve of the modules, which are connected by relays to an electronic load. These data are stored together with the meteorological data in a database for post-treatment and analysis. Aim of these I/V test benches are: Monitoring of the characteristic parameters of the PV modules under natural irradiance. Precise data for novel models for PV modules of any technology Determination of the nominal power and the energy production Comparison of the performance of different technologies of PV modules under different climatic conditions In-situ monitoring of degradation processes. 1.3. Variable Illumination Measurements The Variable Illumination Measurements (VIM) method, consists of scanning the I(V)-curve under logarithmically varying illumination levels. This variation is achieved by using the natural variation of the outdoor illumination. Variations in the illumination spectra are not critical for the analysis done here [5] except for the multijunction thin film module. The measured I(V)-curves are condensed into characteristic parameters which are: the short circuit current Isc, the open circuit voltage Voc, the fill factor FF, the "open circuit resistance" Roc=(V/I)I=0, which is used to measure the series resistance Rs, and the "short circuit resistance" Rsc=(V/I)V=0, which is not equal, but used to measure the parallel resistance R p. These parameters are presented as a function of illumination level indicated by the short circuit current Isc, (Figures 2 to 7). The data can be interpreted straight forward using a simple model described in the next section. All data are normalised to one single cell of one square centimetre. Aim is not to apply any temperature correction to the curves measured, but to apply the temperature correction to the extracted parameters. In this paper, temperature correction is applied to Voc, Isc, and the module
0,7

Figure 1: The equivalent circuit of the model used. The current sink (dashed) stands for recombination losses in the i-layer of amorphous silicon modules. power using the coefficients from the datasheets. The model Figure 1 shows the equivalent circuit of the model used for the interpretation of the data. Each element shown represents one technological issue relevant for the modules which is expressed in the following equation (1):

I (V ) = I ph + I ph + I01e
V IRs n1kT / e

di2 eff [Vbi (V IRs )]

V IRs n2kT / e

(1)

1 + I02e

V IR s 1 + , Rp

where Iph is the photogenerated current, I01 the saturation current of the first diode, n1 its quality factor, I02 the saturation current of the second diode, n2 its quality factor, e the elementary charge, k Boltzmanns' constant, T the modules temperature, Rp the parallel resistance and Rs the series resistance. The terms on the right side of this equation are detailed here: The current source Iph represents the photogeneration. Note, that Iph is the only parameter requiring the measurement of the illumination level; it is obtained from Isc-data at solar noon under clear irradiance conditions. the current sink represents recombination losses in the i-layer of amorphous silicon modules [2]. Note that this term is not used for wafer based modules. The loss current is directly proportional to the photogenerated current Iph, and is directly related to the effective product eff in the i-layer. This term is the photogenerated current Iph multiplied by the ratio of the i-layer thickness di over the carrier drift length

eff [Vbi (V IRs )]/ di

in

that

layer

(the

electrical field is determined by the voltage over the i0,8

G2 (a-Si)
0,7 0,6

H (a-Si)

0,6

A (c-Si)
Voc [V/cell] 0,5 0,4 0,3 0,2 0,1

0,5

First Diode
Voc [V/cell] 0,4

G1 (a-Si)

0,3 0,2

Second Diode

0,1 0 0,001 0,01

Parallel Resistance

0,001

0,01 0,1 Solar Irradiance in terms of Isc [Isc / Isc@STC]

0,1 1 Isc [mA/cm2]

10

100

Figure 2: Example for the determination of the diode parameters from VIM-data.

Figure 3: Open circuit VIM-data for three a-Si modules. Note that the short circuit current has been normalised to the value under one sun. G1 and G2 are single junction modules of the same type, H is a multi-junction module.

layer, which is the built-in voltage Vbi less the cell voltage V-IRs, and its thickness di). the first diode sows the rectification quality of the cells used, the second diode term presents additional losses the resistances represent shunt current paths (Rp) or contact resistances (Rs) respectively. The parameters of this model are extracted straight forward from the VIM-data as described in [2] and the following chapter. There is no need of numerical fitting procedures. The parameters of the modules examined are shown in the Table. Results 1.4. Open circuit voltage and junction diodes The data of both thin film and wafer based modules need to be modelled by two diodes. For each diode, there is an illumination regime, where the effect of the other elements can be neglected under open circuit conditions. A logarithmic relation of the open circuit voltage with the short circuit current can be approximated from the model: Voc = (n1kT/e) ln(Isc / I01). (2) Voc = (n2kT/e) ln(Isc / I02). (3) For amorphous silicon modules, an approximation leading to this relation is that the second term on the right side of equation (1) has the same value in open circuit and short circuit conditions. The regimes of equation (2) and (3) can be seen in Figure 2, where three regimes of illumination can be distinguished: Under high illumination (more than 1 mA/cm2), the open circuit voltage is determined by the first diode (equation (2)), which has a quality factor close to one for high quality modules, for both wafer based and amorphous silicon technology (Table) Under intermediate illumination, the open circuit voltage is determined by the second diode (equation (3)) with a quality factor of at least three. Note that the second diode term does not affect the outdoor performance of the modules examined, as its effect can be perceived only at illumination levels less than 0,1 suns. However, the optimisation of the second diode is very important for PV modules which are expected to deliver power at small illumination levels (typical small scale applications). Under lowest illumination (below 0.02 mA/cm2), the complete I/V curve is determined by the parallel resistance which limiting the module performance, and
1,E+06

Voc is proportional to Isc (Voc = Rp Isc). 1.5. The short circuit resistance Rsc The slope Rsc=(V/I)V=0, is often said to be equal to the parallel resistance Rp of the module. However, Rsc is affected by both Rp and recombination losses in the i-layer of amorphous silicon modules. Figure 4 shows the Rsc data for two amorphous silicon modules as a function of the illumination level, which is expressed as the ratio of the measured short circuit current and the short circuit current at STC. Two regions can be distinguished: one is determined by the recombination losses in the i-layer, the other by the parallel resistance. 1.5.1. Recombination losses in the i-layer Towards high illumination levels, Rsc is inversely proportional to the illumination level. In this regime, Rsc is dominated by the recombination losses in the i-layer of amorphous silicon modules [2] according to:

V 1 Rsc = eff bi I ph di

(4)

The corresponding illumination regime is marked in Figure 4, Rsc is inversely proportional to Isc. Once the thicknesses of the i-layers and the built in voltages are known, the effective -product eff in the i-layer can be directly determined using equation (4). 1.5.2. Shunt current paths and pin holes It is only towards lowest illumination levels that Rsc is directly equal to the parallel resistance and not affected by other terms of the model [2]: Rp = Rsc=(V/I)V=0 (5) Low parallel resistances show up parasitic ohmic current losses which are caused by pin holes in amorphous silicon modules and reduce the module performance under very low illumination conditions. The corresponding illumination regime is marked in Figure 4, where Rsc is independent of Isc. For the module G1, the low parallel resistance shown affects the performance of the module already below 0,3 suns. This means that the module G1 shown in Figure 4 cannot be recommended for regions with low irradiance, and definitively not indoor applications, where the illumination level is far below. However, the other module G2 examined has a parallel resistance which more than 25 times higher (Table). This means that we have detected a
8 7

G2 (a-Si)
1,E+05 Rsc [Ohm cm2]

Recombination Losses in i-layer

Roc [Ohm cm2]

6 5 4

Roc @ 1 sun

H (a-Si)
1,E+04

A (c-Si)
3 2 1

G1 (a-Si)
1,E+03

Parallel Resistance
0
1,E+02 0,001 0,01 Solar Irradiance in terms of Isc 0,1 [Isc / Isc@STC] 1

Series Resistance
0 0,05 0,1 1/Isc [a.u.] 0,15 0,2

Figure 4: Rsc for the two thin film modules examined. G1 and G2 is a single junction, H a multi-junction module.

Figure 5: Example for the determination of the series resistance Rs from VIM-data. Indicated is the series resistance determined from the fit and the value of Roc under standard conditions.

thin film module G1 with a manufacturing failure. 1.6. Contact resistance The only terms having an effect on the slope Roc are the series resistance and the two diode terms. In the high illumination limit, we may neglect the effect of the second diode and we obtain for Roc: Roc = 1/Isc (nkT/e) + Rs (6) Figure 3 shows the determination of the series resistance from a plot Roc as a function of (1/Isc). We may extract Rs by a linear fit of equation 3 to these data as the intercept of the fitline with the Roc -axis. The series resistance provokes a saturation of the Roc(Isc)data towards high illumination levels (Rs is the high illumination limit of Roc). This leads to a drop of the fill factor and limits the module performance in this illumination regime. The Table shows the series resistances determined for the modules examined. For wafer based modules A, B, C, and D, the share of Rs on Roc under 1 sun conditions is in the range of 50%. This means that the negative effect on the fill factor has to be equally attributed to the quality of the first diode and the series resistance. The share is significantly lower for the single cell F and for the thin film modules. The share is also lower for module E which uses relatively small cells. 1.7. The resulting module performance Assuming a linear relation between the short circuit current Isc and the illumination G, the module efficiency can be expressed as: = Voc FF Isc, STC / GSTC, , (7) where Isc, STC is the short circuit current under one-sun illumination level (GSTC=1000W/m2). This relation eliminates the need for a reference detector once the short circuit current has been determined at one single illumination level. Furthermore, the efficiency is normalised to the value at standard conditions. In analogy to the Performance Ratio, which describes the relative performance of a complete PV system, we propose to use the term Module Ratio MR for the relative performance of the Module: MR = /STC = (P/G) / (PSTC,/ GSTC) , (8) where STC is the module efficiency under standard test conditions, P the module power under real operating conditions with the illumination G, and PSTC the module
1.2

power under standard test conditions. The module ratio data are shown for some samples in Figure 6. Note that the module ratio curves are used by the MotherPV method to compute the annual energy production: after adapting them to the module temperature on site, they are convoluted with the annual distribution function of the solar illumination on site. This directly gives the annual energy production of the specific module technology on the specific site [3]. According to the equations (7) and (8), the module performance (module ratio) is affected by the: carrier generation (not examined here) open circuit voltage achieved fill factor The effects of the equivalent circuit parameters on the open circuit voltage and the fill factor and the related technological issues have been discussed in the previous sections. A drop of the power towards high illumination levels are provoked by the series resistance and the first diode. A drop of power towards slightly lower illumination levels by the second diode term. By this way, a clear link between the cell technology, the equivalent circuit, and the module performance is created. 1.8. Module performance under lowest illumination For small scale applications, such as parking meters, modules are also expected to deliver power under very low illumination conditions. Figure 7 therefore shows the same Module Ratio data using a logarithmic scale for the illumination level. The best candidate for such applications is the amorphous silicon module G2, the limiting factor for the module performance under lowest illumination levels is the second diode. We speculate that the second diode presents shunt paths on the lateral side of each pin structure of the module, a shunt path which is provoked by the back contact metal. Conclusions and Outlook PV-modules of identical nominal power do not perform equal. This paper examined the irradiance dependence of module performance and shows up the technology factors limiting this performance. For this purpose, data from variable illumination measurements (VIM) of the I/V curve are analysed using a modified two diode model. This allowed for a clear separation of four technological key issues determining the
G2 (a-Si) H (a-Si)

H (a-Si)

G2 (a-Si)
1,2

1 MR 1 0.8 Module Ratio MR

Module Ratio

C (p-Si) A (c-Si) E (p-Si)

0,8

0.6

C (p-Si)
0,6

0.4

0,4 0.2 0,2 0 0 0.2 0.4 0.6 Solar Irradiance in terms of Isc 0.8 1 [Isc / Isc@STC] 1.2 0 0,001

E (p-Si)

A (c-Si)
0,01 0,1 Solar Irradiance in terms of Isc [Isc / Isc@STC] 1

Figure 6:.The module ratio of some sample modules. The cross indicates the module performance and illumination under standard conditions. The wide scatter of the amorphous silicon module G2 is related to the stabilisation of the module power.

Figure 7: Same data as in Figure 6, but the illumination level is presented on a logarithmic scale for performance assessment under lowest illumination conditions.

performance of the modules: the rectifying capacity, the second diode, the recombination in the i-layer for amorphous silicon modules, the series and parallel resistance. The VIM-method is a simple tool revealing rich information about PV modules. It is ideal for monitoring a manufacturing process or for certification procedures. References [1] Photon International, September 2007, page 20 [2] J. Merten, J.M. Asensi, C. Voz, A.V. Shah, R. Platz, and J. Andreu, Improved Equivalent Circuit and Analytical Model for Amorphous Silicon Solar Cells and Modules, IEEE transactions on electron devices, Vol. 45, No.2, February (1999) pp.423-429 [3] A. Gurin de Montgareuil, Description of MotherPV, the New Method developed for the Assessment of the Energy Production of Photovoltaic Modules, 22nd European Photovoltaic Solar Energy Conference, Milano 2007 [4] J.Merten, A.Muoz, C.Voz, J.M.Asensi, J.Andreu, Industrial quality control of Amorphous Silicon Modules by Variable Illumination Measurements (VIM), World Conference on Photovoltaic Solar Energy Conversion, Vienna, July 1998 [5] J. Merten, C. Voz, A. Muoz, J. M. Asensi, J. Andreu, The role of the buffer layer in the light of a new equivalent circuit for amorphous silicon solar cells, Solar Energy Materials & Solar Cells 57 (1999) pp. 153-165

APPLICATION OF THE MOTHERPV METHOD TO THE ACCURATE MONITORING OF THE GRID-CONNECTED PHOTOVOLTAIC SYSTEMS Antoine Gurin de Montgareuil *, Frdric Mezzasalma *, Jens Merten ** INES CEA/LITEN/DTS/Laboratory for Solar Systems (L2S) * Solar Outdoor Test Facility - CEA Cadarache bt. 356 - F-13108 St-Paul-lez-Durance, France Phone: (33) 442 25 44 12, Fax: (33) 442 25 73 65 - gdm@cea.fr ** BP 332, 50 avenue du Lac Lman, F-73377 Le Bourget du Lac, France ABSTRACT INES CEA has developed the MotherPV method (Meteorogical, Optical and Thermal Histories for the Energy Rating of PhotoVoltaics) allowing the accurate characterization of the performance of photovoltaic modules. This method has been extended in order to allow the monitoring of the performance of photovoltaic systems, resulting in the MotherPV SYST method. The MotherPV SYST method observes the behaviour of the DC part of the system and the inverters behaviour. For the DC part of the system, module knowledge may be used to determine if the system behaviour is correct. For the inverters part, the dynamic behaviour of the inverter has to be taken into account. The method allows the detection of out-of-range behaviour of the system, leading to an efficient monitoring. It is hoped that next improvements of the method will allow the observation of system ageing. Keywords: system performance, MotherPV SYST method, monitoring, module performance

1 INTRODUCTION INES/CEA has developed the MotherPV method (Meteorogical, Optical and Thermal Histories for the Energy Rating of PhotoVoltaics see [1]) allowing the characterization of the performance of PV modules. The MotherPV method relies on the observation that, for a given short circuit current and a given module temperature level, the maximum power of the module does not vary a lot (even if it depends weakly on the recent history of the module) and that it follows an almost Gaussian distribution function around its average value. Thus, by using a great quantity of measurements, it becomes possible to describe the behaviour of the module as a function of short circuit current and module temperature. Further more, if the relationship between the irradiance measured by an irradiance sensor and the modules short circuit current is well known, one can determine the performance of the module for every irradiance and module temperature. More precisely, the MotherPV method expresses the Irradiance@STC Module power . ratio as a Module power@STC Irradiance function of the irradiance level for a module temperature of 25 C. This ratio expresses the non proportionality of the module power as a function of irradiance at a given module temperature. The MotherPV method also calculates the temperature coefficient of this ratio, which is usually found to be a linear function of the module temperature for a given irradiance level. On the same way, the behaviour of grid connected photovoltaic systems depends, but not exclusively, on the received irradiance and on the temperature of the modules. The main other parameters are the behaviour of the maximum power point tracker of the inverter (especially when important and quick variations of light occur) and also the efficiency of the DC/AC conversion. In the present work, the MotherPV procedures have been applied to a grid connected PV system, leading to a new method called the MotherPV SYST method allowing the monitoring of grid connected PV systems. The first part of the document describes the monitored system and the data recording and handling procedures.

The second part of the document describes the modelling of the behaviour of the DC parameters of the system, depending on the association of the module array and of the maximum power point tracker of the inverter. The third part describes the modelling of the inverters performance and the fourth part the use of the model.

2 DATA RECORDING AND VALIDATION 2.1 The monitored system The system consists in 4 rows of 10 Photowatt P800 modules mounted in series and connected to a Fronius inverter (the inverter at the right in Fig 1).

Figure 1: Inverters of the system test facility The system is part of a 23 kW system allowing the simultaneous test of up to 10 inverters (see Fig. 2).

Figure 2: Modules at the Cadarache system test facility

2.2 The data recording system Data recording has been performed by a National Instrument data logger. Data have been recorded every five seconds during one year in order to be able to observe the transient effects of the voltage and current due to the irradiance variations and to the inverters behaviour. The recorded data have been irradiance (Kipp & Zonen CM11 and SPLITE see Fig. 3), module back temperature, inverter voltage and current (DC and AC sides), AC power, AC energy.

Irradiance as a function of time, clear sky day

1000 Irradiance (W/m) 800 600 400 200 0 00 03 06 09 12 15 Time 18 21 00

Figure 4: One day of irradiance, clear sky day

DC power as a function of time, clear sky day 2500 DC power (W) 2000 1500 1000 500 0

Figure 3: The CM11 and SPLITE irradiance sensors 2.3 Data storage Data have been stored in a Microsoft Access database. At Cadarache, data are alternatively stored in Microsoft SQL Server databases, which allow bigger databases (Access is limited to 2 gigabytes databases). 2.4 Solar parameters calculation Various solar parameters have been calculated: sun altitude and sun azimuth (algorithms used by the French Bureau des longitudes), solar time, air pressured air-mass and sun-module angle of incidence (the tilt angle of modules is 30 ). These parameters allow the limitation of the extracted datasets to the desired air mass and angle of incidence intervals. 2.5 Data filtering Physically out-of-range measurements have been marked and not deleted (for example, records with an irradiance level above 1500 W/m). Then after, during the analysis process, only correct records have been taken into account. 2.6 Data analysis To retrieve data from databases as Access or SQL Server, users perform queries to the database, using a specific language, called SQL. Three kind of data extraction are usually performed: Times series of data during a certain period, usually one day or part of day to observe specific situations like transient effects (see Fig. 4, 5 and 6) Data responding to defined conditions, mostly intervals: irradiance, temperature, angle of incidence Statistical data for given datasets: for example average values of module back temperature as a function of irradiance (see Fig. 7)

00

03

06

09

12 15 Time

18

21

00

Figure 5: One day of DC power, same day

DC voltage at sunrise 400 350 DC voltage (V) 300 250 200 150 100 50 6:40 6:50 7:00 Time 7:10 7:20

Figure 6: DC voltage during the sunrise, same day

Module temperature as a function of irradiance Module temperature ( C) 70 60 50 40 30 20 10 0 0,0 0,5 Number of Suns 1,0

Figure 7: Average module temperature as a function of the irradiance expressed in number of Suns, one year data

2.7 Distribution functions of quantities and outliers elimination In order to fully characterize the normal behaviour of DC and AC parts of the system, all transient effects, specific effects, rare situations have to be discarded. So, the distribution functions (density of probability) of meaningful ratios are plotted and reasonable situations are manually defined. For example, for a Gaussian distribution function, all the cases not too far from the average (less than three times the standard deviation) are considered as good, the others are considered as outliers. Figure 8 shows the distribution function of the ratio between the DC and AC power of the system (it is presented as a normalized density probability function). This ratio is one component of the inverter efficiency.

As noticeable are the ratios between: Irradiance measurement and DC current DC power and AC power (UAC . IAC) / PAC expressing the cos() The systematic discarding of outliers by marking the corresponding records allows a better confidence in the quality of the data used for the modelling, often performed with regressions. Once all the known sources of problems for statistical modelling (regressions) have been studied and discarded, the modelling itself may begin. Without this data filtering, almost no satisfying and meaningful statistical law can be found. At the opposite, rough data filtering may bring its own errors, as the distribution function of the remaining data may be no longer Gaussian. 2.8 Specific data filtering for systems For module characterisation, no further data filtering is required by the MotherPV method. Systems are more complicated, mainly because the inverters behaviour introduces more complexity: even with rather good irradiance conditions (very slight variations in sun irradiance), the maximum power point tracker of the inverter may have a specific behaviour and the DC power is not very reproducible. One can see for example that a bit before noon on the 5th of May, a very clear sky day (see Fig. 10), the behaviour of the inverter maximum power point tracker is very specific (see Fig. 11 and 12).
Irradiance as a function of time, 5 of May
th

Density probability of PAC / PDC, one year data

40 35 30 25 20 15 10 5 0 0

Density probabilty

Peak around 1

10 15 Ratio PAC/PDC

20

Figure 8: Density probability of the ratio PAC/PDC

Irradiance (W/m)

Sometimes, the density probability has to be focused on a specific part of the graph (bins are narrowed) as shown in Fig. 9. In this case, values below 0.70 and above 0.95 will be discarded and marked as outliers.

1000 800 600 400 200 0 00 03 06 09 12 15 Time 18 21 00

Density probability of PAC / PDC around 1

Ratio PAC/PDC

DC power (W)

40 35 30 25 20 15 10 5 0 0,65

Density probabilty

Peak at 0.926

Figure 10: Very good sky conditions on 5th of May

DC power as a function of time, 5th of May 2500

0,75

0,85

0,95

2000 1500 1000 500 0 00 03 06 09 12 15 Time 18 21 00

Figure 9: More accurate density probability of the ratio PAC/PDC around 0.9 For this study, many ratios have been investigated, and especially ratios which are known to be representative of some specific situations; For example, the ratio between the output of the pyranometer and the SPLITE gives indications on the presence of dew, snow, frost, soiling

Figure 11: The DC power of the system is not as smooth as the measured irradiance

Detailed DC power, 5 of May 1950 1940 DC power (W) 1930 1920 1910 1900 11:30

th

This work is still in progress. It will be presented in a future publication comparing different models.

Non proportionality of the module power @25 C Non proportionality coefficient multicrystalline silicon 1,0 0,9 amorphous silicon 0,8 0,7 0,0 0,5 Number of Suns 1,0

11:31

11:32 11:33 Time

11:34

11:35

Figure 12: The maximum power point tracker of the inverter follows a specific strategy We then had to add two steps in the data filtering process: We selected very clear sky days (see an example of these days with Fig. 4 and 10) Then, we selected the moments when the variations of voltage and current are not significantly higher than the variations of light, which usually means that the maximum power point tracker is at the right place This step was a key point in our work. Without this additional filtering, no satisfying laws could be really found.

Figure 13: Non proportionality coefficient of the module power as a function of irradiance for two different module technologies, amorphous and multi crystalline silicon The third reason is that the average non proportionality coefficient during a given period can be used directly for the calculation of the performance ratio of a whole system. This will be fully described in a future publication. In the MotherPV SYST method, the procedure to calculate the non proportionality coefficient of the power of the system at a module temperature of 25C, is the following: Data are carefully filtered (see paragraph 2) The non proportionality coefficient of the system is calculated for a given set of irradiance levels, typically from 0.025 Suns to 1.05 Suns, steps of 0.025 Suns. Because of the great number of data, only records with an irradiance of 0.01 Sun around the considered irradiance are used for the calculations (see Fig. 14) The temperature coefficient of the non proportionality coefficient is calculated for the same irradiance levels. If necessary, only one average temperature coefficient is used, for simplicity reasons (see Fig. 15) The average module temperature is calculated for the same irradiance levels (see Fig. 7 and 16) The non proportionality coefficient of the power of the system at a module temperature of 25 C is calculated from the three previous datasets (see Fig. 17) 3.2 Results The found laws are rather smooth, indicating a good data filtering. Nevertheless, as it can be seen if Fig. 6, the maximum power point trackers of inverters do not follow exactly the maximum power point of the module array and there are discrepancies between the non proportionality coefficient of the system (see Fig. 17) and the usual non proportionality coefficient of one module (see Fig. 13). There may be less smooth behaviours of the DC side of the system.

3 MODELLING OF THE DC BEHAVIOR OF THE SYSTEM 3.1 Method As described formerly in [1] and remembered in the introduction, the MoterPV method uses the ratio Irradiance@STC Module power . as a function of Module power@STC Irradiance the irradiance level for a module temperature of 25 C. This ratio can be described either as the relative efficiency of the module either as the non proportionality coefficient of the module. In a formal point of view, the use of the ratio is strictly equivalent to the use of the module power as a function of irradiance. It has been introduced for several reasons. The first one is that it allows a normalized presentation of the non proportionality of the module power coefficient as a function of irradiance, since all the non proportionality coefficients equal one by definition when the irradiance is one Sun. Thus it is very easy to compare the behaviour of different technologies (see Fig. 13). The second reason is that the non proportionality coefficient can be modelled as a function of the number of Suns. Several models are used by the different laboratories. The model developed and used at INES/CEA can be described in formula 1:

Coef =A + B ln(S ) + C (S 1) + D (T 25) (1)

Non proportionality coefficient

Furthermore, in a system, shadows, soiling and other parameters may bring discrepancies between modules current-voltage electrical characteristics (IV curves) and the behaviour of the module array is not proportional to the behaviour of one single module.

Non proportionality coefficient of the system power @ 25 C 1,0 0,9 0,8 0,7 0,0 0,5 Number of Suns 1,0

Non proportionality coefficient of the system power @ module operating temperature Non proportionality coefficient 1,0 0,9 0,8 0,7 0,0 0,5 1,0 Number of Suns

Figure 17: Non proportionality coefficient of the power of the system at a module temperature of 25 C

4 MODELLING OF THE INVERTER BEHAVIOR 4.1 Method This point will be less detailed than the modelling of the behaviour of the DC side of the system, because part is commonly very well studied. The ratio between the input and the output power is calculated expresses the conversion efficiency of the inverter. Here, this efficiency is expressed as a function of the DC power rather than a function of the AC power, as it is usually done. This is mandatory if one calculates the performance ratio of a system as a product of several successive conversion coefficients (soiling/shadow coefficient, module array non proportionality coefficient, inverters efficiency). The key point for the determination of the inverters efficiency is the use of extremely stable conditions: clear sky days, periods with a smooth behaviour of the inverter. The standard deviation of the inverters efficiency in real operating conditions has to be measured. 4.2 Results They can be seen in Fig. 18. The inverters efficiency seems to follow a standard behaviour. It would be very interesting to compare this inverters characterization to the characterization performed in stable and reproducible indoor conditions.
Inverter conversion efficiency 0,95 0,90 Inverter efficiency 0,85 0,80 0,75 0,70 0,65 0,60 0 500 1 000 1 500 Inverter DC power (W ) 2 000

Figure 14: Non proportionality coefficient of the power of the system at the operating module temperature

Non proportionality coefficient

Temperature coefficient of the non proportionality coefficient of the system power @ module operating 0,0E+00 temperature 0,0 0,5 1,0 -2,0E-03 -4,0E-03 -6,0E-03 -8,0E-03 -1,0E-02 Number of Suns

Figure 15: Temperature coefficient of the non proportionality coefficient of the power

Operating temperature for the considered days Module temperature ( C) 70 60 50 40 30 20 10 0 0,0 0,5 1,0 Number of Suns

Figure 16: Module operating temperature for the considered days

Figure 18: Inverter conversion efficiency as a function of DC power

5 APPLICATION OF THE MOTHERPV SYST METHOD TO THE MONITORING OF THE SYSTEMS BEHAVIOR 5.1 Method As the DC voltage is monitored, DC and AC side are considered separately. The previous characterization of the DC behaviour allows the calculation of the DC power for the measured irradiance and module temperature at any time. On the same way, the characterization of the inverters behaviour allows the calculation of the inverters output for a given DC input power. As the MotherPV SYST method deals with statistics, only averages may be considered. So, deviations from the average are normal. Only average behaviours are significant. Thus, one has to observe if the behaviour of the system for a given period is correct and more precisely if the distribution function of the recorded values during the last period is not too far from the predicted distribution function. So, if during the last period the number of outliers is too important and/or if the average behaviour of the system is too far from the rated behaviour, something might be wrong. 5.2 Results Preliminary results show that the method could be successively applied to the monitored data. Thanks to the MotherPV SYST method, it has been clearly seen (Fig. 19) that there has been a sudden change in the behaviour of the DC part of the system after six months of monitoring. This change had not be observed before the application of the method mainly because temperature variation brings so much variation in the system output that even important changes in the system behaviour can not be detected.
Changes in system DC power @ 25 C DC power @ 25 (W) C 2500 2000 1500 1000 After 26th of June 500 0 0,0 0,5 Number of Suns 1,0 Before 26th of June

6 CONCLUSION The MotherPV SYST method allows the monitoring of photovoltaic systems thanks to a preliminary characterisation of the system behaviour as a function of irradiance and module temperature. It considers a statistical point of view and uses average values and standard deviations. The next development of the method will be to reduce the time needed to fully characterize the system by using the standardized procedures of characterisation of modules and inverter and applying the results to the whole system. The method can be used to detect unexpected problems, as it has been seen in the present monitoring. It is expected that thanks to its accuracy the MotherPV SYST will be able to monitor the ageing of photovoltaic systems, after the characterization of the DC and AC parts of the system during of the first months of systems life.

REFERENCES

[1] A. Gurin de Montgareuil, Description of MotherPV, the new method developed at INES / CEA for the assessment of the energy production of photovoltaic modules, 22nd European Photovoltaic Solar Energy Conference, 2007, Milan

Figure 19: Sudden change in DC power normalized at a module temperature of 25 C after a certain date. The two lines represent averages of 6 month datasets The next steps of this work will be the application of the method to every day of the period, trying to detecting days with a strange behaviour of inverter (see for example the variation of inverters power at noon during a clear sky day Fig. 11 and 12). .

SHADOWING EFFECTS ON GRID CONNECTED PV SYSTEMS

Nicolas Chaintreuil, Franck Barruel, Xavier Le Pivert, Herv Buttin, Jens Merten INES CEA/LITEN/Laboratory for Solar Systems (L2S) BP 332 Savoie Technolac, 50 avenue du lac Lman, 73377 Le Bourget-du-lac, France Phone : +33(0)4 79 44 46 11 - E-mail : nicolas.chaintreuil@cea.fr ABSTRACT The aim of this paper is the study of the impact of partial shading on a grid connected photovoltaic (PV) plant. Two technologies of PV modules are examined: multi-crystalline silicon and amorphous silicon. Two types of studies are made: On one hand, shadowing losses are assessed instantaneously under stable solar irradiance conditions. On the other hand, these losses are summed up during a five weeks period. Different types of shadows are tested. This study shows better results for amorphous silicon systems face to shadows. Partial shading may lead to complete system failures when the inverter stops operating when the Maximum Power Point voltage gets out of range. The results of these two cases are compared. Keywords: grid connected, shading, system performance

INTRODUCTION

Today, development of grid connected PV systems due to good purchase price of energy requires having losses the lowest as possible. It is important to have a good ROI (Return On Investment). In France, there is a special prime for building integrated PV (BIPV) systems. For these systems, however, partial shading is difficult to be avoided. Importance of shadows on a grid connected PV system was known but they have not been quantified yet. Problem is that there are many types of partial shadows on a PV plant. This study analyzes six types which are the most representative of the realistic cases. Furthermore, the thermal impact of a shadow on the PV cells is poorly known. This information would permit to draw a vision of long term risks of degradation due to overheating. Tests are made on two technologies of PV modules: multi-crystalline silicon and amorphous silicon. It was based on measurements of two identical grid connected PV systems. Steady state tests are performed to compare the influence of different types of shadow on different electrical architectures of the PV array. Furthermore, a long period measurement of five weeks is performed to determine the energy losses suffered by the two grid connected installations of different module technologies under the shadow of a pole. This study has been realised within the framework of a project named SOLUTION PV which has the goal to research an optimisation of PV system architecture. 2 DESCRIPTION OF PV ARRAYS STUDIED

The modules can be connected in series or parallel (see Figure 3). Each array has one PV inverter SMA Sunny Boy SB1100 connected to the grid. These arrays are positioned at south.

Figure 1: Multi-crystalline array structure.

2.2 Amorphous PV arrays Two identical amorphous silicon PV arrays of 744Wc were used. They consist of six modules Uni-Solar ES124. Each cell of these modules has a surface of 0.085m. It represents a rectangle of 35.6cm (14in) per 23.9cm (9.4in). The modules are connected in series. Each array has one PV inverter SMA Sunny Boy SB1100 connected to the grid. These arrays are positioned at south but slightly shifted 6 to the east.

The PV arrays examined are located at the INES headquarter close to Chambry in the Rhne-Alpes region in France. 2.1 Multi-crystalline PV arrays Two identical Multi-crystalline PV arrays of 1.5kWc were used. They consist of nine modules Photowatt PW1650 each. The cells used in these modules measure 12.55cm (4.9in) per 12.5cm (4.9in). There surface is 0.016m.

STEADY STATE MEASUREMENTS

In these tests, we study the impact of different types of shadows on a PV plant (without inverter). We compare different electrical architectures and quantify the losses. For this part, we used I-V curves tracers which are connected directly to the complete PV arrays. The tracer type is PVPM6020C with a reference cell named Solar Radiation Sensor (calibrated). One array was exposed to partial shading whereas the other served as a reference. I-

V curves tracers permitted to determine maximum power point (MPP) of each array. The difference between the two arrays has been found to be less than 0.5%. This test was made at 829W/m and at a temperature of 45C in back of module. 3.1 Multi-crystalline PV arrays

3.1.1 Shadowing losses on array level In this part, six types of shadows were tested, which are shown in Figure 2.

The parallel architecture presents better results for shadows 1, 4, 5 and 6. For the shadowing 2 and 6, the difference between series and parallel may be up to a factor 2. There is no clear relation between the surface shaded and the corresponding losses because of the series connection of the cells. This explains the drastic impact of shadow 1, which only covers 1 single cell: the power loss is 25 times the loss of active surface in the case of the series architecture, and 11 in the case of the 3x3 architecture. Note that this loss is limited due to the bypass diodes - losses would be more important and could lead to the destruction of the modules due to the hotspot effect.

3.1.2

Shadowing losses on system level

Figure 2: The Six types of shadows examined. Two architectures were tested with these shadows:

Shadowing can cause additional losses on system level, as it has important effects on the Maximum Power Point of the array; two points are discussed here. Firstly, the Maximum Power Point may be difficult to be found by Maximum Power Point Tracker, as the shadowing causes the presence of two Maximum Power Points, which is shown in Figure 4.

Figure 3: Electrical architectures examined: series (left) and 3x3 (right). I-V curves tracers are placed on plus and minus terminals of the complete array. Figure 3 illustrates the deformation of the array I/V curves due to shadowing. The steady state power losses are synthesized in table I: Degree of module shading 0,15% Array Power loss Series 3x3

Figure 4: Example of two MPP curve for parallel architecture and shadow 2 (red: no shading, blue: shading). In this case, the Maximum Power Point Tracker can stop on the local maximum point at 92V and forget real maximum power point at 68V. Should the Maximum Power Point Tracker bet on the wrong Maximum Power Point, the loss due to shading would be 31.5% instead of 29.2%. Secondly, the voltage of the Maximum Power Point can vary significantly due to shading and going out of operating voltage range of inverter, which is shown in Figure 4, where the Maximum Power Point voltage is reduced by 38%.due to shadow 5. In this case, if operating voltage range of inverter starts at 180V, the inverter will stop operating and the shadowing losses are 100%.

Shadow 1 3,7% 1,7% (1 cell) Shadow 2 11,1% 12,6% 29,2% (72 cells) Shadow 3 13,9% 22,2% 36,8% (90 cells) Shadow 4 12,5% 18,3% 17% (81 cells) Shadow 5 11,1% 36,5% 30,5% (72 cells) Shadow 6 2,62% 16,7% 7% ( 17 cells) Table I: Steady state shadowing losses for the multicrystalline PV array. As we can see in this table I, the series architecture is better for shadows 2 and 3. However, in this case the orientation of the strings is very important and can significantly influence results.

(error of 5%) and losses on Shadow 2 are too small compared measure precision, so it explains difference.
700

600

500

P d c (W )

400

300

200

100

Figure 5: Maximum power point moving out of the operating range due to shadow 5 (series architecture, red: no shading, blue: shading).

0 0 50 100 Vdc (V) 150 200 250

Figure 7: Example of it can be an out of range voltage curve for shadow 5 (red: no shading, blue: shading). Figure 7 shows a reduction of the Maximum Power Point voltage under shadow to 123V, which represents a reduction in voltage of 22%, whereas the power is reduced by 26%. With the multi-crystalline modules, we did not observe such a close relation of voltage and power loss under shading. The reason is the presence of a by pass diode in each cell of this module: the cells shadowed are by-passed, which favors a reduction of the voltage and not of the module current under shading. These variations of voltage can stop running inverter for same reasons as said for multi-crystalline PV arrays.

3.2 Amorphous PV arrays Amorphous PV modules are interesting to study because of their conception. They have one diode per cell and present a good immunity to face shading. Tests are made on 6 types of shadows:

4 Figure 6: Amorphous PV array shadows. In this case, only series architecture was tested. As done with the multi-crystalline PV arrays, I-V tracers are placed on plus and minus terminals of the complete array. The precision of the power measurement is about 0.91%. Its the difference between the two I-V tracers and modules dispersion. We consider this difference in these measurements because some results values are near 1%.. Degree of Array module Power shading Loss Shadow 1 0.7% 1.3% Shadow 2 0.8% 0.3% Shadow 3 3.3% 4. 9% Shadow 4 4.3% 7.6% Shadow 5 14 % 25%. Shadow 6 1.4% 2% Table II: Steady state shadowing losses for the amorphous PV array. The results are shown in Table II. This table shows a better correlation between shading surface and losses of amorphous PV array than multi-crystalline PV array. This is due to the presence of one bypass diode per each cell. The factor between surface shaded and the loss in array power is around 1.6 in most cases. Only shadows 1 and 2 is different because Shadow 1 shading surface is an estimate

LONG TERM MEASUREMENTS

4.1 Multi-crystalline PV array These measures have been made in January 2008 at INES headquarter close to Chambry. Two identical PV arrays of nine PW1650 modules with a total array power of 1500W have been examined. Inclination of the array was 33, the array was oriented south. The pole with a diameter of 5 cm was used to create partial shading of the array. Identical inverters are used to connect these arrays to the grid. One array has been exposed to shadowing with a pole as shown in Figure 8, and one array served as a reference.

Figure 8: Shadowing by a pole of the Multi-crystalline array. The arrow indicates the daily coverage of the shadow on the array. In this case shadow is out of array after 2 P.M. (in January). We measured both DC power and AC power. To eliminate inverter losses we compare only DC power here. Measure difference between the two channels is

about 1% without shadow, it is the measure precision. Results indicate losses and DC production per day. It permits to verify if there is a link between losses and production (linked to sunshine).

Figure 11: Multi-crystalline one good day results in December 2007 (red: no shading, blue: shading, green: shading losses). In this case, day losses sum up to about 7%. It is easy to see when shadow of the pole goes out of PV array (at 13H on horizontal scale). 4.2 Amorphous PV array Figure 9 indicates that losses are linked to DC Production (linked to sunshine) due to dispersion of shadow when sunshine is low. Effectively, if sunshine is very important (without indirect sunshine), shadow was very net and contrasted. It results bypass of shading cells. If sunshine reduces or indirect sunshine becomes important, shadow will be less net and contrasted. Shading cells will not be automatically bypassed and losses of power (compared to a not shading array) will be lower (see Figure 10). This part of the study was made from 1st to 30th June 2008. Two identical PV arrays are used. Each array represents 744Wc with 6 ES-124 modules. Inclination of the array was 30, the array was oriented south. The pole with a diameter of 5 cm was used to create partial shading of the array (Figure 12). Two identical inverters are used to connect these arrays to the grid.

Figure 9: Daily shadowing losses (a) and DC-production of the reference array (b) for the Multi-crystalline array.

Figure 12: Shadowing by a pole of the amorphous array. The arrow indicates the daily coverage of the shadow on the array. In this case, the shadow covers the whole array during the whole day. Measure precision is about 1% and includes PV modules dispersion and Maximum Power Point Tracker differences. Inverters are identical. Results are presented as done in the previous section. The relation between production and losses is the same as multi-crystalline tests. Dispersion of shadow depends of intensity of sunshine and direct or indirect sunshine.

Figure 10: Multi-crystalline one cloudy day results (12th January 2008) (red: no shading, blue: shading, green: shading losses. On Figure 10 difference between the two channels is directly linked to power production. When production is strong, the difference increases. The DC losses on January 2008 sum up to 4.5% (Figure 9). This result is not negligible in view of the small surface covered (2% in the worst case). It is important to note that shadow is out off array near a half day. So, losses, if array is shading throughout the day, can be almost doubled.. For a good day, we observe the curve on Figure 11:

Figure 13: Amorphous long period results. In this case, shading losses sum up to 1% throughout test period. This results is very close to the measurement uncertainty. This means that no significant shading losses of the amorphous silicon array have been detected.

HOT SPOTS

It is clear that shaded cells do not generate current, but have to pass through the current produced by the rest of the string. The corresponding cell heating has been observed with thermal camera (see Figure 12):

Figure 14: Thermal behavior of multi-crystalline PV array under the shadow of a pole. Figure 14 shows thermal impact on cells shading by a pole. Difference of temperature between overheating cells (under shadow) and not shading cells is about 15C in this case. Air temperature is around -1C and sunshine around 200W/m (morning of a sunshine day). This results permits to conclude that a repeating shadow on a multi-crystalline PV array can generate degradations on heating cells.

CONCLUSION AND OUTLOOK

This study has helped to quantify the losses suffered by partial shading of PV power plants based on multicrystalline and amorphous silicon technology. Several mechanisms have been identified which reduce the performance of PV generators under shading: Reduction of current generation Blocking of current by shaded cells Creation of local maxima on the array I/V curve which may lead to operation of the array far away from the optimum operation voltage Reduction of the Maximum Power Point voltage until running out of the operation range of the inverter In this case production is stopped! Steady state test have shown that amorphous silicon arrays are less sensitive to shading than the multicrystalline arrays. This is due the presence of a bypass diode for each cell of the amorphous silicon module employed. Shadowing losses are higher, when all modules of the array are connected in series. Long term tests with a pole shadow on the array also show that amorphous PV systems are less affected by partial shadowing than multi-crystalline systems. Long term studies on different sites are needed in order to quantify the annual productivity loss under shading of different module technologies.

DESIGN OF A MANAGEMENT AND SIMULATIONS TOOL FOR SOLAR CAR PARK

Thang Vu, Franck Barruel, Jens Merten INES CEA/LITEN/Laboratory for Solar Systems (L2S) BP 332 Savoie Technolac, 50 avenue du lac Lman, F-73377 Le Bourget-du-lac, France Phone : +33(0)4 79 44 45 67 ; E-mail : franck.barruel@cea.fr

ABSTRACT The concept of Solar Mobility consists of making the maximum use of the synergy of the fluctuating energy flow of PV systems and the grid coupled storage inherent to electrical cars. This paper shows that this concept does make sense and that negative impacts on the electricity grid due to massive penetration of PV systems and massive demand form electrical vehicles can be balanced. It is the energy management system which has to assure this balance. This paper presents a simple tool to evaluate the grid impact of the energy management strategy, which has to be designed considering a global vision taking into account the state of the electricity grid, the state of charge of the vehicles, and the state of the PV production .For demonstration, a sample strategy is simulated as a real case on the INES microgrid. Keywords: solar energy, electric vehicle, Solar Mobility 1 INTRODUCTION peak hour and 32 minutes in peak hour. It is clear that the autonomy of most electric cars exceeds by far the distance required to go to work and back.

Energy became a great debate of our modern civilizations. The traditional sources of energy are neither inexhaustible nor ecologically beneficial, so it is appropriate to integrate, within our lifestyles, energies known as renewable. The photovoltaic systems use the energy best distributed in the world: that of the sun. In France, currently, thousands of achievements emphasized qualities of photovoltaic solar electricity: its reliability, its autonomy, its weak influence on the environment. So in the absolute we can even guarantee part of our energy needs via Photovoltaic for our dwelling, our work place, but the field of transport remains problematic. Recently, the electric car returns in the foreground. The success or not of this novel mode of transport will be strongly related on the new performances of storages elements and to the (re)definition of the real need of the user. The objective of this article is to show an interest (and feasibility) on convergence between solar energy and the electric vehicles. It is not a question to work on a concept of car with integrated PV panels but of a parking lot equipped with PV panels supplying the cars. For that, the first part is the assumptions with the existing statistics on transport and the PV. The second part will relate to the development of management tools of the EV batteries connected to a solar installation and the network. The limits of this tool will be given. 2 TOWARDS A CONVERGENCE VEHICLE AND PHOTOVOLTAIC. ELECTRIC

Figure 1: Median distance of home-work displacement of migrant employees alternating in 2004 [1] 2.2 Calculation of energy necessary to be transported From the preceding figures, we seek an estimate of electrical energy necessary to ensure the vehicle complete his daily route home-work. For that we fix the following assumptions: Average output of the thermal vehicles under real conditions: TV=0.15% Average output of the EV (total): EV = engine x electronic x battery = 0.65% Average ICV (interior calorific value) of Gasoline: ICVa = 10.04 kWh/l Average consumption of the vehicles in 2004: Ca= 7.16 L/100 km Then, via the following formula it is possible to go back to the necessary energy EC. EC = (Dm x Ca x ICVa x EV)/TV With Dm = median displacement distance (km) After calculations, for the route home-office (7.9 km), half of French would require 3 kWh per day to go to work and back with an electric vehicle.

2.1 Statistic According to INSEE [1] (French statistic national institute), by including the people who reside and work in the same commune (27% employee) for which the distance between home and office and the time to work are conventionally regarded as null, the average distance of residence-work is 25,9 km. For half of the employees, the distance is lower than 7.9 km. The average duration of the travel residence-work would be 26 minutes in off-

Obviously that does not cover all displacements possible with a conventional car, but it appears that for the way to work and back, the amount of energy required is far less than we could imagine. For recall, an existing standard electric vehicle has a battery capacity between 15 and 30 kWh, which is equivalent to an autonomy of 80 to 160 km. It is noticed that 3 kWh/day is a value very compatible with the daily production energy of a small sized solar installation. Indeed, if we now look at the chart of France radiation, we observe that the average yearly production is 1150kWh/kWp. So 1kWp by each EV could meet the EV energy requirement for homeoffice displacement. It represents 7m of PV module (with 14% of photovoltaic efficiency). 3 kWh/day = 1080kWh/year 1kWp 7m of PV This remains average calculations but they show that there does not need a great quantity of energy to go on its work place. In this context and with the quick and constant raise of the fuel price (and soon the electricity price) why not to consider a coupling between the electric vehicles and the solar installations? Although it remains to prove, the emitted CO2 amount will be certainly decreased. There remains the economic aspect but taking into account the economic situation previously quoted, there still a solution towards PV + EV. However, if this solution is applied at a private individual, that poses some disadvantages. The solar installations produce only in the course of the day, when the private individual left to work! If we think storing this energy in battery the day, and charging the electric vehicles the evening, the energetic output decreases enormously. Thus it is necessary to create an operating process where we can directly transfer solar energy towards the electric vehicles batteries. In our study we considered that the vehicles are reloaded the day on the work place. Indeed most of the time, the vehicle remains motionless and useless from morning at the evening. Moreover, according to the work place, surfaces available would allow the creation of solar car parks. 3 MANAGEMENT ALGORITHM CONCEPT

refill: Normal mode (or slow charge) of which the duration is several hours Quick charge mode in 1 hour for the emergencies. In the evening, when the employee returns in his home, it connects the EV to the house socket. The battery is not obliged to be immediately charged, it can provide part of its capacity to decrease the consumption of energy of the house at the rush hour for example (typically around 20:00). At 1:00 to 6:00 am, in off-peak hour, it can be charged in order to reach the level necessary to ensure the route home-office When charging, the model of storage behaves like the model of load. When discharging, it behaves the same manner but with a negative sign for the powers: P = PStockage , Q = QStockage
V , V P = PStockage . V Q = QStockage . V lim lim The SOC is determined by its remaining energy. In reality, we never discharge or charge the batteries at 100%, We put a limit for the SOC from 10 to 90%. We put also a coefficient (to be fixed by user) for the output efficiency of battery. This model is quite simple. It does not take into account the electrochemical aspect. For example, for the cycles of discharge, the evolution of its SOC can be regarded as linear with the constant power. On the other hand, in the cycles of recharge, we can impose the constant power only in the range < 80%. Over 80%, because of the parasitic phenomena in the battery, to have a good output, we must recharge in floating mode i.e. the tension is constant and that the current decreases with time. Then the evolution of the SOC is not linear any more.
2
2

Figure 3: SOC Evolution 3.2 Algorithm First of all, it is decided that the battery must reach a certain level of SOC at the end of day's work. Normally, the vehicles remain with the car park in the work hour (from 8:00 to 17:00) in 94% of the cases. I.e. the vehicles are motionless more than 8 hours. Thus if it is the case of an electric vehicle, the battery must have enough time to be recharged completely (the slow recharge lasts only 6 hours). Then the first priority must be the guarantee to reach the SOC required in end-of-day. Then, if many electric vehicles require at the same time a quick recharge, it can create a strong demand of instantaneous power. In this situation, the cables can be congested even not supported. Thus if we do not want to remake all the installation or to support a great voltage drop, we propose the following solution: to discharge the other batteries if they have enough energy or to suspend their recharge. Although the solution of transfer of battery energy from one to another does not give a good

3.1 Proposed scenario

Figure 2: Daily Evolution of SOC It is considered that: The solar panels are installed on the work place. Maximum 8 m2 of PV can be installed for each vehicle in car park (the surface of a parking space is 11m). The morning, starting from 8:00, the EV will be charged during the work hour. It is supposed then that there are two possible modes of

energetic efficiency, it does not remain less interesting. In the majority of the cases, the quick recharge never exceed 1 hour. The energy loss is not much. In comparison with the cost of reinstall, this loss is reasonable. Lastly, with the exhaustion of the energy sources, the price of electricity increases quickly. Currently, there are price for off-peak hour/peak hour, but in a near future, why not a variable price with time. The difference in tariff between peak hour and offpeak hour can vary from 2 to 3 times. Thus it is necessary to charge the batteries in the off-peak hours, and during the peak hours, to make them discharge or suspend the recharge. There are 2 interests. First, we reduce the cost of total electricity in the course of the day. Second, we can reduce the maximum consumption power. That makes possible to sign a weaker subscription of electricity. Here the summary of the principles by set of priorities: If there is enough time, the required SOC must be satisfied. If there is a quick recharge, it is priority. Thus either the non priority batteries recharge the priority battery, or they are disconnected to decrease the total required power. At peak hours, the batteries can discharge to limit the maximum power, or to stop charging. We also made some assumptions: The battery state of charge is perfectly known. Until now, there are many methods to estimate this famous SOC, but none guarantees 100% of exactitude. The average error always lies between 10 and 20%. The minimal SOC is to 10%, we never go lower of this value. The quick recharge is at the beginning, at 8:00 of the morning when we has just connected the batteries. This choice is made because it is the worst case. We want to observe the bad influences of a strong demand of power on the network. If we make the quick recharge later, other batteries have already time to be recharged; therefore they have easily an enough high level of SOC to output energy. The peak hour tariff is 3 times more expensive than in off-peak hour (30 c vs. 10 c). It has one peak hour period in the work hour. If we inject power to the network, the resale price is equal to the tariff of electricity during this time. Limit: If we dont have enough time, the required SOC is not guaranteed. 4 DEVELOPMENT OF MANAGEMENT TOOLS

With the current version of PSAT [2] (Matlab toolbox for electric power system), we can only make static simulations. I.e. the parameters of the components are fixed (power of load, production etc). Thus to observe the impact of daily load variation we must change manually the elements parameters step by step. That thus makes impossible the quasi-static analysis of a network. In order to mitigate this problem we developed new functions. The first step has been to update the PSAT to include hourly load and production profiles

Figure 4: Voltage profile in p.u for 9 houses connected to a distributed LV grid The second step has been to develop a man-machine interface which includes all objectives named before. In addition to the voltage profile and the SOC of the battery, the total cost per day is shown.

Figure 5: Man-machine interface 5 ALGORITHM TESTED FOR SOLAR CAR PARK CONCEPT The example study is a solar station with 6 electrical cars, each of 30kWh connected to the INES microgrid. Two of these are service cars with a booster charge option. A PV plant of 5kWp (about 6m/car) is simulated with meteorological data at Le Bourget du Lac.

On the assumption that the market is really born and that the private individuals and professionals adhere to the principle, the companies will have to place at the terminals disposal to refill for the personnel EV car. However to add terminals of refill for ten vehicles is not without consequence on the dimensioning of electrical installation. It is thus advisable to have a simulation tool with which we can test management strategies for the vehicles batteries refill by measuring the impact on the low tension network. 4.1 Software development

Figure 6: INES network 5.1 Quick recharge without management tool Now it is considered that there are 2 cars (5 and 6)

charging in booster mode (30kW each). The powers distribution of the batteries from 1 to 4 is the same (5kW each). For batteries 5 and 6, the quick recharge mode last only a half an hour. With this configuration, the total power request from the car park is approximately 80kW. This one exceeded the acceptable maximum power of the line. An important voltage drop is then inevitable.

Figure 10: Recharge powers distributions When batteries 5 and 6 are in quick recharge, the batteries from 1 to 4 discharge so that the maximum power does not exceed 60kW. After half an hour, when the quick recharge is finished, the non-priority batteries (1 to 4) return in normal recharge. If we fixe the peak hours from 11:00 to 13:00, batteries 1 to 4 discharge this times. After 13:00, the batteries recharge up to the required SOC. This figure described the variation of SOC of battery 2.

Figure 7: Voltage drop occur Figure 11: SOC Evolution of battery 2. This problem also appears when we want to increase the number of EV to the car park. For example if there are more than 14 vehicles in normal recharge, the required power will exceed the acceptable maximum power by the line. 5.1 Quick recharge with management tool The simple energy production by PV cannot cover all the recharge instantaneously. Under these conditions here the result of simulations with the strategy of management defined by our algorithm. The SOC variation in our example is very recommende because it does not exceed the high and low threshold of the operating range. On the other hand, it is obvious that the number of cycles increases compared to a normal use. 7 CONCLUSION AND FACILITIES AT INES

Figure 8: Voltage stay in norm The batteries recharge functions like the proposed algorithm. The red arrow (8-8h30) indicates the quick recharge, the green arrows (8h30-11h and 13h-17h) the normal recharge and the blue arrow (11h-13h) the discharge of the batteries during the peak hours.

Figure 9: Car park consumption power For better distinguishing operation from each battery, we look at the following figure:

We just showed that the concept electric vehicle for home-work travel is interesting. Energy necessary to ensure the travel is compatible with the daily production of PV panels. The connection of the EV to the network does not have only advantages. The quick recharge can involve voltage drops of lines. The idea charging the batteries by batteries during this time can be an answer, even between cars (case study in this paper) or with stationary batteries under the solar car park. However, it should not be forgotten that the lifespan of the batteries is not unlimited. It is inversely proportional with the depth of discharge. But we know that with an adequate management, we can decrease this effect. The management tools are far from being perfect. However we have a good base to imagine scenarios which would not be limited to individual but collective displacements. Overall, we sought to validate the broad outlines of a concept which still requires advanced technologies where the battery will remain critical. Li-ion technologies are promising on the increase in the cycle number and possible depths of discharge. To validate the solar mobility concept, a solar park will be installed and plugged at INES with 3kWp, 2 electric cars and some electric bicycles. This solar park is connected to INES micro-grid (PRISMES) and monitoring to control the power management system [4]. Finally beyond technological advances, there remains one about it, and not the least; will of the people to change their means of transport and their behavior. And then which cannot be sulky the pleasure of taking one day its pump assistant, full of sun, please! 6 REFERENCES

[1] B. Baccani, F. Smcurbe, G. Thomas, Les dplacements domicile-travail amplifis par la priurbanisation, Insee Premire n1129, Mars 2007. [2] http://www.power.uwaterloo.ca/~fmilano/psat.htm [3] F. Milano, PSAT Documentation, Mars 2007 [4] : F. Barruel, O.Wiss, H.Buttin, S.Grehant, J. Merten, PRISMES : The INES Micro-grid platform. 23rd european PV solar energy conference, Valence 2008.

PRISMES: THE INES MICROGRID PLATFORM

Franck Barruel, Herv Buttin, Olivier Wiss, Sbastien Grehant, Jens Merten INES CEA/LITEN Laboratory of Solar System (L2S) BP 332 - Savoie Technolac, 50 avenue du lac Lman F. 73375 Le Bourget du Lac - France Phone: +33 (0)4 79 44 45 67 E-mail : franck.barruel@cea.fr;

ABSTRACT Applied research on PV systems is in expansion but it must be linked to industrial development for rapid and visible results. In this context, a new microgrid facility PRISMES will be realised at INES. This paper presents objectives and test possibilities with it. This contribution is aimed at the presentation of the facilities themselves. 1. OBJECTIVES OF MICROGRID 2 DC grid of 10kW (250-750V) 2 DC grid of 20kW (250-750V) Each grid can work in micro-grid configuration or can be connected to the main transformer of 1,2MVA at INES for grid-injection. The components of the PRISMES platform are dispersed over several buildings on INES campus and can be connected to each AC or DC grid: On top of PUMA 2&3 buildings, PV solar plant of 60kWp, multicrystalline technology, divided in 5kWp string. The electrical architecture of each string is modular. Moreover all chassis have a manual one axis tracker system with four inclinations (0, 10, 35 and 45). In front of LYNX building, PV solar plant of 10kWp, multicrystalline technology, divided in 2kWp string. On top of LYNX 2 building, PV solar plant of 3 kWp amorphous silicon technology. 26 different inverters (3 of 700W, 2 of 1.1kW, 12 of 2kW, 8 of 5kW and 1 of 30kW) in technical room at Puma 2 In LYNX 2 several battery system test benches up to 50kW. The complete battery systems (Pb, Li-ion) including the power electronics for grid coupling are tested. A broad variety of real and simulated DC and AC loads. A SOLAR STATION for electric vehicles. Contains 3kWp PV and a connection point for two cars. Connection to four full-scale instrumented houses (under construction, INCAS project)

The new paradigm of decentralised production on the low voltage grid puts the stress on storage system connected to renewable energy production and then on micro-grid notion. It is necessary to the PV domain to study and analyse this micro-grid. Main objective of PRISMES is the creation on a universal test and development tool for photovoltaic systems, such as: Stand alone hybrid systems with minigrid Grid connected PV systems combined with a storage system The modular structure makes the PRISMES platform flexible enough to allow for the realisation of any type of experimental set-up, such as: Comparison of performances for different electrical architectures of PV plant (fig 1). Effect of storage system connected to the grid. Test of solar inverters under real conditions. Test of a real small village with PV or hybrid sources. Charge of electrical cars directly with PV plant.

fig 1: One string of 5kWp of the INES PV plant 2. FACILITY DESCRIPTION

To reach these objectives the PRISMES platform is divided in three main facilities. 2.1. Network test bench Based at the Headquarter of the National Institute of Solar Energy (INES) in Chambry, the PRISMES platform is a multi microgrid that will allow testing of several parallel experimentations thanks to its 14 available physical networks (fig 2 and fig 7): 1 AC grid of 100kW (400V) 1 AC grid of 30kW (400V) 8 AC grid of 5kW (230V)

100m

Transfo INES

fig 3 : Monitoring and data acquisition 2.2. System test bench

Solar parking

Houses: INCAS

fig 2: The INES Campus with PRISMES interconnection To make the modular structure, a plug and socket system with Marechal decontactor system has been used. 15 power distribution and interconnection boards are installed in every buildings as shown in fig 2.

A system test bench allows assessing performances comparison of complete PV systems of 1.5kWp (until 12 systems in the same time) using different modules technologies (fig 4). Identical inverter, cable section and wire lengths are used for each system. Aims are the assessment of system losses and comparison of the productivity. The PV modules are on top of the PUMA 2&3 buildings close to the 60kWc PV plant.
1.5kWc Techno 11
2kV/400V , 1.2MVA Transformer

DC control

AC control

fig 4: System test bench 2.3. Module test bench fig 2 : distribution and interconnection boards in Lynx 1 building In-situ monitoring of all energy flows is completed by a central control system. This monitoring has been fully conceived and computed by INES engineers. Data of each channel can be visualized on any PC located in INES office (fig 3) such as. AC current, voltage and power at fundamental or with harmonics DC current, voltage and power Temperature sensors on the PV plant Illumination sensor Other data can be taking into account (speed, torque, dynamic) function of the specific loads used for the experiments. The sampling rate of the acquisition is around one per second. This supervisor system takes also into account the security risk over the entire platform. A novel of PV modules test bench for the continuous outdoor scan of I/V curves will be set up on the same area as the 60kWp PV plant and the system test bench, on the roof of PUMA 2&3. The same bench already exists in front of LYNX 1 and on the solar outdoor measurement platform in Cadarache (South France). The I/V test bench has 24 channels for a continuous scanning of the I/V curves of the modules, which are connected by mean of relays to an electronic load. These data are stored together with the meteorological data in a database for post-treatment and analysis. Aim of these I/V test benches are: Monitoring of the characteristic parameters of the PV modules under natural irradiance. Precise data for novel models for PV modules of any technology. Determination of the nominal power and the energy production of the modules under test Comparison of the performances of different technologies of PV modules under different climatic conditions. In-situ monitoring of degradation processes.

1.5kWc Techno 12

1.5kWc Techno 1

1.5kWc Techno 2

1.5kWc Techno 3

fig 5 : PV modules of different technology 3. CONCLUSION Three engineers and technicians of CEA-INES have designed a tool for testing and development of any kind of PV systems and grid coupled storage systems. More than one and half year has been necessary. The official launching is set on September 2008 and first results are scheduled for autumn 2008.

fig 6: Global scheme of PRISMES

SYSTEM MODELLING AND ENERGY MANAGEMENT FOR GRID CONNECTED PV SYSTEMS WITH STORAGE

Yann Riffonneau*, Arnaud Delaille*, Franck Barruel*, Seddih Bacha INES CEA/LITEN/Laboratory for Solar System (L2S), BP 332, Savoie Technolac, 50 avenue du Lac Lman, F.73377 Le Bourget du Lac, France G2Elab, ENSIEG-INPG/UJF-CNRS-BP - 38402-Saint Martin DHres-Cedex-France Phone: +33 (0)4 79 44 46 00 E-mail: Yann.Riffonneau@cea.fr

ABSTRACT This paper presents the modelling and energy management of a grid connected PV system associated with storage. Within the economic, energetic and environmental context, objective of the system is to ensure loads supply at the least cost by optimising the use of solar power. Therefore, due to the complicated operating patterns, an energy management system which decides on energy flow for any moment is necessary. First, we present the system studied. Based on an AC bus typology, we consider all the energy flow possibilities. A model of the system is presented with an innovate algorithm for the estimation of batteries state of charge (SOC). Energy management has been done with a deterministic algorithm. At each step size, the algorithm finds the energy flow pattern which ensures loads supply at the least cost, taking into account availability of PV power, batteries SOC, loads demand and instantaneous electricity grid prices. loads as a power sink, batteries and grid are considered as power sources or sinks depending of the power direction. 1. Introduction Face up to the environmental, economical and political context, integration of renewable energy in electricity production is essential. Due to its many assets, photovoltaic solar power is one of the most promising solutions. However, over voltage in distribution lines network during important feeding time and the mismatch between the PV production and the loads consumption times impose the limits of integration of grid connected PV systems [1]. Adding a storage element to these systems would enable to solve these problems. In such a case, many applications are conceivable, depending of size and technology of the storage element. The high number of power flow direction has to be considering in function of the architecture of the system. To satisfy the objective function of the system, a power flow management algorithm has to be developed. Forecasting of variables in the system such as grid electricity price and solar source availability are necessary to ensure optimization along the period studying. This paper presents the hybrid system and its corresponding power flow possibilities. Modelling of the PV source and storage are displayed. Determinist algorithm has been used to develop the power flow management algorithm. This algorithm and its corresponding objective function are described for an individual economic environment. Simulations have been realized in a perfect prediction context. Results are analysed and discussed. 2. System description In this paper, we propose to integrate a storage element in a grid connected photovoltaic system. The storage technology used in this study is lead acid secondary batteries. Because the storage is associated with a commercial grid connected PV system, topology of the global system is composed of an AC bus. Figure 1 shows the scheme of the system with the possible power flow directions and their corresponding signs convention. Photovoltaic modules are considered as a power source,
PV inverter P_pv < 0

AC bus
P_bat > 0 P_loads > 0

P_grid > 0 Grid

P_bat < 0 PV generator Batteries inverter

P_grid < 0

Loads

Batteries

Figure 1 : Scheme of the system studied The size of the system corresponds to the values of Table 1 with: Ploads_max = Maximum loads power (kW) Ppv = Photovoltaic generator peak power (kWp) Pbat_max = Maximum batteries power (kW) Cbat = Total batteries capacity (Ah) Vbat = Total batteries voltage (V) Ploads_max Ppv Pbat_max Cbat Vbat 2.8 3.5 2 200 120 kW kW kW Ah V

Table 1 : System specifications The reference loads consumption profile and grid electricity prices are presented Figure 2. The PV purchase price is a constant equal to 0.3 /kWh. The on peak and off peak periods are based on the actual end user electricity contract, from the grid operator GEG in Grenoble, France [2]. Prices values have been evaluated considering to be expected in a future liberalised electricity market. The loads power demand profile has been created with information about french electric consumption from [3]. City of simulation is Chambry, in France. Hourly irradiation and temperature values for the year 2007 are collected from experimental measure at INES institute, Chambry.

3. System modelling 3.1. Photovoltaic generator The PV module current/voltage behaviour at the maximum power point (MPP) is described through linear equations (1), (2) and (3) [4]. The module temperature Tmod (C) is obtained from the nominal operating cell temperature NOCT (C), the ambient temperature Tamb (C), and the irradiance E (W/m). Taking into account the temperature effect, output current and voltage are obtained from the standard test conditions (STC). By this way, the only parameters needed are temperature coefficients CImpp (A/C) and CVmpp (V/C) , the NOCT, STC irradiance E0 equal to 1000 W/m, STC current Istc (A) and voltage Vstc (V), and cell temperature Tstc (C). These data are always given by the constructor. Tmod = Tamb + E/800 (NOCT 20) (1)

Qc(t) = (b1(SOC) b2(Qd) |Ibat(t)| dt)

(7)

The coefficient a1 is function of the discharge current in the battery and of the reference current I0 corresponding to the reference capacity Cn0. It permits to calculate the equivalent charge quantity extract of the battery during discharge process with the equivalent reference current I0 whatever the current rate Ibat. It corresponds to the Peukert equation where n is the Peukert coefficient. a1 = (Ibat(t) / I0) ^ (n 1) (8)

Impp = [Istc + CImpp (Tmod Tstc)] E/E0 Np (2) Vmpp= [Vstc+ CVmpp (Tmod Tstc) + 0.026 log (E/1000)] E/E0 Ns (3) Total current and voltage of the PV generator are obtained by multiply module current and voltage respectively by the number of modules in parallels Np and in series Ns. As the equations shows, we consider the PV generator working at its MPP. Behaviours of the MPP tracker (MPPT) and inverters are not modelled here. The mean efficiency of power electronic is taking into account with a ratio of 92% applied to the output power of the PV generator. 3.2. Lead acid battery This section presents only the state of charge. We do not model the voltage characteristic because it is not needed for energy management. We propose here an innovating estimation of SOC from [5] describes by equations (4) (5) (6) (7). It takes into account the variation of the quantity of charge in the process as a function of the current rates and temperature. Values needed to SOC estimation are the reference capacity and its corresponding current. Cnx (Ah) is the new nominal capacity after the discharge number x. This nominal capacity tapes off with ageing process after each discharge. Evolution of this quantity is described in the subsection about state of health (SOH); C(t) (Ah) is the available quantity of charge in the battery at time t; Q(t0) (Ah) is the initial quantity of charge in the battery at time t0; Qd(t) (Ah) is the quantity of charge which take part in the discharge process; Qc(t) (Ah) is the quantity of charge which take part in the charge process; ai and bi are balanced coefficient used to take into account the work conditions. SOC = Cn(t) / Cnx C(t) = Q (t0) Qd(t) + Qc(t) If Ibat < 0 Qd(t) = (a1(Ibat) a2(T) |Ibat(t)| dt) If Ibat 0 (4) (5)

The Peukert coefficient depends of the current rate and is generally in the range 1.2-1.6. Values of n is given by Table 2 as a function of the current Ibat(t). The coefficient a2 is used to take into account of the temperature effect. It is a function of the reference temperature Tref equal to 20C and the ambient temperature Tamb. A corrective factor of 0.75% is introduced from the French policy NF C58-510 on lead acid batteries for solar application. a2 = 1 / (1 + 0.0075 (Tamb(t) Tref)) (9)

The coefficient b1 corresponds to the charge efficiency of the battery. Discharge efficiency is assumed to be equal to one, and losses appear only during charge process. Efficiency depends of the SOC and the current rate. In our model, b1 is not a continuous value and is a function of SOC. Table 3 shows values of b1 used in our model for lead acid batteries. Table 2 : Values of the Peukert coefficient as a function of current
Ibat n I50 1.2 I20 1.3 > I20 1.5

Table 3 : Values of coefficient b1 function of SOC for lead acid batteries

SOC (%) 0 20 20 50 50 80 80 100 b1 0.9 1 0.7 0.6

The coefficient b2 is an innovate point in the SOC estimation. It permits to take into account the influence of the last discharge on the next charging process. b2 correspond to the rate between the equivalent quantity of charge extract from the battery during discharge process calculated with equivalent coefficients a1 and a2 (= Qd), and the same variable calculated without the equivalent coefficient. b2 = Qd / (|Ibat| dt) (11)

This coefficient allows to maintained feasibility of calculi during irregular cycling process such as in photovoltaic application. 4. Optimisation algorithm

(6)

The objective of the power system is to supply loads with the least cost while satisfying the constraints imposed. The problem will be solved with hourly step size. Equations (18) and (19) show the constraint imposed to batteries on SOC and batteries power Pbat. 0.2 < SOC < 0.9 Pbat 2 kW (18) (19)

DODref = Reference depth of discharge of batteries (%). It assumed to be equal to 80%. So the discharge energy cost is: Bat_dch_cost = 0.12 (23)

Referring to fig. 2, the algorithm has to find the batteries and grid power schedule which satisfy the objective function along the entire simulation period. We define below calculus of decision variable energy cost and cost functions used by the algorithm to find the optimal power flow solution. The problem is formulated based on the discretization of a 24 hours future prediction horizon with a step size of one hour [6]. All the conceivable power flow directions in the system are taking into account. Each one is optimal for a particular range value of the decision variable. At each discrete time instance, from real time and predicted values of solar power availability, loads consumption and batteries SOC, the algorithm calculates the decisions variables and applies the corresponding optimal power flow direction. 4.1. Energy cost Because we want to supply loads with the least cost, we need to estimate energy cost of each source. This subsection present the method to obtain energy cost of the photovoltaic generator and batteries. Photovoltaic energy cost is obtained from equation 20.

Charge energy cost Bat_ch_cost (/kWh) is the energy price of the source used to charge batteries divided by the batteries charge efficiency effbat estimated to 0.85. If batteries are charge with PV, this quantity is equal to PV_cost, if it is charge with the grid, it corresponds to energy price from the grid (24). PV_cost / effbat if charged from PV Bat_ch_cost = (24) grid_cost / effbat if charged from grid

The batteries energy cost Bat_cost (/KWh) is the sum of the charge and discharge energy cost (25). Bat_cost = Bat_dch_cost + Bat_ch_cost (25)

4.2. Cost functions The optimum power direction is choice in function of the cost function results. Two cost functions are defined (Present_cost_fct and Future_cost_fct) to choice the optimal state of batteries (charge (Pbat > 0), discharge (Pbat < 0) or idle (Pbat = 0)). Present_cost_fct is defined as the difference between the grid cost and Batteries cost (24). Present_cost_fct = grid_cost Bat_cost (26) It is calculated to estimate feasibility of batteries discharge operation (so when Ppv < Ploads). If the Present_cost_function is positive, grid_cost is higher than Bat_cost, so discharge of batteries to supply loads is the least cost operation. Otherwise, loads are supply by the grid and batteries are in idle state. In order to estimate feasibility of batteries charge operation, predictions are necessary. A prediction matrix has been developed containing the future loads power demand (f_Ploads (kW)), future grid electricity price (f_grid_cost (/kWh)), and future PV power availability (f_Ppv (kW)). The algorithm finds into the prediction matrix the periods when PV power is not enough to supply loads. It reads the future grid electricity price at this time and compares it to the battery energy cost calculated as a function of the charge source. The future cost function Future_cost_fct is defined as the difference between the future grid electricity price, gains from purchase PV power PV_gain (/kWh), and the batteries cost (25). If PV modules are the batteries charge source (so when Ppv > Ploads), we have to submit to the future electricity price the gain obtained by selling PV power to the grid instead of charging batteries. By this way, we can decide at any moment if it is cheaper to charge batteries or to feed the grid. If PV modules are not the charge source, it means PV power is equal to zero, so PV_gain is zero. Future_cost_fct = f_grid_cost PV_gain Bat_csot

PV _ cos t =

Inv pv Ypv Life pv

(20)

With : Invpv = Grid connected photovoltaic system investment cost (/kWp). From [7], this value is assumed to be equal to 6000 /kWp in France. Ypv = Annual photovoltaic yield (kWh/kWp). From experimental measures, this value is assumed to be equal to 1200 kWh/kWp in France. Lifepv = Life time of the photovoltaic generator (years). This value is estimated to 20 years. So the photovoltaic energy cost is: PV_cost = 0.25 /kWh (21)

Batteries energy cost is composed of two terms, the discharge energy cost and the charge energy cost. The discharge energy cost Bat_dch_cost (/kWh) is a constant obtained from equation (22).

Bat _ dch _ cos t =

Invbat N ref DODref

(22)

With: Invbat = Batteries investement cost (/kWh).Taking into account the future cost reduction achieved by the massive employment of grid couple storage system, this value is estimated to 100 /kWh for lead acid technology. Nref = Reference cycle life of batteries (number of cycle). It is assumed to be equal to 1000 for a reference depth of discharge of 80%.

(27)

If battery energy cost is lower than the future grid electricity price, batteries are charged from the corresponding source; otherwise, batteries are not used. By this way, batteries charge is anticipated to supply loads when PV power is not enough. 5. Simulations results Two hypotheses have been assumed during simulations. Firstly, the power flow from batteries to the grid is not considered in our algorithm.
4
Loads consumption profile PV production profile

Secondly, we assumed that predictions are perfect. This means that the true values of variables match exactly with the predictive values. This situation is not realistic but simulations with perfect predictions are necessary to study the algorithm behaviour in optimal conditions. Figure 3 and Figure 4 show the results of the optimisation algorithm for the 11 of July 2007.

0.5
Electricity prices

3 Power (kW)

0.4

0.3

0.2

10 12 14 Time (h)

16

18

20

22

0.1 24

Figure 2 : Loads consumption profile, electricty prices and PV output power corresponding to the input data of the simulation during the 11 of July 2007
4
Batteries energy Grid energy

Energy (kWh)

-1

-2

10 12 14 Time (h)

16

18

20

22

24

Figure 3 : Energy flow diagram of the batteries and grid energies corresponding to output data from the algorithm of the simulation during the 11 of Jult 2007.

Electricity prices (/kWh)

1 0.9 0.8 Batteries SOC 0.7 0.6 0.5 0.4

10 12 14 Time (h)

16

18

20

22

24

Figure 4 : Batteries SOC during simulation from the 10 to 11 of July 2007 As expected, batteries are charged with the grid during low price periods and are discharged to supply the loads during PV unavailability and high prices periods. The maximum battery power is respected and its SOC is kept in the range allowed. The total energy spent during simulation is obtained with equation (28). Ei (kW) is the energy quantity from the source i (PV, grid or batteries) and Ei_cost (/kWh) is the price of the energy from the source corresponding to the power Ei. Money spent = (Ei Ei_cost) (28)
[6]

[3]

OLIVIER SIDLER, Cabinet Sidler ENERTECH , Les usages de llectricit dans le btiment (electrical used in building), Septembre 2003, http://sidler.club.fr A.RICAUD, Photopiles solaires, polytechniques et universitaires romandes, 1997 Presses

[4] [5]

A.DELAILLE, Dveloppement de mthodes dvaluation de ltat de charge et de ltat de sant des batteries utilises dans les systmes photovoltaques, Thse de doctorat de luniversit Pierre et Marie Curie, 2006 P.L ZERAS & al, Model-based optimal control of a hybrid power generation system consisting of photovoltaic arrays and fuel cells, Journal of power sources, Vol 181, pp.327-338, 2008

The total money spent during on day is equal to 10 . This is less than the money spent using only the grid as power source, which is equal to 15 . The annual benefits from the system are equal to 1825 . In this economical context, the algorithm assures to supply loads at the least cost and reduce electricity cost. 6. Conclusion Energy storage associated to grid connected PV systems enables an optimal control and use of solar power. The algorithm developed in this paper assures loads supply at the least cost over the entire period studied. The algorithm determines for any moment the power flow in the system. It takes into account the real time and predictive values of loads demand, electricity grid prices, batteries SOC and solar power availability. A significant value of money is saved from the system use. Simplicity of the algorithm developped enables to implement it easily into a microprocesor in order to develop an experimental energy management system which will be tested at INES institute in Chambry. 7. References
[1] Y.UEDA & all, Detailed performance analyses results of grid-connected clustered PV systems in Japan first 200 systems results of demonstrative research on clustered PV systems, 20th EUPVSEC, 2005 Gaz Electricit de Grenoble (GEG), Tarif heures pleines/heures creuses pour le particulier, 2008

7 HESPUL association, Energy efficiency and renewable energies, grid connected photovoltaic systems, Energy information of Rhne, France, 2007, http://www.hespul.org

[2]

STATISTICAL APPROACH FOR THE GUARANTEE OF PHOTOVOLTAIC ENERGY OUTPUT A. Gurin de Montgareuil INES CEA/LITEN/DTS - Solar Outdoor Laboratory, Cadarache, 13108 St-Paul-lez-Durance, France, Tel. + 33 442 254 412, fax + 33 442 257 365, gdm@cea.fr ABSTRACT For economic reasons, the guarantee of long term energy output for grid connected photovoltaic modules becomes more important. The knowledge of the nominal peak power of the module does not allow good predictions, because the module efficiency in real conditions is different from the module efficiency in standard conditions. Furthermore, even with an excellent knowledge of the module behaviour, the prediction of energy output can not be more accurate than the meteorological predictions. Thus, it becomes necessary on one hand to perform module efficiency characterizations under a larger number of irradiance and temperature conditions and on the other hand to know as perfectly as possible the irradiation and temperature conditions is the candidate site. Two prediction methods developed at CEA are proposed : the Module Back Temperature method and the MotherPV method. Both rely on short measurement campaigns at a reference Solar Outdoor Laboratory and on the perfect knowledge of long term statistical behaviour of meteorological data at the site where the energy output has to be guaranteed. The two methods are applied and discussed on a real case, but for the same site. They show a very good accuracy. They have to be thoroughly tested with different sites for the measurement campaign and for the prediction. Keywords: PV module, performance, energy rating

INTRODUCTION

For a given site, long term behaviour of meteorological parameters such as irradiance, ambient temperature, sky temperature and wind speed and direction, are remarkably stable. The frequency distribution function of these parameters does not vary a lot from one year to another. As the electrical, optical and thermal behaviour of PV modules mirrors the behaviour of the meteorological parameters, the theoretical maximum energy output of grid connected modules (i.e. with perfect maximum power point tracker and inverter) is as constant and predictable as the long term behaviour of the meteorological parameters, as long as the ageing problems do not interfere. The prediction methods described in this article propose to gather accurate long-term statistics on the site where the modules will be placed, then to realize a short measurement campaign in a reference Solar Outdoor Laboratory, and finally to calculate the energy output for the given site. Both methods rely more or less explicitly on the fact that the average module efficiency for a given period is the energy barycentre of efficiencies for sub-periods. When considering periods with the same irradiance, the corresponding formula may be written as follows :
T = (G, (G )).
G

This temperature coefficient is the energy barycentre of the temperature coefficients of module efficiency [1]. The second method, called the MotherPV method (for Meteorological, Optical, Thermal Histories for Energy Rating in Photovoltaics) uses directly meteorological statistics and module performances expressed as functions of irradiance, and then applies relation (1). Both methods are compared in [2] with other prediction methods. This article describes briefly the meteorological statistics encountered at the solar outdoor laboratory, then the studied module behaviour during two years. At last, it describes in detail the two methods and their results.

2 THE METEOROLOGICAL STATISTICS AT THE SOLAR OUTDOOR LABORATORY ARE STABLE The irradiance, the ambient temperature and the module back temperature have been recorded during two years for one crystalline module in open field, south oriented and tilted at the latitude of the site (43 39). The irradiance has been measured with a pyranometer, the ambient temperature with a resistive sensor and the module back temperature with a thermocouple, placed at the centre of a cell near the centre of the module. The back of the thermocouple has not been highly insulated, just slightly protected against the wind gusts. Then, the irradiation and the average temperatures have been expressed as functions of irradiance for two consecutive years. Each year begins in March. The irradiation repartition is almost constant for the two years (see fig. 1). The ambient temperatures for the second year are a little higher than for the first year (see fig. 2). As shown in [3], the rise of module temperature above ambient temperature is, for a given site, almost proportional to the irradiance (see fig. 3). So the use of a constant heat transfer coefficient is pertinent, but it has to be adapted for each site and module set up. The use of the NOCT value is only pertinent if the wind and sky temperature average effects are similar to the wind and sky temperature encountered during the NOCT determination (wind speed = 1 m/s).

where h (G ) is the derivative function of the cumulative distribution function of irradiation expressed as a function of irradiance, (G ) is the average module back temperature at irradiance G, (G, (G )) is the average module efficiency at irradiance G, HT is the irradiation received during the time period T. The first method, called Module Back Temperature method, relies on the empirical law showing that the average module efficiency for a given set of periods is a linear function of the average module back temperature.

h(G ) dG HT

(1)

0,305
2,5E-03 Irradiation distribution 2,0E-03 1,5E-03 1,0E-03 5,0E-04 0,0E+00 0 500 1000 Irradiance (W/m) Year 1 Year 2

Cell ratio (A/W)

0,300 0,295 0,290 0,285 1/3/02

30/8/02

1/3/03 date

30/8/03

29/2/04

Figure 1: The distribution function of irradiation as a function of irradiance is almost constant

Ambient temperature (C) 30 25 20 15 10 0 200 400 600 Irradiance (W/m) 800 1000 Year 2 Year 1

Figure 4: The cell ratio expressed in A/W shows a strange behaviour during the winter 2002-2003 There is a very slight decrease of the ratio for the two spring periods, but there is a strange rise of the ratio during the winter of the first year. This has not been explicated yet nor been seen for the other modules in test. Nevertheless, the cell ratio is abnormally high during this period (almost 3%). The seasonal spectral effect is not evident. 3.2 The behaviour of the cell voltage at the maximum power point If a problem occurs with the serial resistance of the module, it can be seen on the value of the module voltage at the maximum power point. The observed voltage Vcell max is the module voltage divided by the number of cells. Thus, the average Vcell max is calculated for every day, translated to 25 C and to 1000W/m with calculated coefficients. The results are shown in fig. 5.
0,438 0,436 0,434 0,432 0,430 0,428 0,426 0,424 0,422 0,420 1/3/02 Voltage at max. power (V)

Figure 2: The average ambient temperature as a function of irradiance is slightly higher for the second year
30 25 Temperature rise (C) 20 15 10 5 0 -5 0 200 400 600 800 1000 Irradiance (W/m) Year 2 Year 1

30/8/02

1/3/03 date

30/8/03

29/2/04

Figure 3: The average rise of module back temperature above ambient is almost a linear function of irradiance

Figure 5: The voltage at the maximum power point decreases suddenly at the end of February 2003 The voltage drop between the two periods is around 1.4 %. It may be due to a connection problem, but this has not been investigated yet. 3.3 Consequences The measured behaviour of the module varies slightly during the two years. The standardized cell efficiency for high irradiance levels (see fig. 6) increases during the first winter (pyranometer problem ?), then lowers during the second year (connection problem ?). The accuracy of the prediction methods will be influenced by this unpredictable module behaviour. The average module efficiency for all irradiances levels decreases by 1.7 % during the second year.

3 THE MODULE BEHAVIOUR DURING THE TWO YEARS 3.1 The short circuit current behaviour In order to detect ageing effects or spectral answer effects, the ratio
Isc has been measured Gpoa . Cell Surface

for high level irradiances (> 800 W/m), averaged for every day and then translated to 25C, using a calculated temperature coefficient. The results are shown in fig. 4.

12,0% 11,9% 11,8% Cell efficiency 11,7% 11,6% 11,5% 11,4% 11,3% 11,2% 1/3/02 30/8/02 1/3/03 date 30/8/03 29/2/04

As explained below, it is also possible to predict the yearly energy output of a PV module with a one month campaign of measurements. Thus, daily statistics from each of the 24 different months of the two years measurement campaign have been used for the prediction of the energy output for one year. In order to avoid the bad statistical influence of the cloudy days, the days with irradiance below 2 kWh/m have been discarded from the statistics, because during those days, the distribution function of irradiance is too different from the annual distribution. The fig. 8 shows the results for the month of June 2002.
100% 80% Performance ratio

Figure 6: The standardized cell efficiency at high irradiance levels varies during the two years

THE MODULE BACK TEMPERATURE METHOD

60% 40% 20% 0% Yearly perf ormance ratio

4.1 Description For a given site and for a module with no strong ageing effect nor seasonal spectral effect, the monthly module efficiency is a linear function of the average module back temperature (see fig.7). The linear law is verified with a very good standard deviation (2 from 1 to 4 % according to the module technology). The yearly performance ratio follows the linear law with a very good accuracy.
100% 80% Performance ratio 60% 40% 20% 0% 15 20 25 30 35 Ave rage module back te mpe rature (C) 40 Yearly perf ormance ratio

25 30 35 40 45 50 Ave rage module back te mpe rature i n June (C )

Figure 8: Daily performance ratio at SOL site as a function of average module back temperature June 2002. The June month has been an excellent period for the prediction, but all the months have not been as good as the June month, as it is shown in table I. Table I: Error on prediction of energy output for year 1 based on one month measurements for 24 different months
Month Error Month Error 3 4 5 6 7 8 1,8% -0,3% -0,1% 0,5% -1,8% 1,5% 9 2,9% 10 4,6% 11 12 5,6% 5,8% 13 14 5,5% 1,3%

Figure 7: Monthly performance ratio at SOL site as a function of average module back temperature The module efficiency may be calculated as a function of the module back temperature for shorter periods (days), but the law is not well verified for the days with a very different distribution function of irradiation (cloudy days). Thus, for periods or sites with very different distribution functions of irradiation the law will not be verified accurately and the energy prediction can not be performed easily. On the other hand, for sites with a similar distribution function of irradiation, the law will probably be verified. The only problem will be then the prediction of the yearly average of the module back temperature. The module efficiency will be calculated by applying the linear law to the new point. The more accurate the temperature prediction, the smallest the error on the energy yield prediction. The temperature prediction may be done by measurements for yearly periods, or by using three meteorological data [1] : ambient temperature, average irradiance and average wind speed.

15 16 Month Error -0,4% -1,3% Month Error 21 0,3% 22 1,1%

17 18 19 20 0,2% 1,6% -0,7% 0,3% 23 24 2,5% 2,6% 25 26 3,8% 0,9%

The error on prediction for the energy yield of the first year lies between -1.8 and +5.8 %. As expected, the prediction based on the months of the first winter is overestimated, because these months present high module efficiency. Except these 4 months, the standard deviation 2 for the prediction is 3.3 %. The error on prediction for the energy yield of the second year lies between -0.4 and +7.4 %, which is coherent with the observations. 4.2 Discussion The module back temperature method is very easy to perform. It is powerful for long term energy prediction in the site where the characterization campaigns are performed. It probably may be used for sites with similar climates, knowing the yearly average module

temperature. The method is certainly not accurate for sites with too different climates, but it has to be verified.
Error on prediction

6% 4% 2% 0% -2% -4%

Prediction based on the months of year 1 For Year 2

THE MOTHERPV METHOD

5.1 Description This method consists of 3 steps. During the first step, the irradiation and the average module back temperature for the candidate site are expressed as functions of irradiance, either by direct measurements or by prediction methods. The second step consists in determining by a short measurement campaign the module efficiency as a function of irradiance and module back temperature. The final step uses the data provided by the first two steps. For each level of irradiance the new efficiency for the new average module back temperature is calculated by a temperature translation of the efficiency measured during the step 2. Knowing the new efficiency and the irradiation received, the energy production is calculated for each irradiance and then summed up. 5.2 Determining the meteorological statistics The irradiation and the average module back temperature depend on one hand on the climate of the site and on the other hand on the module set up (tilt angle, orientation). The easiest way to determine the meteorological statistics is to use the data of monitored sites. If the distribution function of irradiation on a tilted plane is not known, it seems that the available meteorological data translation methods are not powerful enough to predict it : the methods known by the author just allow to predict the irradiation received on a tilted plane, knowing the irradiation received on an horizontal plane, but no more. 5.3 Determining the module efficiency by a short campaign of measurements During the campaign, it is better to reach the highest possible level of irradiance. At SOL facility, the irradiance level reaches sometimes 1000 W/m during the winter solstice and usually 1080 W/m during the rest of the year. It is also interesting to get different ambient temperature levels during this short period. At SOL facility, several months during spring and autumn show alternate periods of cold winds and mild weather. IV curves are systematically recorded and cell efficiency is calculated for all irradiance and module back temperature levels. It is verified that, for a given irradiance, module efficiency is a linear function of module back temperature. The temperature coefficients are calculated for each level of irradiance. 5.4 Results In fig. 9, the error of the prediction is plotted as a function of the month used for the prediction. As expected, the months of the first winter overestimate the energy yield for the first year, and much more for the second year. The average accuracy is good. Except the 4 months of the first winter, the standard deviation 2 for the prediction is 2.3 %.

For Year 1

ar ap r m ay ju n

ju l au g se p oc t no v de c ja n

Figure 9: Error on the prediction of the energy yield of the first year (black) and year 2 (grey), with the data of the months of the first year. As expected, the energy yield for the second year is overestimated.

CONCLUSION

In this article, two prediction methods developed at CEA have been exposed. Both methods rely on a short measurement campaign at a reference Solar Outdoor Laboratory and on a good knowledge of the statistical meteorological conditions at the candidate site. The first method, the Module Back Temperature method, requires less knowledge of meteorological statistics, but its accuracy for sites with a very different climate is not guaranteed. The second method, the MOTHERPV method, is still under development. It seems that it brings very accurate results as soon as the annual meteorological data for the candidate sites are well known, which may be easily done by monitoring existing systems. This will have to be verified in a future work. Thus, the MOTHERPV method could be used for many sites. The module data would be published by the Solar Outdoor Laboratory performing the measurement campaign and the results would be used by any previously monitored site. In the future, both methods will have to be completed by taking into account the problem of module ageing and the spectral and thermal seasonal effects, especially for thin film modules.

REFERENCES [1] A. Gurin de Montgareuil, An Empirical Synthetic Law between the Module Energy Output and the Meteorological Data, Proceedings of the 3rd WCPEC (2003), Osaka, Japan. [2] Williams & al, Evaluating the State of the Art of Photovoltaic Performance Modelling in Europe, Proceedings of the 20th PVSEC (2005) , Barcelona, Spain. [3] E. Amy de la Bretque, Statistical Properties of PV Modules Thermal Behaviour, Proceedings of the 20th PVSEC (2005), Barcelona, Spain.

fe b

CHARACTERIZATION OF ADVANCED NICKEL-CADMIUM BATTERIES DEVELOPED FOR PHOTOVOLTAIC APPLICATIONS

Arnaud Delaillea, Florence Mattera a, ke Nystrm b, Charles Streb b INES CEA/LITEN/Laboratory for Solar Systems (L2S) BP 332 Savoie Technolac, 50 avenue du lac Lman, F-73377 Le Bourget-du-lac, France Phone : +33(0)4 79 44 46 08 ; E-Mail: arnaud.delaille@cea.fr b SAFT AB Box 709 SE-572 28 Oskarshamn SWEDEN Phone: +46 (0)49 16 80 00 Fax : +46 (0)49 16 81 80 E-Mail: Ake.NYSTROM@saftbatteries.com

ABSTRACT This paper presents a complete characterization of new nickel-cadmium (Ni-Cd) batteries developed for photovoltaic (PV) applications. This study has been performed by INES, the French National Solar Energy Institute, from the end of 2005 to the beginning of 2008. Eight Ni-Cd batteries have been studied according to different standards proper to PV applications, namely: the IEC 62093 standard, which consists in the determination of the faradic and energy efficiencies of batteries for state-of-charge (SOC) comprised between 0% and 50%; the IEC 61427 standard, which consists in the determination of batteries performance through an accelerated ageing test at 40C, and whose the cycling profile represents both daily micro-cycles and season macro-cycles specific to PV applications; the IEC 62124 standard, which consists among other things in the determination of batteries performance in a stand-alone PV system under real irradiation conditions. This paper is divided in four sections. The first one is dedicated to capacity tests performed at different discharge current rates, followed by a discussion about the influence of the current on the capacity delivered respectively by NiCd and Lead-Acid Batteries (LABs). Then, the three other sections deal with the different standards abovementioned. After a brief description of each test, the results obtained on Ni-Cd batteries are presented and also compared to typical results obtained on LABs in the same test conditions. Keywords: batteries, characterization, performance.

INTRODUCTION A new type of nickel-cadmium (Ni-Cd) batteries has been developed by SAFT for photovoltaic (PV) applications. This paper addresses a complete characterization of these new batteries, based on the technical know-how about storage systems of the Laboratory of Solar System (L2S), which is a part of INES, the French National Solar Energy Institute. This characterization has been done from the end of 2005 to the beginning of 2008, including the following tests: determination of batteries capacity at different discharge current rates; IEC 62093 standard, which consists in the determination of the faradic and energy efficiencies of batteries for State-Of-Charge (SOC) comprised between 0% and 50%; IEC 61427 standard, which consists in the determination of batteries performance through an accelerated ageing test at 40C, and whose the cycling profile represents both daily micro-cycles and season macro-cycles specific to PV applications; IEC 62124 standard, which consists among other things in the determination of batteries performance in a stand-alone PV system under real irradiation conditions. This paper is divided in four sections, corresponding to the different tests above mentioned. After a brief description of each test, the results obtained on Ni-Cd batteries are presented, and also compared to typical results obtained on Lead-Acid Batteries (LABs) in the same test conditions.

1 CAPACITY COEFFICIENTS 1.1 Experimental

TESTS

AND

PEUKERTS

The results presented in this section have been obtained on four Ni-Cd batteries, assembled from eighteen Ni-Cd elements SAFT SUN+140, according to the following configuration: Bat_1: six elements in series (Unominal = 7.2 V); Bat_2: six elements in series (Unominal = 7.2 V); Bat_3: three elements in series (Unominal = 3.6 V); Bat_4: three elements in series (Unominal = 3.6 V). The nominal capacities of the elements (and so of the tested batteries) given by the manufacturer are C5 = 128 Ah and C120 = 140 Ah. Several consecutive capacity tests have been performed at 25C (use of a thermostatic bath), down to 1 V/element, at different discharge current rates: C5/5 (I = 25.6 A); C5/10 (I = 12.8 A); C120/120 (I = 1.2 A). The full charge performed before each capacity test was realised at a constant current fixed to C5/10 (I = 12.8 A), during fifteen hours without any floating voltage. Besides, a relax period of one hour was observed between each charge and discharge steps. 1.2 Results Results of capacity tests are summarised in the Table below:

2,50 y = 1,0347x + 2,2409 2 R =1 y = 1,0331x + 2,2315 2 R =1 log t (h) 1,00

2,00

Table 1: Initial capacity measurements (in Ah) at 25C Bat_1

Current rate

1,50

Bat_2 145.4 163.5 144.2 151.1 147.1 157.1 152.7 170.0 149.9 153.7 153.3 153.7

Bat_3 147.8 168.4 150.9 153.3 152.5 152.9

Bat_4 147.4 166.6 150.0 152.7 152.3 152.9

Discharged capacity (Ah)

C5/5 C120/120 C5/5 C5/10 C5/5 C5/10 C5/5 C5/120 C5/5 C5/5 C5/5 C5/5

148.5 167.3 147.1 154.8 150.7 157.8 156.9 174.1 154.7 158.8 158.3 158.6

0,50

Bat 1 (3.6 V) Bat 2 (3.6 V)

0,00 -1,60

-1,40

-1,20

-1,00

-0,80

-0,60 -0,40 - log I (A)

-0,20

0,00

0,20

0,40

Figure 1: Determination of the Peukerts coefficient for two SUN+140 batteries Some Peukerts coefficients have been obtained from two series of capacity measurements (see Table 1), and noted n1 et n2 in the Table below:

We can observe a small influence of the electric history (that means of the previous cycles) on the measured capacity, which is a typical behavior of every electrochemical storage systems. Here is the average value of the C5/5 capacity for each battery, considering the four last cycles (stable values): C5/5 (Bat_1): 157.6 Ah +/- 2.0 Ah C5/5 (Bat_2): 152.7 Ah +/- 1.9 Ah C5/5 (Bat_3): 152.4 Ah +/- 1.0 Ah C5/5 (Bat_4): 151.9 Ah +/- 1.3 Ah We can thus underline here that SUN+140 are oversized compared to the nominal capacity (near 20% higher). From the results of the Table 1, it is then possible to determine the Peukerts coefficient n from the relation (1) below, used to estimate the battery capacity for any fixed current regime:

Table 2: Peukerts coefficients calculated from capacity tests performed on four SUN+140 batteries Bat_1 n1 n2 1.038 1.035 Bat_2 1.038 1.033 Bat_3 1.034 Bat_4 1.033

Peukert coefficient

The Peukerts coefficients obtained are very reproducible and close to 1.03. It can be mentioned here that the Peukerts coefficient obtained on LABs is superior and usually comprised between 1.15 and 1.20. Consequently, the current has a lower effect on the capacity of SUN+140 Ni-Cd batteries than on LABs; for a similar capacity at a low discharge current rate (e.g. at C120/120 as in the Figure below), a SUN+140 Ni-Cd battery will have a higher capacity than a LAB at a high discharge rate, as illustrated by the two graphs below:
150 140 130 120 110 100 90 80 70 0 20 40 60 80 100 120

Cx =

K Ix
n 1

I x .t x = K
n

a)

(1)
Available capacity (Ah)

with: Cx [Ah]: capacity at a discharge current I x ;

I x [A]: discharge current;

tx

[h]: time of discharge;

K and n are some constants, n corresponding to the Peukerts coefficient.

Hence, the Peukerts coefficient can be deduced from the slope of the curve log(t ) = f ( log( I )) , as shown in the Figure below for the batteries Bat_1 and Bat_2:

Ni-Cd, n = 1,03 Ni-Cd, n = 1,04 LAB, n = 1,15 LAB, n = 1,20

Approximately discharge time (h)

b)
100

faradic efficiency Bat_1

Ah_ch Ah_dch Eff.(%)

Available capacity (% of C120/120)

90

Bat_2
Ah_ch DCH Eff.(%)

80

70

60

Ni-Cd, n = 1,03 Ni-Cd, n = 1,04 LAB, n = 1,15 LAB, n = 1,20

50 0 20 40 60 80 100 120

Approximately discharge time (h)

Figure 2: Available capacity (expressed in Ah in Fig. a and in % of C120/120 in Fig. b) versus the time of discharge (equivalent to the discharge current rate), for different Peukerts coefficients n

Cycle 1 Cycle 2 Cycle 3 Cycle 4 Cycle 5 Cycle 6 Cycle 7 Cycle 8 Cycle 9 Cycle 10 Cycle 11

80.9 80.9 80.9 80.9 80.9 80.9 80.9 80.9 80.9 80.9 80.9

79.6 80.6 80.7 80.7 80.7 80.7 80.5 80.7 80.7 80.3 80.5

98.4 99.6 99.8 99.8 99.8 99.8 99.5 99.8 99.8 99.3 99.5

78.6 78.6 78.6 78.6 78.6 78.6 78.6 78.6 78.6 78.6 78.6

77.4 78.4 78.3 78.4 78.5 78.3 78.3 78.5 78.2 78.2 78.3

98.5 99.7 99.6 99.7 99.9 99.6 99.6 99.9 99.5 99.5 99.6

Table 4: Charged and discharged Wh and corresponding energy efficiency Bat_1

Ah_ch Ah_dch Eff.(%)

This difference between Ni-Cd and LABs can result in a big difference in term of autonomy, as illustrated in the section 4 of this paper through the IEC 61424 standard.

Bat_2
Ah_ch DCH Eff.(%)

2 FARADIC AND ENERGY EFFICIENCIES ACCORDING TO THE IEC 62093 STANDARD 2.1 Experimental The faradic and energy efficiencies of SUN+140 Ni-Cd batteries have been determined on the two batteries Bat_1 and Bat_2 (see section 1.1) according to the IEC 62093 photovoltaic standard [1], whose procedure is illustrated in the Figure below. According to this procedure, the faradic and energy efficiencies are measured for SOC comprised between 0% and 50%. In this range of SOC, the efficiencies are maximum for Ni-Cd batteries (as for LABs) since parasite reactions due to gas evolution only occur significantly from higher SOC.

Cycle 1 Cycle 2 Cycle 3 Cycle 4 Cycle 5 Cycle 6 Cycle 7 Cycle 8 Cycle 9 Cycle 10 Cycle 11

337.0 334.9 335.8 335.9 334.7 332.5 335.4 338.9 338.9 336.1 338.6

292.5 292.4 291.5 290.9 293.6 292.3 291.9 289.9 289.9 290.4 292.8

86.8 87.3 86.8 86.6 87.7 87.9 87.0 85.5 85.5 86.4 86.5

325.4 324.4 327.1 325.7 325.4 328.3 327.5 328.7 325.7 326.8 325.8

282.3 284.5 282.2 285.0 284.7 282.0 283.6 282.8 284.3 283.1 281.5

86.7 87.7 86.3 87.5 87.5 85.9 86.6 86.0 87.3 86.6 86.4

Results obtained on the two SUN+140 batteries tested are very reproducible from one cycle to another one, and from one battery to the other one. The faradic efficiency is close to 100%, as a result of the low part of parasite reactions in this range of SOC. The energy efficiency is close to 87%, as a result of overvoltages occurring during charge and discharge processes. The Table below presents the average faradic and energy efficiencies measured by the L2S on a great number of LABs, divided here into three groups: flooded flat (positive) plates, flooded tubular (positive) plates, and valve-regulated lead-acid (VRLA) batteries. All the measurements have been done in the same conditions, that is to say according to the IEC 62093 standard, between 0% and 50% of SOC. Table 5: Average faradic (Ah) efficiency and energy (Wh) efficiency measured on different types of LABs

2 100 % 1 : Charge at C5/5 for 15h 2 : 1h at rest 3 : Discharge at C5/10 down to 1V/cell 4 : Charge at C5/5 up to 50% of SoC according to the measured capacity 5 : Discharge at C5/5 down to 1V/cell

State-of-Charge (SoC)

Repeat at least 5 times up to stabilisation 1 50 % 3

00 % 0

Time

Figure 3: Cycling procedure according to the IEC 62093 standard 2.2 Results Results obtained for eleven consecutive cycles according to the IEC 62093 performed on Bat_1 and Bat_2 are summarized in the two Tables below: Table 3: Charged and discharged Ah and corresponding

Ah_efficiency
(%)

Wh_efficiency
(%)

Flooded flat plates 95 +/- 2 Flooded tubular plates 95 +/- 3 VRLA 97 +/- 2

88 +/- 3 86 +/- 2 90 +/- 4

We can conclude from the two last Tables that SUN+140 Ni-Cd batteries present quite similar faradic and energy efficiencies between SOC = 0% and SOC = 50% than

LABs.

140

120

SUN+140 Ni-Cd batteries

a)

Capacity (% Cn)

3 AGEING PERFORMANCES ACCORDING TO THE IEC 61427 STANDARD 3.1 Experimental An accelerated ageing test have been performed on the batteries Bat_1 and Bat_3 (see section 1.1), according to the IEC 61427 standard [2]. For this test, the two batteries have been placed in a thermostatic bath regulated at 40C during the ageing periods, and at 25C during the capacity tests performed at the end of each IEC 61427 period. A description of the IEC 61427 standard procedure is given in the Figure below. It corresponds to a sequence of two phases: one period A consisting of fifty cycles at low SOC (comprised between 10% and 40%) followed by a period B consisting of one hundred cycles at high SOC (comprised between 75% and 100%). The period A corresponds to the behavior of a battery in a photovoltaic (PV) system during a low irradiance period and worst case scenario (like in winter), while the period B corresponds to a high irradiance period (like in summer).
Period A (low SoC) I 100% Period B (high SoC) x 100 II Capacity test
1 : Charge at C5/10 for 15h 2 : 90% Discharge at C5/10 for 9h 3 : 30,9% Charge at 1,03C5/10 for 3h 4 : 30% Discharge at C5/10 for 3h 5 : Full recharge at C5/10 for 20h 6 : 25% Discharge at 1,25C5/10 for 2h 7 : Floating charge at C5/10 up to 1,6V/cell for 6h 8 : Discharge at C5/5 down to 1V/cell 9 : Charge at C5/10 for 15h 10 : Discharge at C5/5 down to 1V/cell

100

80

Tubular PV LABs
60

SLI LABs
40 20

0 0 150 300 450 600 750 900 1050 1200 1350 1500 1650 Number of cycles
140

b)
120

SUN+140 Ni-Cd batteries

100 Capacity (% Ci)

80

60

Tubular PV LABs

40

SLI LABs
20

0 0 150 300 450 600 750 900 1050 1200 1350 1500 1650 Number of cycles

Figure 5: Capacity (expressed in % of the nominal capacity in Fig. a) and in % of the initial capacity in Fig. b)) versus the number of cycles according to the IEC 61427 standard

75% State-of-Charge (SoC)

5 8 9 10

x 50 40%

I : Rab : Heating up to 40C for 16 h II : Rgh : Cooling down to 25C for 16 h III: Rh : Relax time 1 to 72 h
10%

0 0 Time

III

We can observe form these graphs that after ten periods according to the IEC 61427 procedure, the capacity of SUN+140 Ni-Cd batteries is still superior to 90% of the nominal capacity specified by the manufacturer, and the capacity loss is close to 20% of the initial capacity (initial capacity being equal to about 120% of the nominal capacity). Furthermore, the capacity loss observed on SUN+140 NiCd batteries at C/120 is even inferior to the capacity loss observed at C/5: the capacity loss measured after ten IEC 61427 periods is less than 10% at C/120, while it reaches about 20% at C/5, as indicated in the Table below:

Figure 4: Accelerated ageing test according to the IEC 61427 standard Eleven IEC 61427 periods (each period being equivalent to one hundred and fifty cycles, including the fifty cycles of the Period A and the one hundred cycles of the Period B) have been completed on the two studied batteries. 3.2 Results Results obtained are presented in the Figures below, in which SUN+140 Ni-Cd batteries are indicated in pink. The other colors correspond to two categories of leadacid batteries: Start Lighting and Ignition (SLI) LABs in brown and tubular PV LABs in green. More details about results obtained with LABs can be found in the Reference [3]. These results confirm the good behavior of SUN+140 Ni-Cd batteries in terms of ageing, which offer similar performance than best tubular LABs.

Table 6: Capacity losses after ten IEC periods measured respectively at C/120 and at C/5 Bat_1 Initial C/5 value (Ah) C/5 after ageing (Ah) Capacity losses in Ah Capacity losses in % Initial C/120 value (Ah) C/120 after ageing (Ah) Capacity losses in Ah Capacity losses in % 154.2 118.5 35.7 23% 170.7 157.5 13.2 8% Bat_2 151.5 119.6 31.9 21% 168.4 156.2 12.2 7%

4 PERFORMANCE IN A STAND-ALONE PHOTOVOLTAIC SYSTEM ACCORDING TO THE

IEC 62124 STANDARD 4.1 Experimental The purpose of the IEC 62124 standard [4] is to: ensure the system and load can operate as expected; ensure the PV array is capable of recharging the battery; determine if there is any significant change in the usable capacity during the short-term performance tests. A schematic description of the test procedure is given in the Figure below, in which UBCx correspond to the different capacity tests, FT is a Functional Test and RT a recovery Test:
UBC0 FT
Off On Off On On Off On Off On On

performed on a new ENERSOL 130, at 8 A (C5/5) and 25C. Its average capacity measured at this current rate is 84.5 Ah +/- 0.7 Ah. For this test, three new SUN+140 batteries constituted by ten elements in series (Unominal = 12 V) have been assembled and cycled. In summary, here is the initial measured capacity of each battery: Bat_channel_1 (SUN+140): C5/5 (25.6 A) = 149.8 Ah; C120/120 = 164.8 Ah Bat_channel_2 (SUN+140): C5/5 (25.6 A) = 148.8 Ah; C120/120 = 163.7 Ah Bat_channel_3 (SUN+140): C5/5 (25.6 A) = 148.8 Ah; C120/120 = 163.7 Ah Bat_channel_4 (ENERSOL130): C5/5 (16 A) = 85.2 Ah; C120/120 = 137.2 Ah
Off On

UBC1
On On Off Off Off On Off On

RT
On/Off On/Off On Off

UBC2
On Off Off On

PV PV Charge Load Voltage Tension (V)

VREG

On Off

On Off

72 h

30 min

30 min

72 h

LVD 5h 5h 5h

11
0

Temps

Time

Figure 6: Schematic test procedure according to the IEC 62124 standard

We can underline that, due to the difference between the Peukerts coefficient of a Ni-Cd and a LAB (see the section 1.2), similar capacities for a Ni-Cd and for a LAB at a given discharge current rate result in different capacities at a higher discharge current rate. Furthermore, we can also mention here the difference existing between the nominal capacities (given by manufacturers) and the measured capacities. These facts involve a difficulty in comparing Ni-Cd and LABs. 4.2 Results The IEC 62124 standard has been realized simultaneously on four stand-alone systems, whose three was equipped with SUN+ batteries and the fourth one with a LAB. A Functional Test (FT in the Figure 6) of fourteen days has been performed on each channel, showing that a stand-alone PV system equipped with SUN+140 Ni-Cd batteries work properly under normal system operation. Moreover, these batteries offer more autonomy than the LAB with a similar nominal capacity, the difference observed being higher than 50% during this test! This difference is explained by the two following main factors: the Depth-Of-Discharge (DOD) is limited in the case of the LAB and not in the case of SUN+140 Ni-Cd batteries. Indeed, a Low Voltage Disconnect (LVD) equal to 1 V/ele can be used with Ni-Cd battery, which means that 100% of the capacity can be discharged for this type of battery instead of 80% down to 60% in our test for LABs for which the LVD is generally fixed to 1.9 V/ele in the field. A deep discharge could effectively have a bad impact on the ageing of LABs, while it is without any consequence in the case of Ni-Cd batteries; the Peukerts coefficient is lower for Ni-Cd batteries than for LABs, which means that a higher part of a similar nominal capacity given at C120/120 can be discharged at higher discharge current rates for Ni-Cd than for LABs. In spite of the simultaneous start of the test, we have observed some time-lags in the test progression between channels: the channel 4 was in advance compared to the three other channels, which can be easily explained by the fact that the use of a LAB on this channel results in a lower

An outdoor test bench called PLISE (for PV Lighting System and Evaluation) has been used to perform the IEC 62124 standard. It is composed by four stand-alone PV systems, that is to say four test channels with the following configuration and management strategy: Table 7: Configuration of the 4 channels tested according to the IEC 62124 standard
Channel 1 Battery / Capacity C120 PV array / Power Load / Current - Ah/day Regulator Charge regulation Discharge termination Channel 2 Channel 3 Channel 4 LAB (ENERSOL) /130 Ah SUN+140 Ni-Cd battery (SAFT) / 140 Ah

Polycristalline Silicon Module (PHOTOWATT) / 125 Wp (75 Wp + 50 Wp) Enerstat (TENESOL) PWM at 15,5 V LVD at 10,0 V Dynamic load (SODILEC) / 3 A ; 21 Ah/day Enerstat (TENESOL) Bz Products with MPPT On-off at 14,5/16,5 V Floating at 15,5 V LVD at 10,0 V LVD at 10,0 V Enerstat (TENESOL) On-off at 13,8/14,7 V LVD at 11,4 V

All batteries have been kept at a constant temperature of 20C +/- 3C during this test. We can also underline here that a charge regulator basing on a MPPT (Maximum Power Point Tracking) algorithm has been used on the channel 3, in order to observe if this type of charge controller enables to have a gain of Ah/Wh charged compared to other channels. The channels 1 and 2 correspond to optimized end-of-charge regulation, respectively in PWM mode (Pulse Width Modulation) and in On-Off mode (two voltage thresholds). Finally, the channel 4 includes a flooded LAB with the regulation typically used in this case on the field; this channel can thus be used as a reference of current stand-alone PV system. The characteristics of the LAB ENERSOL 130 are the followings: C120/120 = 132 Ah, C100/100 = 130 Ah, and C5/5 = 81.9 Ah considering a Peukerts coefficient of 1.15. Five conditioning capacity tests have been

Usable Battery Capacity (UBC) and so in shorter UBC0 and UBC1 steps, and also in less Ah necessary to reach Vreg after these capacity tests; the channel 3 is late compared to channels 1 and 2, which can be explained by the relatively bad impact of the MPPT regulator on the charge of the battery 3. Indeed, if the MPPT algorithm is developed to give more amperes from the PV array (and so more ampere-hours), this algorithm is only effective when the battery is not in the end-of-charge regulation mode. Moreover, we have observed during this test that the MPPT regulator used was not effective at low irradiance. Consequently, the MPPT regulator gives in this test no benefice and even less Ah charged per day compared to the channels 1 and 2 without MPPT. By the way, we can note here that a detailed study of the MPPT impact in stand-alone PV systems has been realized by the L2S [5]. Furthermore, a Recovery Test (RT in the Figure 6) of about 20 days have been performed, consisting in discharging completely the batteries in order to observe how they were able to reach the High Voltage Disconnect (HVD) again. The outcomes of this second part of the test are the following: on the channels 1, 2 and 3, the SUN+140 Ni-Cd batteries have performed properly the charges and discharges of the step RT, although they have not reached the HVD before the interruption of the discharge phases (after six consecutive discharges according to the standard); on the contrary, on the channel 4, the LAB has reached the HVD immediately, although the battery was weakly charged! After that, few Ah was charged during each charge step, whereas the daily consumption was maintained to 21 Ah. Consequently, the Low Voltage Disconnect (LVD) has been reached after only six days in the RT phase! Then, since the used regulator did not authorize the discharge to be purchased before reaching the Low Voltage Reconnect (LVR), and since the total of 21 Ah was not reached, the test procedure had to be abandoned on this step! Hence, we can deplore that the LAB has not passed the RT phase, due to a lack of available autonomy! After a new capacity test on the LAB, it appears that the problem observed during the RT step was not due to a premature capacity loss of the battery: effectively, the capacity measured at C5/5 (16 A) is still equal to 85.7 Ah after this test (instead of 85.2 Ah before the test). The results of the different Usual Battery Capacity tests performed during the IEC 62124 standard are added in the Table below: Table 8: Usual Battery Capacity tests UBC0, UBC1, and UBC2
System Units Channel 1 SUN+ C120 = 140 Ah Channel 2 SUN+ C120 = 140 Ah UBC0 Ah 146.2 UBC1 Ah 128.1 UBC2 Ah 118.2 UBC1 / UBC0 % 87.6 UBC2 / UBC0 % 80.8

Channel 3 SUN+ C120 = 140 Ah Channel 4 LAB C120 = 130 Ah

142.8

119.9

106.3

84.0

74.4

76.2

53.6

Test stopped!

70.3

Test stopped!

UBC2 is sensed to give an indication on the State-OfHealth (SOH) evolution of each battery during the test. Nevertheless, due to electric history effect, this capacity test is not very accurate. Hence, new capacity tests have been performed on a specific test bench following the complete capacity test procedure (residual discharge at C/5, rest time of one hour, full charge of fifteen hours at C/10, rest time of one hour, capacity test at C/5 down to 1 V/element). The results given in the Table below show that the capacity losses of batteries are not relevant, contrary to the indication obtained from UBC2 values. Table 9: Estimation of the SOH of batteries at the end of the test
System Units Channel 1 SUN+ Ni-Cd Channel 2 SUN+ Ni-Cd Channel 3 SUN+ Ni-Cd Channel 4 LAB C/5 measured before the test Ah 149.8 148.8 148.8 85.2 C/5 measured after the test Ah 143.1 146.3 146.3 85.7 Estimation of SOH % 96 98 98 101

CONCLUSION The outcomes of this study are the following: The performance of SUN+140 Ni-Cd batteries in term of faradic and energy efficiencies between SOC = 0% and SOC = 50% (according to the IEC 62093) is quite similar than for the LABs: about 100% for the faradic efficiency and about 87% for the energy efficiency. Accelerated ageing tests (according to the IEC 61427) performed on two SUN+140 Ni-Cd batteries and on a lot of different types of LABs showed that SUN+140 batteries present similar performance in term of ageing than the best LABs, that means with tubular positive electrodes. The clearer advantage of SUN+140 Ni-Cd batteries compared to LABs consist in the influence of the discharge current rate on the delivered capacity: for a similar capacity given at C/120, the capacity of SUN+140 batteries is about 1.3 times higher than the capacity of LABs at C/20, 1.6 times higher at C/5, etc. This behavior can thus involve a big difference in term of autonomy between these two types of batteries considering the discharge current rate. Another advantage of Ni-Cd batteries compared to LABs is the possibility to discharge them completely without any impact on their ageing, contrary to LABs which have to be discharged down to SOC = 20% only to protect them from premature capacity loss. Finally, a test of SUN+140 batteries used in three stand-

146.8

124.2

110.4

84.6

75.2

alone PV systems under real irradiation conditions (according to the IEC 62124) have shown that these three systems worked properly, and offered higher hours of autonomy than a fourth PV system equipped with a flooded LAB as reference. The difference observed has reached more than 50%, which can be explained by the two previous remarks. Hence, these two points must be considered with attention in the sizing of stand-alone PV systems, as for economical analyses of PV systems, depending on the type of batteries used.

REFERENCES [1] IEC 62093, Balance-of-system components for photovoltaic systems Design qualification and type approval (1997). [2] IEC 61427, Secondary cells and batteries for photovoltaic energy systems General requirements and methods of test (2002). [3] E. Lemaire-Potteau et al., Assessment of storage ageing in different types of PV systems: technical and economical aspects , Proceedings of the 23rd EU PVSEC - Valencia, Spain (2008). [4] IEC 62124, Photovoltaic (PV) stand alone systems Design verification (2004). [5] M. Vervaart, J. Merten, F. Mattera, Quantitative energy yield assessment of four battery charge regulator topologies , Procedings of the 22nd European Photovoltaic Solar Energy Conference, Milan, (2007).

ASSESSMENT OF STORAGE AGEING IN DIFFERENT TYPES OF PV SYSTEMS: TECHNICAL AND ECONOMICAL ASPECTS

Elisabeth.Lemaire-Potteau, Florence Mattera, Arnaud Delaille, Philippe Malbranche INES CEA/LITEN/Laboratory for Solar Systems (L2S) BP 332 Savoie Technolac, 50 avenue du lac Lman, F-73377 Le Bourget-du-lac, France Phone : +33(0)4 79 44 45 43; E-mail : elisabeth.lemaire@cea.fr

ABSTRACT Successful implementation of storage within PV systems requires a reliable life time estimation, to ensure the economical profitability of projects and, not negligibly, the user satisfaction. However, the main difficulty is that among the different components of a PV system, the storage part is one of the shortest lived, and whose life time is the most difficult to predict. In this paper, the methodology for assessing the storage ageing is described for standalone systems, first for technical aspects dealing with accelerated ageing tests, then for economical aspects with the comparison of storage cost of ownership. Finally, we examine how this approach can be transposed to the case of storage in grid-connected systems. Keywords: Storage, lifetime, testing, economical analysis

1 STORAGE IN THE DIFFERENT TYPES OF PV SYSTEMS PV systems can be classified in two main families: stand-alone and grid-connected systems. In the first case, except for some very specific applications, storage is indispensable to match the energy supply to the users demand. On the other hand, grid-connected systems are most often operated without storage. Nevertheless, the wide deployment of such systems has already led, in some regions, to bottle-neck situations, for instance at times of strong generation and low load. This is particularly true for massive installation of small-medium size PV plants on private houses. Adding a storage device opens up many opportunities for managing the internal energy flow, as well as its behavior from the utility point of view [1]: storage can then make PV predictable. Successful implementation of storage within PV systems requires a reliable life time estimation, to ensure the economical profitability of projects and, not negligibly, the user satisfaction. However, the main difficulty is that among the different components of a PV system, the storage part is one of the shortest lived, and whose life time is the most difficult to predict [2]. In this paper, our approach will first focus on the case of stand-alone systems, where most of our experience has been gained. The methodology for assessing the storage ageing is described, first for technical aspects dealing with accelerated ageing tests, then for economical aspects with the comparison of storage cost of ownership. Finally, we will examine how this approach can be transposed to the case of storage in grid-connected systems.

real operating conditions of PV systems, identifying the constraints the storage is subjected to, knowledge of the storage degradation modes and of their causes, for the different types of storage technologies studied, tear-down analysis of field batteries, and identification and quantification of their degradations. Results of this type were presented at the 22nd EUPVSEC in Milan [3]. A synthetic analysis was performed, based on the ageing data obtained so far at INES (and formerly at GENEC) on more than 150 different batteries of different technologies. Initially, tests were focused on lead-acid batteries, and nickel-cadmium in specific cases [4]: this reflects the present market of storage in PV systems. The approach has now been extended to lithium-ion and nickel-metal hydride batteries, to account for the market evolution for certain types of PV systems. This type of analysis then allows technology comparison, from a technical point of view (life time, ageing), as well as from an economical point of view (initial cost, cost of ownership over a given period, typically 20 years). Another noticeable point is that this technical and economical assessment cannot be dissociated of the sizing issue: depending on the storage technology and of its tolerance to overcharge and/or deep discharge, the storage capacity will not be calculated in the same way. One of the consequences is that the operating constraints of the storage will also differ.

3 2 METHODOLOGY FOR THE ASSESSMENT OF STORAGE AGEING The methodology proposed here is relies on data and expertise of our team in different topics:

TECHNICAL ASPECTS

When examining the suitability of an ageing test to select storage for a given application, two questions need to be answered, which we will now examine successively:

Qualitatively: is the test representative of the application constraints? Quantitatively: what is the acceleration factor? How can the test be used to predict the storage life time?

The existing test procedures mainly use two types of approaches [2, 4]: Simulation of the system operation in an accelerated way, Reproduction of degradations observed in the field. 3.1 Tests based on the battery operating conditions With the first type of procedures, based on the system operation profile, representativeness is rather easily ensured. The present standards [5, 6] use this approach. A schematic evolution of battery state of charge (SOC) during these tests is given in Figure 1. For both procedures, the test profile is a superposition of microcycles, simulating the daily operation, over a macrocycle, representing the seasons. Figure 2 is an example of our synthetic data analysis for the accelerated test procedure proposed in the IEC 61427 standard. The capacity loss is plotted against the number of equivalent full discharges, which is the cumulated discharged Ah, normalised to the battery capacity. This allows comparing batteries with different capacities and tests with different depth of discharge. In the standard, test parameters are given only for lead-acid and nickel-cadmium batteries. For the other types of batteries (here for lithium-ion batteries), the voltage thresholds were adapted following the manufacturers recommendations.

After plotting the capacity loss curves for each battery tested, groups are made according to the battery type. Provided that the number of tested batteries is sufficient, from different manufacturers, these capacity loss areas are representative of the ageing of each battery type. This type of graphs is continuously enriched by adding new test data obtained. Then, for each type of batteries, a general ageing tendency can be derived (full lines on Figure 2). The value of the obtained slope is given in Table 1, as well as the number of batteries tested. It is expressed as the capacity loss (in %) for 100 equivalent full discharges. Of course, there is quite a large incertitude over this slope value, but this is also due to the very large dispersion between batteries of a given type, and even of a given model from different batches. The test results obtained with the procedure of the NFC 58-510 French standard were analysed following the same method. Table 1 also gives the ageing parameters for this test. We can see that capacity loss is noticeably faster with this second procedure, which could maybe inspire a future revision of the IEC 61427 standard. However, practical implementation of the French test is more difficult, which may limit its use. Table 1: Ageing rate for the two standard procedures
Technolog y Pb, solar Pb, tubular Ni-Cd Li-ion Nb bat. 8 7 4 8 IEC 61427 Ageing / 100C - 30% - 5% - 5% - 1.7% NFC 58-510 Nb Ageing / bat. 100C 8 - 60% 5 - 15% ? ?

summer winter
IEC 61427 NFC 58-510

3.2 Tests based on the reproduction of degradations This approach was developed initially for lead-acid batteries. It is based on the knowledge of battery degradations encountered in the field for the different types of systems. Specific procedures were developed in order to induce specifically each type of degradation [7]. These procedures are shown schematically on Figure 3.

Figure 1: Schematic evolution of SOC for the two PV standard test procedures
20
Li-ion Li-

efficiency

stratification

shedding

0 Capacity (% Cnom)

Ni-Cd NiPb tubular

-20

deep discharge

corrosion

-40
Pb "solar" "solar"

Figure 3: Schematic SOC evolution for the degradation based tests Then, depending on the applications, one or several procedures can be selected. The great advantage of this strategy is that testing is much faster: the different tests can be run in parallel, and their duration is much shorter (between 1-2 weeks and 6 months). As a comparison, the IEC 61427 procedure requires more than 2 years testing for the best lead-acid batteries, and after 3 years testing, the lithium-ion batteries have only lost 10% capacity;

-60

-80 0 100 200 300 400 500 Number of equivalent full discharges 950

Figure 2: Synthetic analysis of ageing results obtained with the IEC 61427 test procedure

However, transposition of these procedures to other battery technologies is possible. They will then not induce the same degradations, but will be representative of the battery ability to withstand certain specific constraints of the application. Efficiency, measured between 0 and 50% SOC, is the easiest one. It only requires the definition of 100% SOC for each new battery. For this, following the manufacturers recommendations is most often sufficient. Having a high efficiency battery is important, but we believe that it is particularly important at low SOC, in bad weather conditions: from the energy stored in the battery, the maximum amount has to be given back. Figure 4 shows the results obtained for all batteries tested up to now, and the number of batteries tested of each type. However, it is important to keep in mind that this is only the efficiency of the battery itself. Its use in a PV system generally requires other components, which will lower the overall charge/discharge efficiency of the storage. The components required depend on the battery technology: a lead-acid battery will always need a charge controller, while a nickel-cadmium one can work without, and a lithium-ion battery will require each element to be monitored and charged separately. Minimising the energy losses in these components is crucial in order to benefit for instance of the high efficiency of a given battery.
1 0.95 0.9 0.85 0.8

number of equivalent full discharges is reached about three times faster with this procedure, since the discharge current is higher.
20

Capacity (% Cnom)

Pb tubular

-20

-40

Pb VRLA

-60
Pb SLI Pb solar

-80

50

100

150

200

250

300

350

Number of equivalent full discharges

Figure 5: Synthetic analysis of ageing results obtained with the shedding test procedure Table 2: Ageing rate for the shedding procedure
Technology Pb, solar Pb, tubular Pb, VRLA Pb, SLI Nb bat. 10 3 11 2 Ageing for 100 full discharges - 30% - 15% - 15% - 60% / 50 discharges

The deep discharge procedure examines how the battery withstands a prolonged period (one month) without being recharged, and how it will then recover. Finally, in the corrosion procedure, the battery is simply kept in floating mode, simulating a UPS-type operation. These two procedures are also easily transposable to other battery technologies than lead-acid. 3.3 Relation with field life time The next step in our analysis is now to establish the relation between the ageing data obtained in the laboratory and effective battery life time in the field, and to estimate the acceleration factor of the tests. In the past, this factor has often been assessed only by theoretical methods. The first point to be clarified is the notion of life time in the field. For professional PV systems (e.g. telecom systems), the criterion will often be the same as for other applications of batteries (UPS). Batteries are considered at the end of their life when they have lost 20% capacity. In some other cases, such as offshore marine signals, batteries will be replaced with a fixed periodicity, because on-site visits are extremely expensive. For domestic systems, used in private households, end of life is often determined by users complaining about their battery! The capacity loss in this case can vary between 20 and 80%. A good estimation of the procedures acceleration factor can then not rely on field life time. The correct parameter is the batteries capacity loss after a certain time of operation in a real system. In a previous paper [3], we have reported the expertise of field batteries after operation in SHS in Morocco, and in domestic systems in Guadeloupe island.

Solar

VRLA

Tubular

Ni-Cd Ni-

Li-ion Li-

21 5 Nb. batteries

12

8(elts)

Figure 4: Wh efficiency for 0-50% SOC operation The so-called stratification procedure will in fact test how the battery will suffer from insufficient recharge for several cycles, then how it will recover from that treatment when good recharge is then provided. This is, again, representative of some specific operating conditions it can be subjected to. In the same way, the so-called shedding procedure is a cycling test with 66% depth of discharge, higher current than in the real operation, and high temperature. It could be considered as well as belonging to the first group of procedures, those simulating the system operation in an accelerated way. The results obtained with this procedure have been subjected to the same analysis as for the IEC 61427 procedure (Figure 5). For the moment, only data with lead-acid batteries is available. In the same way, Table 2 gives the mean ageing rate for the different types of batteries. The values obtained are comparable to that of the two standard test procedures (Table 1), but the same

After 20 months of use, the capacity loss of the SHS batteries was 20%. Identical batteries tested with the IEC 61427 procedure reached this capacity loss after 1800 hours (60 equivalent full cycles). The acceleration factor is then around 8 (20 months / 1800 hours). For the domestic systems, the batteries had lost 30% capacity after 9 years. With the IEC 61427 procedure, these batteries lose about 5% capacity for every 100 equivalent full cycles. 30% capacity loss would then be reached after 7200 hours of testing. The acceleration factor is then around 11 (9 years / 7200 hours). These two values of the acceleration factor for the IEC 61427 (8 and 11) are coherent, taking into account the dispersion already mentioned above (3.1).

4.2 Case of domestic systems Domestic systems used for rural electrification typically use flooded tubular lead-acid batteries, and 0.5 to 10 kWp PV array. We have compared here the cost of this type of battery with that of large capacity lithium-ion batteries. In both cases, we have assumed a 12 kWh battery (e.g. 24 V - 500 Ah) with 5 days autonomy. The system site is supposed to be accessible quite easily (typically by car), and maintenance to be performed every year for lead-acid, and every 10 years for lithium-ion storage. Table 3 shows the results obtained for this system, for the two battery technologies. We observe that, even including maintenance, the lead-acid solution is still the most economical. Of course, these figures are quite dependent on the initial hypotheses, especially the cost of the lithium-ion battery, which is quite different from one manufacturer to another. However, the life time can also be rather dependent on the manufacturer. Table 3: Economic assessment for domestic systems
Battery technology Initial battery cost () Expected life time (years) Total number of cycles Total discharged energy (kWh) Cost of ownership (/kWh) Cost with maintenance (/kWh) Li-ion 12 000 20 7300 17 520 0.68 0.70 Pb 1 800 8 2920 7 008 0.26 0.53

ECONOMICAL ASPECTS

Besides these technical aspects, the economical aspects of the storage selection also need to be analysed. We propose here a simple approach taking into account initial costs, expected life, and maintenance costs. As an example, this methodology is applied here to the comparison of lead-acid and lithium-ion batteries for two different types of PV systems: domestic systems for rural electrification, and street furniture (parking meters, street lighting). 4.1 Methodology We have based our approach on the comparison of cost of ownership. However, initial investment cost can mitigate these conclusions: even if one technical solution is the most economical one in the long-term, the available financial effort at the time of the system installation could also limit the use of the most expensive storage technologies. The cost of ownership is expressed as the cost per kWh supplied to the loads by the storage. This cost also corresponds to that of each kWh discharged from the storage. It is calculated in several steps. First, the total amount of energy discharged during the entire battery life is taken from the technical assessment results, if available. If not, it is calculated using the following steps: 1. the theoretical total number of cycles performed during the whole battery life is calculated from the estimated field life time and the daily number of cycles, 2. the daily depth of discharge (DOD) is deduced from the system autonomy and the maximum allowed DOD, 3. these two figures, combined with the battery sizing, give the total discharged energy. The total cost is the addition of the initial cost for purchasing the storage, and of the maintenance cost over the battery life time. Then, the cost of ownership, as defined just above, can be calculated as the ratio between the total discharged energy and the total cost. Using this method only requires a few simple calculation steps, and has the great advantage of being technology-neutral: it does not depend on the storage technology considered, and enables comparison of the different battery technologies and types.

4.3 Case of street furniture Street furniture includes all objects installed in a city for satisfying the users requirements. A lot of these objects need energy for their operation: public lighting, information boards, parking meters, public shelters for transport users, etc. The cost for connecting these systems to the conventional electricity grid is often rather high, which justifies the use of PV energy. This market of PV-powered street furniture is growing strongly because it also gives a good image of the local public community. In some cases, the operating constraints can be rather strong: non-optimum PV module orientation, high DOD, high temperature variation The battery needs to be sealed, and we have taken the hypothesis of a valveregulated lead-acid battery (VRLA), with a rather short life time due to the strong constraints (2 years only). The compared technology is again lithium-ion, which is less sensible to these constraints, and the assumed life time is still 20 years. These systems are smaller (a few 100 Wp), and we have taken the case of a 0.6 kWh battery (e.g. 12 V 50 Ah), with 3 days autonomy. Table 4 shows the cost of ownership obtained for these systems. In this case, lithium-ion batteries, even if they are most expensive, prove to be competitive on the long term, and the small system size can make the initial cost affordable for the municipalities. Table 4: Economic assessment for street furniture
Battery technology Initial battery cost () Expected life time (years) Li-ion 600 20 Pb 120 2

Total number of cycles Total discharged energy (kWh) Cost of ownership (/kWh)

7 300 1 314 0.46

730 131.4 0.91

5 EXTENSION OF THIS APPROACH TO STORAGE IN GRID-CONNECTED SYSTEMS The next step is now to extend this study to the more recent use of storage in grid-connected PV systems. This issue of storage assessment in grid-connected configuration is not limited to PV systems: without being associated to PV modules, storage systems can be installed and shared at different levels from central installations (several hundred MW) to dispersed installations across the networks or at the customer (from 10 MW down to the kW range, for instance plug-in hybrid cars). Up to now, the applications of storage in electricity grids are mainly driven by the regulatory framework and by the tariff structure. These applications are very different in terms of operating constraints, and the sizing and storage technology selection will need to be adapted to these constraints. The selection of the best suited storage for a given application needs to be based on precise criteria and associated test procedures. These are well defined for the traditional applications, such as uninterruptible power supply (UPS) or stand-alone configuration, but this is not the case for grid-connected applications. Especially in this context, storage systems can be operated in energy mode or in power mode. The storage technologies can also be very different from the traditional ones in PV, and one will also find very large batteries, such as redox-flow or sodium-sulfur batteries. Their present cost is around 1 k / kWh. However, the need for common testing methods is now getting critical. Indeed, in the present situation a different test profile is used for each test to be realised, and this prevents any reliable comparison between test results. The approach used for stand-alone systems cannot be directly transposed to grid-connected systems, because the availability of field monitoring data, and even more of used field batteries, is limited. The only possibility is then to start from the applications, and more precisely from their design operating profiles, and to simplify and classify them while keeping their representativeness. In the design of the test profiles, one major difference with the case of stand-alone systems often lies in the daily DOD, which is limited in this latter case in order to comply with the autonomy requirements. For gridconnected systems, autonomy is in most cases not really meaningful, and this leads to stronger cycling conditions (higher power, higher DOD etc). The test procedures must reflect these conditions while accelerating the test compared to real operation. This transposition is the most difficult step, and relies only on the teams knowledge and expertise, since very little field data is available for the procedures validation.

The synthetic data analysis presented in this paper forms the basis for answering the question of assessing storage life time in PV systems: For stand-alone systems, technical assessment leads to an estimation of the acceleration factor of the IEC 61427 standard test, based on the coupling of field data, battery expertise and synthesis of the available test data. A simple methodology is proposed for the economical assessment. For grid-connected systems, directions for developing suitable test procedures are given. Further work on this topic should include: Based on this methodology, development of a software for supporting the decision-making process and storage design, Extension of this approach to other storage technologies, Development of a similar approach for gridconnected applications.

ACKNOWLEDGEMENTS This paper includes results of projects funded by ADEME (French Agency for Energy Management and Environment) and by the European Commission.

REFERENCES [1] Energy management for grid-connected PV plants combined with battery storage devices, C. O. Heyde, H.-D. Musikowski, Z. A. Styczynski, 21th European Photovoltaic Solar Energy Conference, 2006, Dresden [2] Results and comparison of seven accelerated cycling test procedures for the photovoltaic application, E. Potteau, D. Desmettre, F. Mattera, O. Bach, J.-L. Martin, P. Malbranche, Journal of Power Sources 113 (2003) 408-413. [3] Lessons learned on storage for different types of stand-alone PV systems: monitoring and on-site expertise, E. Lemaire-Potteau, A. Delaille, F. Mattera, B. Rouvire, M. Chafik, G. Olivier, M. Robert, 22nd European Photovoltaic Solar Energy Conference, 2007, Milan [4] Test procedures of storage systems based on leadacid batteries, E. Lemaire-Potteau, P. Malbranche, Storage for Renewable Energies Conference, 2003, Aix en Provence [5] IEC 61427-1 standard: Secondary cells and batteries for solar photovoltaic energy systems General requirements and methods of test [6] NF C 58-510 standard: Lead-acid secondary batteries for storing photovoltaically generated electrical energy [7] D. Desmettre, F. Mattera, P. Malbranche, S. Mtais, Publishable report of the QUALIBAT project EU JOR3-CT97-0161 (1999)

CONCLUSION AND FURTHER WORK