Sunteți pe pagina 1din 9

1132 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 52, NO.

6, JUNE 2005

Analysis of the Parasitic S/D Resistance in


Multiple-Gate FETs
Abhisek Dixit, Anil Kottantharayil, Member, IEEE, Nadine Collaert, Mike Goodwin, Member, IEEE,
Malgorzata Jurczak, and Kristin De Meyer, Senior Member, IEEE

Abstract—The multiple-gate field-effect transistor (FET) is a [9]–[11]. These devices have been reported to achieve the target
promising device architecture for the 45-nm CMOS technology performance [12] with CMOS compatible processing.
node. These nonplanar devices suffer from a high parasitic For a CMOS technology to keep up with the downscaling,
resistance due to the narrow width of their source/drain (S/D)
regions. We analyze the parasitic S/D resistance behavior of the improved carrier transport and low parasitic source/drain (S/D)
multiple-gate FETs using a novel, S/D geometry-based analyt- resistance are required [13]. As the device operation in
ical model, which is validated using three-dimensional device MuGFETs relies on the use of narrow fins, drive currents could
simulations and experimental results. The model predicts limits be severely degraded if special provisions are not made to mini-
to parasitic S/D resistance scaling, which reveal that the contact mize [14]. The most commonly used model for series resis-
resistance between the S/D silicide and Si-fin dominates the par-
asitic S/D resistance behavior of multiple-gate FETs. It is shown tance estimation in planar bulk MOSFETs [15] is inappropriate
that the selective epitaxial growth of Si on S/D regions alone may for MuGFETs due to the three-dimensional (3-D) architecture
be insufficient to meet the semiconductor roadmap target for of these devices. The main shortcoming of the existing model is
parasitic S/D resistance at the 45-nm CMOS technology node. its inability to capture series resistance issues arising from the
Index Terms—Analytical model, Fin field-effect transistors narrow width of S/D regions and current conduction on planes,
(FinFETs), fully depleted, series resistance, silicon epitaxy, sil- perpendicular to the S/D contacts. Although Kedzierski et al.
icon-on-insulator (SOI) MOSFET, small geometry, source/drain [16] have addressed this issue and proposed a technology solu-
(S/D), (110) transport. tion, an analytical understanding of parasitic series resistance in
the MuGFETs is desirable.
I. INTRODUCTION This paper is organized in seven sections. In Section II, the
device structure is explained and details of the device simula-

W ITH THE scaling of CMOS into sub-90-nm technology


nodes, control over short-channel effects (SCEs) be-
comes increasingly difficult in planar-bulk architecture. In order
tions are given. The analytical model for is described in
Section III. The process flow used for Omega-FET [7] fabrica-
tion is briefly described in Section IV. Both the procedure used
to achieve the target subthreshold performance [1], high channel for extraction of the parasitic S/D resistance, and validation of
doping concentrations are needed in planar bulk technologies.
the analytical model using 3-D device simulations and experi-
This gives rise to significant dopant-induced fluctuations in mentally measured data are discussed in Section V. The key de-
threshold voltage, and degradation in channel mobility [2]. vice parameters that dominate behavior of the MuGFETs
A transition from planar bulk to the double-gate architecture are identified in Section VI and their minimization is discussed.
could facilitate the target subthreshold performance while also The conclusions are given in Section VII.
keeping the channel-doping concentration low, if ultrathin Si
films and metal gates can be used to control SCEs and adjust II. DEVICE STRUCTURE AND SIMULATIONS
the threshold voltage [3], respectively.
One of the promising device architectures for sub-65-nm An n-channel double-gate MuGFET structure [5] has been
CMOS technology nodes, originally proposed in 1980s as modeled using ISE-MESH [17] and is shown in Fig. 1(a). Cross
XMOS [4], relies on multiple gates to control SCEs. More sections along the lines A-A and B-B are shown in Fig. 1(b)
refined members of this family of multiple-gate field-effect and (1c), respectively. In the simulation mesh of the MuGFET
transistors (MuGFETs) are FinFETs [5], tri-gate FETs [6], structure shown in Fig. 1(a), no grid points have been placed in
Omega-FETs [7], Pi-gate FETs [8], and gate-all-around FETs the S/D spacers. Abrupt, box-shaped S/D extensions and heavily
doped S/D (HDD) profiles have been defined analytically. The
S/D-silicide regions have been modeled as 3-D, area depen-
Manuscript received November 5, 2004; revised February 17, 2005. This dent distributed resistances [17]. In the device simulations using
work was supported by the IMECs Industrial Affiliating Program under the
Industrial Affiliation Program (IIAP) on the Emerging Alternative Devices ISE-DESSIS [17], the contact resistivity of the S/D-silicide has
(EMERALD). The review of this paper was arranged by Editor R. Singh. been set to a constant value of cm corresponding to
A. Dixit and K. De Meyer are with the Inter-University Microelectronics the target for 65-nm technology node [1]. Dashed rectangles in
Center (IMEC) Leuven, Leuven B-3001, Belgium and also with the Depart-
ment of Electrical Engineering (ESAT), Katholieke Universiteit Leuven (KUL), Fig. 1(c) show the hypothetical location of the S/D-silicide. The
Leuven B-3001, Belgium (e-mail: Abhisek.Dixit@imec.be). thickness of an additional Si layer on the HDD regions ,
A. Kottantharayil, N. Collaert, and M. Jurczak are with the Inter-University has been included as one of the variables in the device structure.
Microelectronics Center (IMEC) Leuven, Leuven B-3001, Belgium.
M. Goodwin is with Texas Instruments Inc., Dallas, 75243 TX USA. This additional Si layer is used to mimic the Selective epitaxial
Digital Object Identifier 10.1109/TED.2005.848098 growth (SEG) of Si on the HDD regions with the assumption
0018-9383/$20.00 © 2005 IEEE
DIXIT et al.: ANALYSIS OF THE PARASITIC S/D RESISTANCE 1133

TABLE I
NOMENCLATURE, SYMBOLS, AND VALUES OF THE PARAMETERS
USED IN THE SIMULATED DEVICE STRUCTURE

Fig. 1. MuGFET structure used in simulations: (a) 3-D view, (b) cross section
along the line A-A , and (c) cross section along the line B-B .

that is the same on the top surface and sidewalls of the Si


fin.
Drift-diffusion equations, coupled with the Continuity and
Poisson’s equations, have been solved for the device structure
shown in Fig. 1(a) using ISE-DESSIS [17]. In case of industry
standard (100) Si wafers, the mesa etch results in (110) fin side-
walls [16]. Due to this reason, the electron mobility has been
adjusted for conduction on (110) fin sidewalls in device sim-
ulations. The choice of a double-gate structure for this work
helps in avoiding mobility issues due to the current conduc-
tion on two different crystallographic planes, e.g., in case of a
triple-gate structure, the top surface is (100) whereas the side-
walls are (110). A double-gate operation has been ensured in
device simulations by the use of a thick oxide on the top surface
of the fin, as can be seen in Fig. 1(b) and (1c).
Device parameters used in the structure shown in Fig. 1(a) and Fig. 2. Schematic top view of a MuGFET with two fins. The contributions
to the parasitic resistance arising from different regions of the S/D geometry
their symbols are given in Table I. Two of the parameters listed are shown. Only one half of the device is shown as the device is symmetrical
in Table I, namely and have been used as the sim- about the center of the gate. The top surface of the fin is shown bare whereas
ulation variables. All of the device parameters have been kept it is always covered by the gate stack. The symbols shown in this figure are
explained in Table I.
fixed at their reference values shown in Table I, unless specified
otherwise.
shown in Fig. 2 is that the HDD regions can be wider than the
III. ANALYTICAL MODEL FOR THE PARASITIC S/D S/D extensions, only when an SEG of Si has been done in the
RESISTANCE IN MUGEFTS HDD regions. Thus, exists only in presence of this SEG
layer on the HDD regions. Resistance in Fig. 2 is the area
Parasitic series resistance in the S/D regions of a MuGFET
dependent contact resistance between the S/D-Si and S/D-sili-
has contributions from its components arising from different
cide regions.
parts of the S/D geometry. These components and their loca-
In the case of graded S/D extension doping profiles, is
tions in the S/D regions are shown schematically in Fig. 2. In
not easily separable from the active channel resistance, i.e.,
Fig. 2, is the accumulation resistance in the region of S/D
is considered as a part of [18]. As the device simulations
extensions overlapped by the gate. is the spread resistance
have been done with abrupt, box-shaped S/D extension doping
due to the spread of current from the thin accumulation layer
profiles, can be ignored. The resistivity of the accumulation
into the S/D extensions. is the resistance of the S/D exten-
layer can be expressed as [18]
sions beneath the S/D spacers (not shown in Fig. 2). is the
resistance due to the spread of current from the S/D extensions
into the wider HDD regions. An assumption made in the device (1)
1134 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 52, NO. 6, JUNE 2005

Fig. 4. Contour plot of electron current density (A/cm ) simulated at V =


Fig. 3. Schematic illustration of the spread resistance model: (a) commonly 0:05 V and V = 1:0 V for n-channel MuGFETs. This reference device
used model for planar-bulk MOSFETs (b) extended model for MuGFETs. assumes an overlayer of 20-nm Si on the HDD regions to mimic the SEG. Planes
A and B represent the interfaces between S/D silicide and HDD-Si, where the
drain current is collected.
where is the gate-bias dependent electron mobility in the
accumulation layer present in the regions of S/D extensions un-
the process conditions that determine the junction depth of the
derneath the gate, is the gate oxide capacitance. It can be
S/D extensions.
seen from (1) that as becomes infinitely large, dimin-
The sheet resistance in the S/D extensions in Fig. 2 can
ishes to zero. As is extracted at large values of , even
be modeled as
in cases when the S/D extension doping profiles are not abrupt,
e.g., in fabricated devices, can be safely ignored for the pur-
(4)
pose of modeling.
The most commonly used model for estimation of the spread
resistance in planar MOSFETs [19] is illustrated schematically where is the length of the S/D extensions, which is as-
in Fig. 3(a). According to this model, the spread resistance is sumed to be equal to the S/D spacer bottom width in this model.
given by To model the parasitic resistance contributions arising from
the HDD regions and the S/D silicide contacts, 3-D nature of
current conduction in MuGFETs needs to be considered. In
(2)
order to trace the path of conduction current, a 3-D device sim-
ulation has been done using ISE-DESSIS [17] for the reference
where is the sheet resistance of the source region, is the device. This reference device assumes values of the constants
source junction depth, W is the channel width, and is the and variables same as those given in Table I, except for ,
channel thickness. We have extended this model for the case which is assumed to be 20 nm. The electron current density
of double-gate MuGFETs as shown in Fig. 3(b). The modified contours resulting from this device simulation are shown in
expression for this spread resistance can be written as Fig. 4, where only the active layer (Si) has been shown for the
sake of clarity. It can be seen from Fig. 4 that the conduction
current is collected on the two planes A and B, where the S/D
(3)
silicide makes contact with the HDD regions.
The fraction of conduction current, collected on the plane A
where is the resistivity of the S/D extensions. A constant ac- in Fig. 4, goes directly into the S/D silicide and does not en-
cumulation layer thickness of 1.5 nm has been assumed for counter any resistance other than the contact resistance, which
in (3) [18]. The channel width (W) seen in (2) has been modified is modeled as
to be the width per channel in (3), i.e., for the MuGFETs.
Further, the junction depth in (2) has been modeled by par- (5)
titioning the fin into two identical halves and approximating the
spread resistance as a parallel combination of two identical re- where is the thickness of the S/D silicide and is the
sistances, as seen in Fig. 3(b). Thus, has been replaced by contact resistivity of the S/D silicide. The above equation is
in (3) and a prefactor of 0.5 has been used to account valid only for . For the cases when ,
for the parallel combination of the two resistances. The model i.e., the contact between S/D silicide and Si fin on the plane A
in (3) is valid for MuGFETs with narrow fins only as will be vanishes, the model assumes . Another implicit as-
lesser than for the devices with wide fins. The validity sumption in (5) is that , i.e., the fully
of the model for a particular fin width will then be governed by silicided S/D case is not considered.
DIXIT et al.: ANALYSIS OF THE PARASITIC S/D RESISTANCE 1135

The fraction of conduction current, collected on the plane B


in Fig. 4, first encounters an additional spread resistance .
The expression for this second spread resistance using our ex-
tended spread resistance model, as developed in (3), is given by

(6)

where is the resistivity of the HDD regions. The channel


width (W) in (2) is replaced by the width per channel, i.e.,
, in (6). Again, the device has been di-
vided in two halves and channel thickness in (2) has been
replaced by in (6). The junction depth in (2) has
been replaced by half the width of HDD regions after SEG, i.e.,
, in (6). Here, the width of the HDD regions after Fig. 5. Tilted SEM image of a multifin Omega-FET, taken after SEG process.
SEG of Si is given by Inset shows a transmission electron microscopy image of the cross section along
the line A-A .

(7)
two resistances arising from the division of conduction current
It should be noted here that (6) is valid only when in two fractions, i.e., one collected on the plane A and the other
, as for the values of the assumption of con- collected on the plane B. The resistance is given by
duction current spreading in only two directions is not valid and
it is expected to spread in all three directions.
In order to model the contact resistance for the fraction of (12)
conduction current collected on the plane ‘B’, transmission line
model [20] has been used. According to this model, in case of
the current conduction parallel to a semiconductor-metal inter- where the resistances and are given by
face, a minimum contact length exists before this conduction
current is actually transferred from the semiconductor to the
metal. This length is known as the transfer length (13)
and is given by (14)

(8) The right-hand sides of (13) and (14) can be evaluated using
(5)-(9). Thus, is given by (10), which is the resultant equa-
tion of this analytical model.
Using the transmission line model [21], contact resistance for
the fraction of conduction current collected on the plane B is
given by IV. DEVICE FABRICATION
N-channel Omega-FETs [7] have been fabricated on sil-
(9) icon-on-insulator (SOI) wafers. Starting with an SOI stack of
75-nm Si on 150-nm buried oxide, fins were patterned using
where is the physical length of the HDD regions. 193-nm optical lithography and dry etched. The devices were
To estimate the collective contribution of the planes A and annealed in an H ambient that helps with rounding the fin
B, has been modeled as a series combination of two resis- corners as well as smoothing the fin sidewalls, as is evident
tances and as from the inset in Fig. 5 [22], [23]. A thin, low-temperature
sacrificial oxide was then grown and used as a screen oxide
during -adjust implants. After well-anneal, this oxide was
(10) removed with wet etch. A stack of 2.5-nm nitrided oxide (phys-
ical thickness) and 100-nm-thick poly-Si was deposited. The
where the factor of two accounts for the resistance from both poly-Si gates were predoped and then patterned using 193-nm
the source and drain sides. In (10), is the resistance between optical lithography. Large tilt extension and halo implants were
the gate and S/D spacer edge, given by followed by a 950 C spike anneal. A low-temperature nitride
recessed spacer process was used and an SEG process was done
(11) on S/D regions. A tilted scanning electron microscope (SEM)
image of a multiple fin device, taken after the SEG process, is
where and are given by (3) and (4), respectively. Re- shown in Fig. 5. Low tilt HDD implants were followed by a
sistance in (10) is modeled as a parallel combination of the 1050 C spike anneal and a standard two-step NiSi process.
1136 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 52, NO. 6, JUNE 2005

Fig. 6. Measured gate characteristic curves for single-fin n-channel


Fig. 7. Total resistance as a function of gate voltage for the fabricated devices
Omega-FETs with 61-nm-high fins and 40-nm-long gates. Solid and dotted
with 18- and 120-nm-wide fins and 40-nm-long gates. The parasitic S/D
lines show the measurements on devices with 18- and 120-nm-wide fins,
resistance has been extracted from these curves at a gate voltage of 5 V. Squares
respectively.
and circles show measurements on devices with 18- and 120-nm-wide fins,
respectively. Solid and dotted lines have been fitted to the measurements on
devices with 18- and 120-nm-wide fins, respectively.
V. EXTRACTION OF THE PARASITIC S/D RESISTANCE AND
VALIDATION OF THE ANALYTICAL MODEL
The total resistance of a MOSFET can be written as [18]

(15)

The absolute source currents are plotted as a function


of the gate-to-source voltage in Fig. 6 for single fin
nm devices with 18-, 120-nm-wide fins and
40-nm-long gates. At low and infinitely large ,
diminishes asymptotically and becomes equal to .
The total channel resistance at low has been extracted as a
function of for the – curves shown in Fig. 6 and is
plotted in Fig. 7. A first-order exponential curve fitting on the
experimental data generates the asymptotic part of the curve.
It can be seen from Fig. 7 that for high values,
becomes constant, and is taken as .
In order to validate the analytical model, 3-D device simula-
tions have been done for the reference device with variable fin
Fig. 8. Parasitic S/D resistance as a function of fin width. The analytical model
widths. has been extracted from simulated – curves. has been compared with 3-D device simulations using ISE-DESSIS. Reference
A comparison of these extracted values and those calcu- case values of the device parameters have been used and R has been extracted
lated using the analytical model for the same set of device pa- at V = 50 mV for these n-channel MuGFETs.
rameters is shown in Fig. 8. A constant electron mobility of 37
cm in S/D extensions and HDD regions, corresponding of cm , and an as-measured S/D-HDD length
to the result of device simulation shown in Fig. 4, has been used of 150 nm have been used as inputs to the analytical model.
in the analytical model throughout this work. As can be seen The values of , extracted from experimentally measured
from Fig. 8, our analytical model is in good agreement with 3-D – curves, have been compared with the analytical model
device simulations. in Fig. 9 as a function of fin width. The analytical model data
For further validation of the analytical model using experi- in Fig. 9 shows values calculated using (10). It can be
mental results, a few input parameters have been modified from seen from Fig. 9 that the model agrees very well with the ex-
the previously used values. The S/D spacer bottom width perimental results within the range of validity of the analytical
has been set to a value of 53 nm as measured on the fabricated model, as imposed by the narrow fin width requirement, i.e.,
devices. A contact resistivity of cm has , in (3).
been assumed for the S/D nickel silicide on the heavily doped It should be noticed that even though the model was basi-
Si S/D regions in the fabricated devices. Also, an as-measured cally formulated for a double-gate MuGFET, i.e., (3), it agrees
SEG thickness of 23 nm, an extension doping well with the experimental data measured on fabricated Omega-
DIXIT et al.: ANALYSIS OF THE PARASITIC S/D RESISTANCE 1137

Fig. 9. Parasitic S/D resistance as a function of fin width. The analytical model
has been compared with experimental results. The experimental data shows
statistical mean of measurements over 5 dies and error bars show the standard
deviation of the results from the mean value. Fins wider than 60 nm do not meet
W
the criterion of narrow fins, i.e., < x 2 in S/D extensions, hence the
model (3) is not valid for the devices with fins wider than 60 nm. Fig. 11. Parasitic S/D resistance as a function of SEG thickness. Reference
case values of device parameters have been used as input to the analytical model
and 3-D device simulations.

VI. MINIMIZATION OF THE PARASITIC SOURCE/DRAIN


RESISTANCE
The short channel performance of a MuGFET is strongly de-
pendent on the width of the fin, necessitating the use of narrow
fins for shorter gate length technologies [7]. In order to
evaluate the performance of a MuGFET relative to a planar
MOSFET, effective width of the device must be taken
into consideration. In case of triple-gate devices, could
be written as

(16)

To compare values shown in Figs. 8 and 9 with those


projected by the semiconductor road map [1], could be
normalized as

Fig. 10. Parasitic resistance as a function of fin width. The contribution of m m (17)
spread resistance, present in the S/D extensions, to the total parasitic S/D
resistance is shown for double-gate and triple-gate cases for the models (3) and where can be calculated from (16). The gravity of the
(10).
problem of series resistance in MuGFETs can be clearly seen in
Fig. 8, where due to a reduction in the fin width from 65 to 20
FETs (triple-gate devices). From the perspective of the model nm, is seen to increase by a factor of three.
(10), the only difference expected in the behavior of a triple-gate For the minimization of , the S/D width , as given
device as compared to a double-gate device should be in by (7), must be increased. An SEG of Si on the HDD regions has
given by (3). This justifies the approximation, made in (3), of been reported as a technology solution for this [16]. In order to
modeling this spread resistance as a parallel combination of estimate the potential of SEG of Si on the HDD regions in over-
spread resistors distributed on all the current carrying faces of coming the parasitic S/D resistance, has been plotted as a
the fin. Thus, with an increasing number of current carrying function of thickness of the SEG layer in Fig. 11. To arrive at the
faces (three for the triple-gate device as compared to two for the results shown in Fig. 11, the reference values of device param-
double-gate device), is expected to decrease further. The eters as described in Table I with variable have been used
magnitudes of and for a double- and triple-gate device as input to the model (10) and the 3-D ISE-DESSIS simulations
are compared in Fig. 10 using the models of (3) and (10). As can [17]. It can be seen from Fig. 11 that improves drastically
be seen from Fig. 10, is much smaller than . Therefore, while going from no SEG device, i.e., , to a device
even for the case when (3) is not valid e.g., for a Omega-FET with SEG.
being a triple-gate device, results of the model (10) still show The parasitic S/D resistance along with its two main
good agreement with experimental results of Omega-FET de- components (11) and (12) has been plotted as a func-
vices in Fig. 9. tion of in Fig. 12. A device with a constant fin width
1138 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 52, NO. 6, JUNE 2005

Fig. 12. Parasitic resistance as function of S/D width. Components of R in Fig. 13. Parasitic resistance as function of S/D width. The components of
(10), i.e., R and R are shown. The S/D width is increased by SEG of Si in resistance R are shown. The S/D width is increased by SEG of Si in HDD
HDD regions while keeping the fin width fixed at 20 nm. Over the whole range regions while keeping the fin width fixed at 20 nm. The resistance R is almost
of S/D widths, the parasitic S/D resistance of the device is determined by the same as R and therefore the curve for R is not visible in this semi-log plot.
component R . Over the whole range of S/D widths, component R is determined by the contact
resistance on plane B, i.e., R .

nm has been assumed in Fig. 12 and is in-


creased by increasing in (7). As can be seen from Fig. 12,
the use of SEG reduces only the component of whereas
component stays the same with increasing . This is ex-
pected as SEG is done in the HDD regions and the component
exists between the channel and HDD regions. An upper limit
on could be the spacing between adjacent fins (S2), as can
be seen from Fig. 2. Competitive pattern density demands S2
values which are less than 100 nm [24]. This means that the
maximum usable per fin is less than 120 nm and the max-
imum usable thickness of the Si SEG layer on the HDD regions
is less than 50 nm for a multifin MuGFET with 20-nm-wide fins.
For the device under consideration in Fig. 12, with this upper
limit on and hence on , is expected to be al-
ways dominated by the component .
Component of the parasitic S/D resistance is primarily
due to contact resistances encountered on the planes A and B
(Fig. 4). As can be seen from (12), component is the resul- Fig. 14. Parasitic resistance as a function of S/D silicide contact resistivity.
tant of the parallel combination of contact resistances and The parasitic resistance values shown in figure have been normalized using (17).
. Various components of as seen from (12), (13) and (14) The dash-dotted line without symbols represents ITRS R specification (160
have been plotted as a function of in Fig. 13. The device
0

m) for the low operating power 45 nm technology node. The solid- and
dashed-lines with different symbols represent total R and its R component
assumed in Fig. 13 is the same as that in Fig. 12. It can be visu- respectively. The dotted line without symbols represents R component of R .
alized from (13) and (14) that amongst the two of and , The square symbols represent the reference case device, as defined in Table I,
with T = 50 nm. Circles represent the same device as squares but with
has lesser potential for the minimization of as its area is 50% reduced spacer bottom width, i.e., W = 20 nm. Triangles represent the
actually limited by the narrow fin i.e., is a small number. same device as circles but with 5 times higher doping in S/D extensions, i.e.,
Therefore, the key to parasitic S/D resistance minimization in N 2
= 5 10 cm .
MuGFETs is the reduction of the resistance . It can be seen
from Fig. 13 that SEG on the HDD regions reduces via re- in presence of an SEG layer on the HDD regions, further min-
duction of , which is the dominating part in (14) as imization of could be achieved only by increasing
is expected to be much lower than . It is clear from (5), for the same . In this perspective, S/D silicide contacts
(13) and Fig. 13 that for a fixed silicide thickness, increasing wrapping around the HDD regions can be useful [25], i.e., addi-
increases until it is forced to zero in accordance with tional contact area from the HDD sidewalls, which are normal
validity condition for (5) i.e., for . to the plane B (Fig. 4), can bring further reduction in .
But since comes in parallel with while determining , Technology guidelines for the minimization of in
is seen to decrease with increasing in Fig. 13. Thus, MuGFETs are drawn from the analytical model by plotting
DIXIT et al.: ANALYSIS OF THE PARASITIC S/D RESISTANCE 1139

parasitic resistance as a function of S/D silicide contact re- [4] T. Sekigawa and Y. Hayashi, “Calculated threshold-voltage character-
sistivity in Fig. 14. For the reference device in Fig. 14, a S/D istics of an XMOS transistor having an additional bottom gate,” Solid
State Electron., vol. 27, pp. 827–828, 1984.
silicide contact resistivity of cm is needed to meet the [5] B. Yu, L. Chang, S. Ahmed, H. Wang, S. Bell, C.-Y. Yang, C. Tabery,
International Technology Roadmap for Semiconductors (ITRS) C. Ho, Q. Xiang, T.-J. King, J. Bokor, C. Hu, M.-R. Lin, and D. Kyser,
target for for the low operating power 45 nm technology “FinFET scaling to 10 nm gate length,” in IEDM Tech. Dig., 2002, pp.
251–254.
node [1]. Amongst the components of (10), dominated [6] B. S. Doyle, S. Datta, M. Doczy, S. Hareland, B. Jin, J. Kacalieros, T.
by S/D extension resistance and contact resistance , Linton, A. Murthy, R. Rios, and R. Chau, “High performance fully-de-
is seen to dominate the parasitic S/D resistance for the pleted tri-gate CMOS transistors,” IEEE Electron Device Lett., vol. 24,
no. 4, pp. 263–265, Apr. 2003.
reference device in Fig. 14 for all the values of the contact [7] F.-L. Yang, H.-Y. Chen, F.-C. Chen, C.-C. Huang, C.-Y. Chang, H.-K.
resistivity lower than cm . This implies that the Chiu, C.-C. Lee, C.-C. Chen, H.-T. Huang, C.-J. Chen, H.-J. Tao, Y.-C.
contact resistivity specification could be relaxed by reducing Yoo, M.-S. Liang, and C. Hu, “25 nm CMOS Omega FETs,” in IEDM
Tech. Dig., 2002, pp. 255–258.
the component . This has been illustrated in Fig. 14 by [8] J.-T. Park, J.-P. Colinge, and C. H. Diaz, “Pi-gate SOI MOSFET,” IEEE
reducing the spacer bottom width by 50%. It should be noticed Electron Device Lett., vol. 22, no. 8, pp. 405–406, Aug. 2001.
here that at low S/D-HDD lengths , local variations [9] S. Zhang, R. Han, H. Wang, and M. Chan, “A self-aligned
gate-all-around MOS transistor on single-grain silicon,” Electrochem.
in silicide resistivity may affect S/D silicide contact resistivity Solid-State Lett., vol. 7, no. 4, pp. G59–G61, 2004.
[26]. However, this length dependence of contact resistivity has [10] S. Monfray, T. Skotnicki, B. Tavel, Y. Morand, S. Descombes, A. Talbot,
not been taken into account in our model owing to the micro- D. Dutartre, C. Jenny, P. Mazoyer, R. Palla, F. Leverd, Y. Le Friec, R.
scopic nature of this phenomenon [27]. A further relaxation in Pantel, M. Haond, C. Charbuillet, C. Vizioz, D. Louis, and N. Buffet,
“SON (silicon-on-nothing) P-MOSFETs with totally silicided (CoSi )
contact resistivity specification could be achieved by the use Polysilicon on 5 nm-thick Si-films: the simplest way to integration
of higher S/D extension doping, as can be seen from Fig. 14. of metal gates on thin FD channels,” in IEDM Tech. Dig., 2002, pp.
However, the realization of such heavily doped S/D extensions 263–266.
[11] F.-L. Yang, D.-H. Lee, H.-Y. Chen, C.-Y. Chang, S.-D. Liu, C.-C. Huang,
might require integration of novel doping processes, such as T.-X. Chung, H.-W. Chen, C.-C. Huang, Y.-H. Liu, C.-C. Wu, C.-C.
vapor phase doping, in the CMOS processing. Chen, Y.-T. Chen, Y.-H. Chen, C.-J. Chen, B.-W. Chan, P.-F. Hsu, J.-H.
Shieh, H.-J. Tao, Y.-C. Yeo, Y. Li, J. W. Lee, P. Chen, M.-S. Liang, and
C. Hu, “5 nm-gate nanowire FinFET,” in Symp. VLSI Tech. Dig., 2004,
VII. CONCLUSION pp. 196–197.
[12] J. Kedzierski, E. Nowak, T. Kanarsky, Y. Zhang, D. Boyd, R. Carruthers,
An analytical model for the parasitic S/D resistance in C. Cabral, R. Amos, C. Lavoie, R. Roy, J. Newbury, E. Sullivan, J. Bene-
dict, P. Saunders, K. Wong, D. Canaperi, M. Krishnan, K.-L. Lee, B. A.
MuGFETs has been developed and validated with the help of Rainey, D. Fried, P. Cottrell, H.-S. P. Wong, M. Ieong, and W. Haensch,
3-D device simulations and experimental results. The parasitic “Metal-gate FinFET and fully-depleted SOI devices using total gate sili-
S/D resistance has been seen to increase with decreasing fin cidation,” in IEDM Tech. Dig., 2002, pp. 247–250.
[13] D. A. Antoniadis, “MOSFET scalability limits and ‘new frontier’ de-
width. The contact resistance in the S/D-HDD regions has been vices,” in Symp. VLSI Tech. Dig., 2002, pp. 2–5.
shown to dominate the parasitic S/D resistance behavior of the [14] N. Collaert, A. Dixit, M. Goodwin, K. G. Anil, R. Rooyackers, B. De-
narrow fin devices in the range of interest. The SEG of Si on groote, L. H. A. Leunissen, A. Veloso, R. Jonckheere, K. D. Meyer, M.
Jurczak, and S. Biesemans, “A functional 41-stage ring oscillator using
the S/D-HDD regions is shown to be a potential solution for scaled FinFET devices with 25-nm gate lengths and 10-nm fin widths
minimizing the parasitic S/D resistance of narrow fin devices. applicable for the 45-nm CMOS node,” IEEE Electron Device Lett., vol.
Even in the presence of an SEG Si layer on the HDD-regions, it 25, no. 8, pp. 568–570, Aug. 2004.
was observed that the S/D silicide contact resistivity should be [15] S. D. Kim, C.-M. Park, and J. C. S. Woo, “Advanced model and analysis
for series resistance in sub-100 nm CMOS including poly depletion and
lowered to meet the ITRS target for the low operating power 45 overlap doping gradient effect,” in IEDM Tech. Dig., 2000, pp. 723–726.
nm technology node. However, reducing the S/D spacer bottom [16] J. Kedzierski, M. Ieong, E. Nowak, T. S. Kanarsky, Y. Zhang, R. Roy, D.
width and increasing the S/D extension doping, the requirement Boyd, D. Fried, and H.-S. P. Wong, “Extension and source/drain design
for high-performance FinFET devices,” IEEE Trans. Electron. Devices,
of a very low S/D silicide contact resistivity could be relaxed. vol. 50, no. 4, pp. 952–958, Apr. 2003.
[17] User’s Manual, ISE-TCAD, 2004.
[18] Y. Taur and T. H. Ning, Fundamentals of Modern VLSI Devices. Cam-
ACKNOWLEDGMENT bridge, U.K.: Cambridge Univ. Press, 1998.
[19] G. Baccarani and G. A. Sai-Halasz, “Spreading resistance in submicron
The authors would like to thank the IMEC-Pilot line for MOSFETs,” IEEE Electron Device Lett., vol. 4, no. 2, pp. 27–29, 1983.
the device fabrication and AMSIMEC for the measurement [20] H. H. Berger, “Contact resistance on diffused resistors,” in Proc. ISSCC,
1969, pp. 162–163.
support, the CMOSDR/DIP group for encouraging discussions, [21] S.-D. Kim, C.-M. Park, and J. C. S. Woo, “Advanced model and analysis
V. Vrankx for her secretarial support, and S. Biesemans for the of series resistance for CMOS scaling into nanometer regime—Part I:
overall direction of the IIAP EMERALD. Theoretical derivation,” IEEE Trans. Electron. Devices, vol. 49, no. 3,
pp. 457–466, Mar. 2002.
[22] Y.-K. Choi, L. Chang, P. Ranade, J.-S. Lee, D. Ha, S. Balasubramanian,
REFERENCES A. Agarwal, M. Ameen, T.-J. King, and J. Bokor, “FinFET process re-
finements for improved mobility and gate work function engineering,”
[1] The International Technology Road Map for Semiconductors (ITRS), in IEDM Tech. Dig., 2002, pp. 259–262.
Semiconductor Industry Association, San Jose, CA, 2003. [23] W. Xiong, G. Gebaram, J. Zaman, M. Gostkowski, B. Nguyen, G. Smith,
[2] K. Takeuchi, R. Koh, and T. Mogami, “A study of the threshold voltage D. Lewis, C. R. Cleaveling, R. Wise, S. Yu, M. Pas, T.-J. King, and J. P.
variation for ultra-small bulk and SOI CMOS,” IEEE Trans. Electron. Colinge, “Improvement of FinFET electrical characteristics by hydrogen
Devices, vol. 48, no. 9, pp. 1995–2001, Sep. 2001. annealing,” IEEE Electron Device Lett., vol. 25, no. 8, pp. 541–543, Aug.
[3] V. P. Trivedi and J. G. Fossum, “Scaling fully depleted SOI CMOS,” 2004.
IEEE Trans. Electron. Devices, vol. 50, no. 10, pp. 2095–2103, Oct. [24] K. G. Anil, K. Henson, S. Biesemans, and N. Collaert, “Layout density
2003. analysis of FinFETs,” in Proc. ESSDERC, 2003, pp. 139–142.
1140 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 52, NO. 6, JUNE 2005

[25] H. Kam, L. Chang, and T.-J. King, “Impact of source-drain doping Malgorzata Jurczak received the M.Sc. and Ph.D.
profiles and contact schemes on FinFET performance in the nanoscale degrees in electrical engineering from the Warsaw
regime,” in Proc. IEEE Silicon Nanoelectronics Workshop, 2004, pp. University of Technology (WUT), Warsaw, Poland
9–10. in 1991 and 1997.
[26] C.-N. Liao and K.-C. Chen, “Current crowding effect on thermal char- In 1991, she joined WUT where she worked on
acteristics of Ni/doped-Si contacts,” IEEE Electron Device Lett., vol. 24, modeling of MOS SOI devices. In 1994, she was
no. 10, pp. 637–639, Oct. 2003. with NMRC, Cork, Ireland and in 1997 with Kyung
[27] M. Cannaerts, O. Chamirian, K. Maex, and C. V. Haesendonck, “Map- Hee University, Seoul, Korea. In 1998, she joined
ping nanometer-scale temperature gradients in patterned cobalt-nickel CNET Grenoble, France Telecom, Grenoble, France.
silicide films,” Nanotechnology, vol. 13, pp. 149–152, 2002. She was involved in the development of 0.18- and
0.12-m CMOS process and alternative approaches
Abhisek Dixit was born in Kanpur, India. He for sub-0.1-mm CMOS. In 2000, she joined Inter University Microelectronics
received the M.S. degree in physics, with a spe- Center (IMEC) Leuven, Leuven, Belgium. From 2000 to 2003, she was
cialization in electronics, from the University of the IMEC Coordinator of the JDP program with Philips on device process
Kanpur in 1997 and the M. Tech. degree in materials integration for 90- and 65-nm CMOS. From 2003, she has been the Group
science and engineering from the Indian Institute of Leader of the CMOS Device Implementation Projects (45-nm CMOS, FUSI
Technology (IIT) Bombay, Bombay, India, in 2002. gates, strained-Si, HK+MG integration, FD SOI, and MUGFET). She is also a
His thesis was on device engineering for sub-100-nm Project Manager of IMEC Industrial Affiliation Program (IIAP) on emerging
CMOS technologies. He is currently pursuing his alternative devices (EMERALD). She has participated in several European
Ph.D. degree at the Katholieke Universiteit Leuven projects:, including ACE, ULTRA, MEDEA, and HUNT. She was the Project
(KUL), Belgium. Coordinator of the European IST ARTEMIS project. Currently, she is on the
He joined the Industrial Research and Consultancy Technical Advisory Committee of the IST Integrated project Nano-CMOS. She
Center (IRCC), IIT, as a Junior Research Fellow in 1998, and worked on projects holds 15 French and American patents and has authored and coauthored more
sponsored by the Indian Space Research Organization (ISRO), Indo-French Col- than 60 papers.
laboration for the Promotion of Advanced Research (IFCPAR), and Depart- Dr. Jurczak received the Best Paper Award at the ESSDERC’2000 Con-
ment of Science and Technology (DST). In September 2002, he joined the De- ference and the Paul Rappaport Award for the best paper published in IEEE
partment of Electrical Engineering (ESAT), KUL, as a Predoctoral Student. TRANSACTIONS ON ELECTRON DEVICES in 2000. She has been a member of
He is now with the Inter University Microelectronics Center (IMEC), assigned the scientific committees of IEDM in 2002 and 2003, and the ESSDERC
to the IMEC Industrial Affiliation Program on Emerging Alternative Devices conferences in 2000, 2004, and 2005.
(IIAP-EMERALD) in the CMOS Device Research Department of the Silicon
Process and Device Technology (SPDT) division. His current research interests
include analytical modeling of MuGFET parasitics, and FEOL integration of
sub-45-nm multiple-gate devices and circuits.

Anil Kottantharayil (M’98) received the B. Tech.


degree in electronics and communication engi- Kristin De Meyer (S’73–M’79–SM’00) received the
neering from the Regional Engineering College M.S. and the Ph.D. degrees from the Katholieke Uni-
Calicut, University of Calicut, Kerala, India, in versiteit Leuven (KUL) Leuven, Belgium, in 1974
1993, the M. Tech. degree in electrical engineering and 1979, respectively.
from the Indian Institute of Technology, Bombay, From 1974 to 1975, she was a Fellow of the In-
India, in 1997, and the Dr. Ing. degree (summa cum stitute for Scientific Research in Industry and Agri-
laude) from the Faculty of Electrical Engineering, culture (IWONL), Laboratory for Physics and Elec-
Universität der Bundeswehr, Munich, Germany, in tronics of Semiconductors, KUL, working on CCD
2002. His Ph. D. dissertation was on low-voltage filter design. From October 1975 through September
hot-carrier effects in MOS devices. 1979, she was a Research Assistant of the National
Since November 2001, he has been with the Inter University Microelectronics Fund for Scientific Research (NFWO) at the same
Center (IMEC), Leuven, Belgium. His interests are in the areas of MOS device laboratory on the design, development, and characterization of two-phase buried
physics, device integration, characterization, simulation, and modeling. He has channel CCDs. From November 1979 to November 1980, she was granted an
authored or coauthored more than 25 papers in peer-reviewed international jour- IBM World Trade Post-Doctoral Fellowship and was with the IBM Thomas
nals and conference proceedings. J. Watson Research Center, Yorktown Heights, NY, researching the develop-
Dr. Kottantharayil was a recipient of the “Youth and Knowledge” fellowship ment of EAROMs using Si-rich oxides. At the end of 1980, she returned to
from Siemens AG during his stay in Germany. KUL as a Senior Research Assistant of the NFWO, and in October 1982, as a
Research Associate of the NFWO. Her activities moved from KUL to the In-
teruniversity Microelectronics Center (IMEC), Leuven, where she initially re-
Nadine Collaert was born in Dendermonde, Belgium, in 1972. She received the mained Research Associate of the NFWO. In October 1989, she became a reg-
M.S. and Ph.D. degrees in electrical engineering, both from the Katholieke Uni- ular employee of IMEC. From 1985 through 1992, she was Head of the Process
versiteit Leuven, Leuven, Belgium, in 1995 and 2000, respectively. Her Ph.D. and Device Modeling and Simulation Group, working on process optimization
dissertation was on modeling and characterization of a novel vertical Si–SiGe techniques, parameter extraction, analytical and numerical device modeling,
heterojunction MOSFET. and computational physics. Also, Statistical Process Control Activities (SPC)
Currently, she is a Researcher in the CMOS device research department at were monitored through her group. In January 1993, she was put in charge
Inter University Microelectronics Center (IMEC). Her research interests include of deep-submicrometer MOS technology and exploratory field-effect devices.
the theory, design, and technology of multiple-gate devices. Since October 1986, she has also been a part-time Professor at KUL. She is
the Coordinator for IMEC in the ESPRIT 962 EVEREST, STORM, NOVA,
Mike Goodwin (M’98) received the Ph.D. degree from the University of North ULTRA, VAHMOS2000, FASEM, and ULIS projects. She was also involved
Texas, Dayton, in 1982. in the ESPRIT ACCESS Program, ADEQUAT, and ACE. She also managed
He joined Texas Instruments (TI), Inc., Austin, TX, in 1982. For the next 12 the HCM-SUSTAIN network on deep-submicrometer silicon technology with
years, he developed advanced infrared detector devices under various grants and 21 partners from academia and research institutes. She authored or coauthored
programs sponsored by the U.S. Government. He then transferred to the Silicon over 200 publications and organized summer courses and the SISPAD’98 Con-
Technology Development Department, TI, and as part of a team developed high- ference, and the ULIS 2004 workshop.
performance transistors for the 180- and 90-nm nodes. For the last two years, Dr. De Meyer is a member of the ITRS working Group on Process Integration,
he has worked at Inter University Microelectronics Center (IMEC) Leuven, Devices and Structures (PIDS). She is currently Editor of the IEEE ELECTRON
Leuven, Belgium, as a TI assignee developing multigate transistor structures. DEVICE LETTERS.

S-ar putea să vă placă și