Sunteți pe pagina 1din 98

Master of Science in Physics Thesis

Waveguide Photodiodes
– with high mode overlap and sensitivity

Henrik Ingerslev

April 12, 2004


ii

The front page shows an electrical eye diagram for a 10QW SOA-PD device with a 320µm
long SOA. The input power is −12dBm, it is modulated at 10Gb/s, with a pseudo random
bit sequence of 107 − 1, 85mA is applied to the SOA, and −1volt to the PD.
This measurement shows an interesting effect of the history of the bits (section 3.4.2).
Abstract

Characterization and optimization of two different types of photodiodes, for use


in optical fiber communication systems, are presented. The first type is a PIN
waveguide photodiode, with a bulk active layer of InGaAs. The second type is a
PIN waveguide photodiode, with an integrated semiconductor optical amplifier,
and it has an active layer of InGaAsP quantum wells.
For the first photodiode a total of 22 different designs are characterized, and in
the light of this characterization the best design is chosen. The mode overlap is
furthermore optimized from among different optical fibers, and a maximum of
∼ 95% is achieved.
For the second photodiode three different heterostructures, with 3, 8, and 10
quantum wells, each with several different lengths of the semiconductor optical
amplifier, is characterized. A maximum sensitivity of ∼ −24dBm is achieved
by optimizing the length of the semiconductor optical amplifier.

iii
iv
Preface

The work presented in this master of science (M.Sc.) in physics thesis was car-
ried out at the Research Center COM, Technical University of Denmark, in the
optoelectronic group, and at GiGA, an Intel company.
The work was supervised by Professor Jørn M. Hvam, head of the optoelectronic
group at COM, Assistant Professor Kresten Yvind, COM, and Poul Erik Lin-
delof, vice chairman of the Nanoscience Center at Copenhagen University. I am
very grateful for the support and guidance given by Jørn M. Hvam, and Kresten
Yvind, and I would also like to thank Ph.D. Jesper Hanberg, GiGA-Intel, for
growing the first type of photodiode.
This thesis presents characterization and optimization of two types of waveguide
photodiodes, for use in an optical fiber communications system.
The project began in January 2003, and in the spring term I had a course on
the physics of diode lasers (5 ECTS), and as part of my master education I also
made a presentation on photonic crystals (5 ECTS) in the fall term.

COM, DTU
April 2004
Henrik Ingerslev
hij@com.dtu.dk
hij@fys.ku.dk

v
vi
Contents

1 Introduction 1
1.1 Optical fiber communication . . . . . . . . . . . . . . . . . . . . . 1
1.2 Project description . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Structure of thesis . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Bulk Photodiode 5
2.1 Device design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 PIN Junction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2.1 Bandstructure . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2.2 I-V curves . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Light absorption . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4 Waveguide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.5 Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.5.1 Bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5.2 Sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.6.1 Comparison of devices . . . . . . . . . . . . . . . . . . . . 43
2.6.2 Improvements . . . . . . . . . . . . . . . . . . . . . . . . . 46

3 Photodiode with integrated SOA 49


3.1 Device design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.1.1 Quantum wells . . . . . . . . . . . . . . . . . . . . . . . . 51
3.2 QW Photodiode . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2.1 PIN Junction . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2.2 Light absorption . . . . . . . . . . . . . . . . . . . . . . . 55
3.2.3 Waveguide . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.3 QW SOA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.3.1 Optical gain in semiconductor material . . . . . . . . . . 58
3.3.2 Amplifying cavity . . . . . . . . . . . . . . . . . . . . . . 63
3.3.3 Total amplification . . . . . . . . . . . . . . . . . . . . . . 65
3.4 SOA-Photodiode . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.4.1 Bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.4.2 Sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.5 Summary and discussion . . . . . . . . . . . . . . . . . . . . . . . 77

vii
viii CONTENTS

4 Conclusion 81

Abbreviations 83

Bibliography 85
List of Figures

1.1 Block diagram if a receiver. . . . . . . . . . . . . . . . . . . . . . 2

2.1 Schematic figure of the PIN WGPD . . . . . . . . . . . . . . . . 6


2.2 Density of occupied states for an intrinsic and doped semiconductor 8
2.3 Depletion approximation for the electrical potential . . . . . . . . 10
2.4 Bandstructure for PIN5 . . . . . . . . . . . . . . . . . . . . . . . 12
2.5 I-V curves for all devices (no incident light) . . . . . . . . . . . . 13
2.6 Absorption and emission processes . . . . . . . . . . . . . . . . . 16
2.7 Calculated absorption coefficient . . . . . . . . . . . . . . . . . . 17
2.8 Quantum efficiency versus incident wavelength for PIN5 . . . . . 18
2.9 I-V curves for all devices with optical input powers . . . . . . . . 21
2.10 Photocurrent for PIN8, at two different wavelengths . . . . . . . 24
2.11 Schematic of the refractive index in the PD . . . . . . . . . . . . 26
2.12 Effective index method calculation of the mode in PIN6 . . . . . 27
2.13 Farfields of three different optical fibers . . . . . . . . . . . . . . 30
2.14 Schematic of the electrical setup for dynamical measurements . . 32
2.15 Model used to calculate the bandwidth of the PDs . . . . . . . . 33
2.16 Block diagram of the setup for measuring the bandwidth. . . . . 36
2.17 Bandwidth: Responsivity versus modulation frequency . . . . . . 37
2.18 Block diagram of the setup for the sensitivity measurements . . . 40
2.19 Optical and electrical eye diagrams . . . . . . . . . . . . . . . . . 40
2.20 Sensitivity: BER versus optical input power . . . . . . . . . . . . 42
2.21 New diagram of external components . . . . . . . . . . . . . . . . 46

3.1 Schematic figure of the SOA-PD devices . . . . . . . . . . . . . . 51


3.2 Structure of bandgap through one QW from the 10QW device . 52
3.3 I-V curves for all QW PDs, with different optical input power . 54
3.4 Bandstructure of the 8QW device . . . . . . . . . . . . . . . . . . 56
3.5 Photocurrent versus incident optical wavelength . . . . . . . . . . 57
3.6 Schematic illustration of the integrated SOA-PD device . . . . . 58
3.7 Calculated gain spectrum . . . . . . . . . . . . . . . . . . . . . . 59
3.8 Optical output spectrum from an AR coated 10QW SOA . . . . 61
3.9 Photocurrent from the PD versus incident wavelength, and SOA-
current, for three different SOA-PDs . . . . . . . . . . . . . . . . 62

ix
x LIST OF FIGURES

3.10 I-V curves of the PD at different SOA-currents . . . . . . . . . . 64


3.11 Photocurrent from PD versus SOA-current . . . . . . . . . . . . 66
3.12 Amplification of seven different SOAs versus SOA-current, and
incident optical power . . . . . . . . . . . . . . . . . . . . . . . . 68
3.13 Bandwidth: Responsivity versus modulation frequency . . . . . . 72
3.14 Sensitivity: BER versus optical input power . . . . . . . . . . . . 76
3.15 Sensitivity versus length of SOA . . . . . . . . . . . . . . . . . . 77
List of Tables

2.1 Growth data for all bulk WGPDs. . . . . . . . . . . . . . . . . . 7


2.2 Breakdown voltages, and breakdown fields . . . . . . . . . . . . . 13
2.3 Responsivity of all bulk PDs . . . . . . . . . . . . . . . . . . . . . 19
2.4 Key parameters of the modes in the bulk PD waveguides . . . . . 29
2.5 Measured and calculated bandwidths of the bulk PDs . . . . . . 34
2.6 Calculated and measured sensitivity for five bulk PDs . . . . . . 43
2.7 Data sheet for all the bulk PDs . . . . . . . . . . . . . . . . . . . 44
2.8 Key parameters for six PDs including PIN5 from GiGA . . . . . 45

3.1 Growth data for the three different SOA-PDs . . . . . . . . . . . 50


3.2 Calculated parameters for the QW material . . . . . . . . . . . . 52
3.3 Key parameters of the modes in the SOA-PD waveguides . . . . 57
3.4 Measured and calculated bandwidths of the SOA-PDs . . . . . . 74
3.5 Data sheet for the SOA-PDs . . . . . . . . . . . . . . . . . . . . . 78

xi
xii LIST OF TABLES
Chapter 1

Introduction

1.1 Optical fiber communication


An optical fiber communication system consists basically of a transmitter, an
optical fiber, and a receiver. The transmitter converts the data from electrical
form to optical form, and couples the optical data into the optical fiber. The
optical fiber then transfers the data over long distances to the receiver, that
converts the data from optical form back to electrical form.
There are two modulations formats for the optical data signal, either return to
zero (RZ), or non return to zero (NRZ). As the name implies; in the RZ format
the optical signal goes to zero between each bit (binary digit) even if there are
multiple bit1’s after each other, whereas in the NRZ format the optical signal
will stay at the bit1 level between multiple bit1’s.
The transmitter part consists of a laser (light amplification by stimulated emis-
sion of radiation), and for bit rates up to 10Gb/s direct modulation of the laser
is commonly used. But, if the bit rate has to be higher for example 40Gb/s,
external modulation is used, with a Mach-Zehnder modulator, or an electro ab-
sorption modulator (EAM) as the external modulator.
An optical fiber are today made with a core and cladding layer of silica glass,
because this material has a very low loss of about 0.2dB/km for light at a wave-
length of 1550nm [5]. The core and cladding layer are doped differently so that
the core have the highest refractive index, and the guiding mechanism in the
fiber is total internal reflection. There is another important quantity to consider
for optical fibers, besides loss, namely dispersion. Dispersion is caused by effects
like 1) the refractive index of silica depends on the optical wavelength, and 2)
core and the cladding layer has different refractive index. And by engineering
the optical fiber in a certain way the dispersion minimum can be made to coin-
cide with the loss minimum, namely at 1550nm. An optical mode at 1550nm
is therefore widely used in optical fiber communication today. But even thou
loss and dispersion is being minimized in optical fibers today, regeneration and
amplification of the optical signal along the optical fiber will still be needed for

1
2 CHAPTER 1. INTRODUCTION

Figure 1.1: Block diagram if a receiver.

very long transfer distances (> 100km) [5].


Figure 1.1 shows a block diagram of a receiver, and the components of such a
receiver can be arranged in three sections; a front end, a linear channel, and a
data recovery section.
The front end consists of a photodiode (PD) possibly with an integrated semi-
conductor optical amplifier (SOA) (see section 3.4), and an electrical preampli-
fier that amplifies the electrical signal for further processing.
The linear channel consists of an amplifier and a low pass filter. The amplifier
gain is controlled automatically to fix the average output voltage to a specific
level, irrespective of the input voltage from the front end. The purpose of the
low pass filter is to cut off high frequency noise.
The data recovery section consists of a decision circuit, and a clock recovery cir-
cuit. The latter finds the bit rate of the received signal, which gives information
about the time period of the bit slot. The decision circuit compares the output
from the linear channel to a threshold value at sampling times determined by
the clock recovery circuit, and decides whether the signal corresponds to a bit1
or a bit0.
To increase the total bit rate in an optical fiber communication network, two
methods are developed; optical time division multiplexing (OTDM), and wave-
length division multiplexing (WDM). In WDM systems several different base
rate channels, each at a different wavelength, are coupled into one fiber. The
OTDM systems works by delaying each of the base rate channels (each at the
same wavelength), by a certain amount of time with respect to the next channel,
and then couple all of the channels into one fiber.

This thesis will be dealing with the PD and the SOA from the receiver. There
are a number of ways to make a PD that can absorb photons at a wavelength
of 1550nm, both Ge and InP/InGaAs materials can be used. The junction can
be either PN or PIN (see section 2.2), and the light can be incident either onto
the surface (called a surface PD), or into a waveguide1 from the facet (called a
waveguide photodiode (WGPD)).
The InP/InGaAs material has the advantage over Ge that InGaAs has a di-
1 The waveguide will be introduced in section 2.4
1.2. PROJECT DESCRIPTION 3

rect bandgap, whereas Ge has an indirect bandgap, so the absorption i much


higher for InGaAs then for Ge. Another advantage is that InGaAs tends to have
smaller dark current, which is due to a smaller intrinsic carrier concentration
(see equation 2.4). The PIN junction has the advantage over a PN junction
that we can control the size of the depletion layer, which is used to optimize key
time parameters for the PD (see section 2.5.1). The surface PD and the WGPD
each have their advantages; the surface PD has a large area for the incident
light to hit, whereas for the WGPD the incident light has to be aligned very
carefully to the waveguide, which makes it hard to fabricate. But, the large
area of the surface PD also results in large capacitance, which in turn gives a
large RC time and consequently a low bandwidth (see section 2.5.1). So, for
high bit rates at 1550nm the PIN WGPD made from InP/InGaAs are preferable.

1.2 Project description


The work presented in this thesis focuses on the PD part of a receiver. Specifi-
cally on two different types of PDs:
1) A PIN WGPD made with a bulk absorbing layer of InGaAs, which is
lattice matched to an InP substrate.
2) A PIN WGPD with an integrated SOA, made with a number of InGaAsP
quantum wells as the active layer, which are strained to an InP substrate.
At GiGA-Intel they have designed a receiver unit, and the bulk PD is designed
to be the PD part of this receiver.
The PD with integrated SOA is made from materials that are not specifically
designed for this purpose. The materials are made by COM, for mode-locked
lasers [15].
The fabrication part of this project is to cleave the ready-made wafers into dies,
for the second type of PD this involves cleaving the SOAs with different lengths.
Then mount the dies onto a substrate, for either DC or AC measurements, and
bond gold wires from the die to the substrate.

The goal of the project is to characterize and optimize the two different
types of PDs, for use in optical fiber communication systems, and to find ways
to improve them.
Intel made the following requirements on the PD before they began developing
it (at an incident Wavelength of 1550nm):
• Responsivity ≥ 1A/W (see section 2.3)
• Bandwidth ≥ 10Gb/s (see section 2.5.1)
• Dark current ≤ 1nA (see section 2.2)
• Bias voltage = 3 − 5volt (see section 2.2)
4 CHAPTER 1. INTRODUCTION

To verify if these requirements has been met, the characterization of the bulk
PD involves measurements like; current voltage graphs with different optical
input powers, photocurrent versus incident wavelength, breakdown voltages,
dark currents, bandwidths, and sensitivities.
The optimization part for the bulk PD consists of optimizing the responsivity2 ,
be using many different optical fibers to couple the light into the PD.
For the PD with integrated SOA, the characterization involves the same things
as for the bulk PD, but also measurements on the SOA that shows the gain
spectrum, total power amplification at 1550nm, and threshold level (which is
where it starts to lase, see section 3.3).
The optimization part of the PD with integrated SOA consists, like the bulk
PD, of finding an optimal optical fiber, but more importantly it also consists of
optimizing the length of the SOA, so that the integrated device gives a maximum
sensitivity3 .
I have developed several different LabVIEW4 programs from scratch, for making
all these measurements, which includes programs to make surface plots (see
figure 3.9 and 3.12).

1.3 Structure of thesis


After this introduction chapter, the thesis is divided into two main chapters;
chapter 2 is devoted to the bulk PD, and chapter 3 is devoted to the PD with
integrated SOA.
In chapter 4 we summarizes what has been made, emphasizes the advantages
and disadvantages of the two types of PDs, and concludes wether the goals from
the project description has been met.

2 Theresponsivity will be defined in equation 2.11


3 Thesensitivity will be defines in section 3.4.2
4 LabVIEW is a graphical development software for test, measurement, and control, pro-

vided by National Instruments Corporation


Chapter 2

Bulk Photodiode

In this chapter we will be dealing with the first PD, namely the bulk PIN WGPD
made from InGaAs/InP.
The chapter starts with a description of a number of different heterostructure
materials that the devices are made from. The PIN junction is first charac-
terized electrically in section 2.2, and then with absorption of incident light in
section 2.3. The waveguide in the device, and mode overlap, is considered in
section 2.4. Then we turn to dynamical measurements of bandwidth and sensi-
tivity in section 2.5. And the chapter finishes with a comparison of the different
designs, and some ideas on how to improve it.

2.1 Device design


A total of twelve different designs of the bulk PIN WGPD have been made.
They are named PIN1 through PIN10, PIN12 and PIN13, PIN11 does not ex-
ist!, and measurements are performed on all except PIN1 because it were the
first design and did not work well.
The overall design of all these devices are similar, and consist from top to bot-
tom of a top contact layer, which is designed to give good electrical contact
to the metal leads. Down through the device we then have a layer of positive
doped (P-doped) InP, then a smaller layer of undoped InP, called a standoff
layer. After that we have three small undoped layers of InGaAsP with decreas-
ing bandgap, which are called grading layers. Then we reaches the layer that
absorbs the incoming light, it is made of In0.53 Ga0.47 As and is called the active
layer. Then we again have three layers of InGaAsP, also called grading layers,
now with increasing bandgap. After that we reaches the last layer, which is a
negatively doped (N-doped) InP layer.
All these layers are lattice matched to InP, and grown in the 100 direction on a
N-doped InP Wafer. The structure can be seen schematic in figure 2.1, where
the length of the device is cleaved to about 200µm. The P-doping is made with

5
6 CHAPTER 2. BULK PHOTODIODE

Figure 2.1: Schematic figure of the PIN WGPD with light incident onto the
facet from an optical fiber. To the left is a zoom-in around the active layer.

with Zn atoms, and the N-doping is made with Si atoms.


A ridge is made by lithography, and on both sides of the ridge an insulating ma-
terial called BCB (benzocyclobutene), is filled. The ridge is made in two designs;
one with a width of 1.5µm and another with a width of 2.0µm. The differences
between the eleven designs are the thickness of the active layer, standoff layer,
and grading layers, height of ridge, and the doping concentration of the P-side.
A summary of all these parameters for the eleven devices, are listed in table 2.1.

2.2 PIN Junction


A junction with a P-doped region followed by a N-doped region is called a P-N
junction, and by introducing an intrinsic (undoped) region between the P and
N side we have a PIN junction. If the materials in the junction are the same,
the junction is called a homojunction, and if there are involved different ma-
terials in the junction, it is called a heterojunction. Since all devices in this
project have P-doped, N-doped, and intrinsic regions, and made from different
InGaAsP alloys, these junctions are all PIN heterojunctions.
2.2. PIN JUNCTION

PIN2 PIN3 PIN4 PIN5 PIN6 PIN7 PIN8 PIN9 PIN10 PIN12 PIN13
P-doping [1017 cm−3 ] 4 4 4 4 4 4 4 4 4 11 28
P-side [nm] 2465 2465 2465 2315 2315 2315 1515 2315 3115 2315 2315
Standoff layer [nm] 150 0 0 150 150 150 950 950 150 150 150
3×Grading layer [nm] 27 27 10 10 10 10 10 10 10 10 10
Active layer [nm] 400 100 50 50 80 30 50 50 50 50 50
3×Grading layer [nm] 27 27 10 10 10 10 10 10 10 10 10
N-side [nm] 1000 (the same for all)
N-doping [1017 cm−3 ] 19 (the same for all)
Substrate [µm] 150 (the same for all)
Substrate doping [1017 cm−3 ] ∼ 70 (the same for all)

Table 2.1: Growth data for all bulk WGPDs.


7
8 CHAPTER 2. BULK PHOTODIODE

Figure 2.2: (a) Fermi’s distribution function for an intrinsic semiconductor at


300K, (b) density of states (3D), (c) density of occupied states for the intrinsic
semiconductor, and (d) shows the density of occupied stated for an n-doped
semiconductor

2.2.1 Bandstructure
Before vi collect the P, I and N materials to a junction, let’s first look at each
material separately. The energy distribution for electrons and holes (absence
of an electron in an otherwise filled band) are governed by Fermi’s distribution
function (see figure 2.2(a)), and the density of states can be calculated on the
basis of our knowledge of the crystal (see figure 2.2(b)). These two functions
multiplied give us the density of occupied states (see figure 2.2(c)) [1]. In intrin-
sic materials there are equal densities of electrons and holes, which is obtained
through the Fermi energy from Fermi’s distribution function. That is, the Fermi
energy adjusts the ratio between electrons and holes.
If we introduce doping, lets say donor atoms in the crystal, we will have more
electrons and fewer holes (though the product of electron and hole densities are
still the same), this will make the Fermi energy shift upwards as can be seen in
figure 2.2(d).
Now lets look at what happens when we combine P, I, and N materials in the
PIN junction. There will be a large carrier concentration gradient across the
junction, which will make the free electrons from the N-side diffuse across the
intrinsic region into the P-side, and free holes from the P-side will diffuse across
the intrinsic region into the N-side. So, this process will lower the Fermi energy
of the N-side, and raise the Fermi energy of the P-side. As this process con-
tinues, the diffused carriers will cause an electric field to be built up, called a
built-in electric field, and this electric field will counteract the diffusion.
Current due to a carrier concentration gradient is called diffusion current, and
current due to an electric field is called drift current. After some time a steady
state situation will be reached where the diffusion current, and the drift current
2.2. PIN JUNCTION 9

are in equilibrium. In this steady state the Fermi energy of the two materials
will have reached the same level.
In the steady state situation the electrons that have diffused from the N-side to
the P-side will be trapped at acceptor atoms, and holes that have diffused from
the P-side to the N-side will be trapped at donor atoms. (An electron/hole that
are not trapped at an acceptor/donor atom will be free to move, and will be
drifted back to the N-side/P-side by the built-in electric field).
There are two important quantities connected with the junction; the depletion
region with width wdep , and the built in potential Vbi . The depletion region is
the entire I region, and the region of the the P and N materials where there are
no more free carriers. The built-in potential across the depletion region, which
is caused by the diffused carriers, is equal to the difference in Fermi energy of
the P-side and N-side, before they were combined.
Because there are no free carriers in the depletion region it is highly insulating,
and by applying an external potential (bias voltage V0 ) to the junction, the po-
tential will lay mainly across the depletion region. The total potential across the
junction will now be a combination of built-in potential, and external applied
potential (Vbi + V0 ). The Fermi energy will now be different on either side of
the depletion region, with a difference equal to the applied bias. The two Fermi
energies are called quasi Fermi levels (see figure 2.4).

We here define the bandstructure as the position of the valence, and con-
duction band through the junction. To plot a bandstructure we need to find
the electrical potential through the junction (from poisson’s equation)[6]:
+
∇2 V = ρ/² , where ρ = e(p − n + ND − NA− ) (2.1)

where V is the electric potential, ρ the total charge density, ² the dielectric
+
constant, p and n are densities of free electrons and holes, and ND and NA− are
doping concentrations of the ionized donors and the acceptors.
Since we know the total charge density (due to diffusion of carriers) is positive
on the N-side, and negative on the P-side, we will approximate the electric field
(E = −∇V ) to a scalar one dimensional problem, where the electric field is
perpendicular to the junction interface (lateral x-direction).
The electrical potential, and the electric field has to satisfy the boundary con-
ditions [6]:
∆V = 0 ²∆E = 0 (2.2)
To find the electrical potential V we need the total charge density ρ, and based
on the discussion above we can approximate it with a step function which is
−|NA− | at the P-side, +|ND +
| in the N-side, and zero in the I region and else-
where, see figure 2.3(a). This is called the depletion approximation[3]. The
width of the depletion region must subsequently be adjusted to match the total
potential (Vbi + V0 ).
The solution to poisson’s equation, and the resulting electric field are shown in
figure 2.3 (b) and (c).
10 CHAPTER 2. BULK PHOTODIODE

Figure 2.3: In (a) the assumed total charge density across the depletion region,
in (b) the magnitude of the electric field, where the direction is vertical up
through the device, and in (c) the electrical potential, is shown. A reverse bias
V0 = −5volt is applied to the junction.
2.2. PIN JUNCTION 11

From the electrical potential in figure 2.3(c) it is now fairly easy to construct
the bandstructure through the device. The bandstructure represents the energy
of the electrons in the valence band, and in the conduction band, so we have to
multiply the potential (in figure 2.3(c)) with −|e|, and at the interface between
two materials we have a discontinuity in energy according to the bandgap dif-
ference. In figure 2.4 the bandstructure for PIN5 at minus one, zero, and plus
one volt bias is shown. The two quasi Fermi levels are the Fermi energy of the
P- and N-side.

2.2.2 I-V curves


In figure 2.5 measured currents versus applied voltage (I-V curves) for all de-
vices are shown, except PIN9, and PIN10 because in DC measurements they
are like PIN8 and PIN5 respectively.
These I-V curves can be divided into three important regimes, starting from
the right, we see a sudden increase in current at small forward bias (positive
to P-side and negative to N-side) around 0.6volt to 0.8volt. Then we have a
section with almost no current, called dark current in a PD. This regime ends at
some reverse bias (positive to N-side and negative to P-side), where the current
again suddenly increases, this is called break down of the diode, and the reverse
voltage at which it happens is called the breakdown voltage.
Breakdown can be attributed to two mechanisms, called Zener breakdown, and
avalanche breakdown. Zener breakdown is caused by tunnelling of the carriers
through the junction, and happens for heavily doped > 1017 cm−3 PN (not PIN)
junctions where the conduction and valence band at reverse bias can get very
close.
For PIN junctions, or lightly doped PN junctions tunnelling is negligible because
the depletion region is much wider. In such devices the breakdown is caused by
infinite avalanche multiplication, called avalanche breakdown. Avalanche mul-
tiplication will occur in junctions with a high electric field and a large depletion
layer, where the carriers can gain enough kinetic energy to excite new carriers
by impact. These new carriers will also gain kinetic energy and by impact excite
new carriers and so forth. (This process can, if it can be controlled, be used to
make gain in a PD, such a PD is called an avalanche photodiode (APD)). But
if the electric field gets high enough, the avalanche multiplication gets infinite,
and the junctions breaks down[4].
All devices in this project are PIN junctions, and based on the discussion above
the breakdown mechanism must be avalanche breakdown. In table 2.2 the mea-
sured breakdown voltages, and corresponding calculated maximum electric fields
(called the break down fields Ebd ) are listed. This maximum electric field is in
the intrinsic region (see figure 2.3(b)), and the width of the intrinsic regions are
therefore also listed in table 2.2.
By looking at table 2.2 we see that for all devices, except PIN12 and 13, the
breakdown field is inverse proportional to the width of the intrinsic region, and
this corresponds well with the breakdown being caused by avalanche multipli-
12 CHAPTER 2. BULK PHOTODIODE

Figure 2.4: Structure of the bandgap through device PIN5 at (a) -1volt bias,
(b) 0 volt bias, and (c) +1 volt bias. Based on the depletion approximation [3].
2.2. PIN JUNCTION 13

Figure 2.5: I-V curves for all devices. The inset is a close up of the forward
current, where the solid black line is equation 2.6.

PIN2 PIN3 PIN4 PIN5 PIN6


Vbd [volt] 25 13 8 12.5 14
wi [nm] 712 262 110 260 290
Ebd [kV /cm] 355 474 586 462 466
PIN7 PIN8 PIN12 PIN13
Vbd [volt] 13.5 35 11 6
wi [nm] 240 1060 260 260
Ebd [kV /cm] 522 334 375 275

Table 2.2: Lists the breakdown voltages (Vbd ), breakdown electric fields (Ebd ),
and the width of the intrinsic regions.
14 CHAPTER 2. BULK PHOTODIODE

cation. Because the number of impact ionization must depend on both the
magnitude of the electric field, and also on the length at which the ionizations
can occur.
The reason that PIN12 and 13 do not follow this pattern is probably because
they are P-doped so highly, that the doping ions begins to diffuse into the in-
trinsic layer[14], causing the intrinsic region to decrease and the electric field to
increase.

At reverse bias, from the breakdown voltage up to zero volt we have the
dark current regime.
When we set reverse bias to the junction, we increase the potential across the
depletion region (which also increases the width of the depletion region), this will
increase the drift current in comparison to the diffusion current. We therefore
have a net drift current of holes from the N-side to the P-side, and a net current
of electrons from the P-side to the N-side. This gives givers a very small current,
called the saturation current, and according [3] it is given by:
Ãs r !
2 Dp 1 Dn 1
Is = e A niInP + + (2.3)
τp ND τn NA−

where A is the current cross section area, niInP is the intrinsic carrier concen-
tration of InP, τp is the life time of a hole in N-side, and τn is the life time of
an electron in P-side.
Using A = 2µm × 200µm (ridge width times length of device), niInP = 1.3 ·
107 cm−3 [11], τp = 1ns, τn = 5ns [9], Dn = 100cm2 /s, and Dp = 3cm2 /s[4], a
current of about ∼ 10−23 A is obtained.
A larger term contributing to the dark current is thermal generation of carriers.
Each time an electron hole pair (EHP) is generated thermally in the depletion
region, the electric field will separate them, and pull the hole to the P-side, and
the electron the to N-side. The rate of generation will depend on the intrinsic
carrier concentration ni , and the generation rate per volume is approximated
by ni /τ , where τ is the life time of the carriers in intrinsic material [3]:
e VInGaAs niInGaAs
Ithermal = (2.4)
τInGaAs
where VInGaAs is the volume of the active layer, and τInGaAs is the life time
of carriers in intrinsic In0.53 Ga0.47 As. The grading layers and standoff layer
are also part of the intrinsic region, so there should also be a similar term for
those layers. But since they have lower intrinsic carrier concentrations, they
give negligible thermal currents compared to the active region. The intrinsic
carrier concentration of InGaAs is niInGaAs = 1012 cm−3 [8], and the lifetime is
τInGaAs = 15ns [11].
For PIN2 this gives about 2nA, and because the rest of the devices has about
the same volume of the active layer they all give about 0.2nA. So according
to these calculations the dark current Idark = Is + Ithermal stems from thermal
generation of carriers.
2.3. LIGHT ABSORPTION 15

Measurements of the dark current is hard because it is so low, and because the
device has to be in complete darkness, but at GiGA they have measurement
equipment for doing this. We measured the dark current on several different
PIN5 devices, and they all gave a dark current about ∼ 0.5nA. Since this re-
sult matches fairly well with the calculated value above, I think it is reasonably
to assume that the dark current stems from thermal generated carriers in the
active layer, and has a value Idark < 1nA for all devices (except PIN2).

The last section of the I-V curve is at forward bias. When we set forward
bias on the junction we reduce the potential across the depletion region (which
in turn also will reduce the width of the depletion region), this will reduce the
drift current in comparison the the diffusion current. Therefore we have a net
diffusion current of holes from the P-side to the N-side, and electron from the
N-side the P-side, this is called minority carrier injection. This diffusion current
is according to [3] given by:
³ ´
I(V ) = Is ee V /kT − 1 (2.5)

where V is forward applied bias, k Boltzmann’s constant and T is the temper-


ature. This is called the ideal diode characteristic [3].
Because of the the small Is , this gives only about 1nA at 0.8volt, which is much
smaller then what we measure in figure 2.5.
Another term attributing to the forward current is recombination current. When
the applied forward bias is equal to the bandgap of the active layer, the minority
carrier injection starts to fill the conduction band, and valence band of the active
layer with electrons, and holes respectively. This will lead to recombinations in
the active layer, which according to [3] is given by:
e Vactive niInGaAs e V /2kT
I(V ) = e (2.6)
2τr
where τr is the recombination lifetime of the electrostatic doped InGaAs. This
lifetime should intuitive be shorter then for intrinsic InGaAs (τInGaAs ) from
above, and it is found to be τr ∼ 1ns for a doping concentration of 1016 cm−3 [10].
This equation, gives a much larger current then equation 2.5, and is plotted in
the insert of figure 2.5, where it fits quit well to the measured I-V curves.

To summarize the electrical PIN junction; breakdown occurs because of in-


finite avalanche multiplication, dark current is due to thermal generation of
carriers in the active layer, and forward current is a recombination current i the
active layer.

2.3 Light absorption


From figure 2.4(a) it is very intuitive how a PD works, an incoming photon will
excite an electron from the valence band to the conduction band, and leave a
16 CHAPTER 2. BULK PHOTODIODE

Figure 2.6: Stimulated absorption and emission, and spontaneous emission (at
small forward bias), f1 and f2 are Fermi distribution functions for the two quasi
Fermi levels.

hole in the valence band. The electron will then be pulled to the N-side because
of the lower electron energy there, and the hole (which has a positive charge) will
be pulled to the P-side. The generated electrons and holes can now be measured
as a current. For this process to occur we need a high electron density in the
valence band, and a high hole density in the conduction band. This is best
obtained at reverse bias as can be seen in figure 2.4 (a) to (c).
Besides absorption, we can also have emission of light, and in figure 2.6 the
processes that can occur when light interacts with matter is sketched; stimulated
absorption, stimulated emission, and spontaneous emission (which is stimulated
by the vacuum field).
The transition rate of these three processes can be quantified using Fermi’s
distribution function for the two quasi Fermi levels, the density of transition
pairs ρr (E21 ) (reduced density of states), and a transition probability, where the
two last contributions often is collected in the famous ”Fermi’s golden rule”[2]:

2π 0 2
Rr = |H21 | ρr (E21 ) (2.7)
~
0
where H21 is a transition matrix element.
To get the transition rates for the three processes we now only need to include
Fermi’s distribution function for the two quasi Fermi levels; f1 and f2 :

R12 = Rr f1 (1 − f2 )
R21 = Rr f2 (1 − f1 ) (2.8)
sp0
R21 = Rrvf f2 (1 − f1 )
2.3. LIGHT ABSORPTION 17

Figure 2.7: Absorption coefficient for bulk In53 Ga47 As (bandgap at 1676nm,
0.74eV), and for InGaAsP quantum well material (effective bandgap at
1585nm). Calculated on the basis of [2] chapter 4. A Lorentzian lineshape
broadening with 13meV broadening of the energy levels is included [2].

0
sp
Where R21 is the transition rate of spontaneous emission from one pair of
transition levels, R12 and R21 are the transition rates for stimulated absorption
and emission. The latter can be used to find the absorption per length, called
the absorption coefficient α:

−1 dNp −1 dNp 1 1
α≡ = = (R12 − R21 ) = Rr (f1 − f2 ) (2.9)
Np dz Np vg dt Np vg Np vg

Where Np is the density of photons, and it can be shown that[2]:

πe2 ~
α(E21 ) = |MT |2 ρr (E21 )(f1 − f2 ) (2.10)
n²0 cm20 E21

where MT stems from the matrix element, and is tabulated in [2] for bulk
In0.53 Ga0.47 As to be |MT |2 ' 4m0 , where m0 is the free electron mass. In this
way the absorption coefficient as function of the photon energy is calculated, and
plotted in figure 2.7. At 1550nm the absorption coefficient can be estimated to
about 0.5µm−1 , and if 5 % of the light is confined in the active region (depending
on the device see table 2.4) we need the length of the device to be a few times
1/(5% 0.5µm−1 ) = 40µm for all the light to be absorbed. The length of all
devices is about ∼ 200µm, so all light will be absorbed (at appropriate reverse
bias).
The generated EHPs can, if they do not recombine, be detected as current, called
photocurrent. To quantify this process we define two very important parameter,
18 CHAPTER 2. BULK PHOTODIODE

Figure 2.8: Measurement of quantum efficiency η versus incident wavelength


for PIN5, at 5volt reverse bias, and 1mW optical input power. At each mea-
surement point the optical fiber was aligned to the device for optimal coupling.

called the responsivity and the quantum efficiency. The responsivity (R):
Current generated
R= (2.11)
Power of incident light
tells us how many ampere of current we will get per watt incident light.
The quantum efficiency (η):
Number of detected carriers
η= (2.12)
Number af incident photons
tells us how many of the incident photons that will generate detectable carriers
to the photocurrent. The quantum efficiency is a number between 0 and 1, and
includes all kinds of loss.
In figure 2.8 a measurement of the quantum efficiency versus incident wave-
length, is shown. The graph is made by measuring the photocurrent at each
wavelength, and then divide it by eP/hν, where P is the incident optical power,
and hν is the photon energy.
The graph shows that the quantum efficiency is highest for the lower wave-
lengths. There are a number of ways to explain this, which we will look at a
little later.

We know that each photon has an energy hν, and that each generated carrier
has a charge e, so we can express the the responsivity in terms of the quantum
efficiency:
ηe
R= (2.13)

2.3. LIGHT ABSORPTION 19

PIN2 PIN3 PIN4 PIN5 PIN6


2µm ridge [A/W ] 0.520 0.657 0.722 0.745 0.712
1.5µm ridge [A/W ] 0.505 0.634 0.711 0.719 0.683
PIN7 PIN8 PIN12 PIN13
2µm ridge [A/W ] 0.640 0.750 0.745 0.748
1.5µm ridge [A/W ] 0.617 0.710 0.687 0.699

Table 2.3: Responsivity measurements on the bulk PDs, all with the same
optical fiber, no AR coating, and reverse bias −5volt, except PIN8 at −15volt
bias. All numbers are the average of measurements on at least 3 different devices.

For the perfect PD η will be equal to 1, which means that every incident pho-
ton generate one detectable EHP. Thus for the perfect PD the responsivity at
1550nm will be 1.25A/W . This is the absolute maximum responsivity we can
get from the PDs (without amplification).
Measurements of the responsivity of the PDs are listed in table 2.3. The best
observed responsivity is about 0.75A/W , so this device has a quantum efficiency
of η = 0.75/1.25 = 60%. At the facet 27% of the incident light is being reflected
(see section 2.4), so if the device were antireflection (AR) coated it could give
a responsivity of about 1A/W and have a quantum efficiency of 80%. The re-
maining loss, which we will look at later, can be attributed to effects like; 1) loss
through the optical fiber (section 2.4), 2) mode overlap between the incoming
mode from the fiber, and the mode sustained in the waveguide (section 2.4),
3) absorption outside the active region, due to free carrier loss (later in this
section), and 4) recombinations of the generated EHPs before they are detected
(section 2.4).

When we send a continuous wave (CW) beam of light into the device, it will
be absorbed and generate EHPs, which will change the I-V curves from figure
2.5. The photocurrent at reverse bias is simply given by the responsivity:

Iphoto = R P (2.14)

However as we increase the bias from reverse to forward we inject more and more
minority carriers; electrons from the N-side into the conduction band, and holes
from the P-side into the valence band, especially into the active layer because
it has the lowest bandgap. So as we increase bias (from reverse to forward)
we begin to fill the conduction band of the active layer with electron, and the
valence band with holes, which is called band filling. This will cause two things;
1) a reduction of the absorption coefficient, because the injected minority car-
riers induces a higher effective bandgap of the active layer, and 2) an increased
EHP recombination rate, because of the increased carrier density. As bias is
increased, the I-V curves will therefore gradually return to equation 2.6.
In figure 2.9 I-V curves for all devices at different input powers are shown, the
20 CHAPTER 2. BULK PHOTODIODE

input powers for PIN number 3, 4, 5, 7, 8 are Pin = 0, 1, 3, 5, and 7 mW with


one optical fiber, and for the rest of the devices the input powers are Pin = 0,
0.25, 0.75, 1.25, and 1.75 mW, because for those measurements a better optical
fiber, that couples more of the light into the device, is used. But the step wise
increase in input power are maintained, so the qualitatively features of the plots
can be compared between all devices.
The first thing to note is the change in current at reverse bias, compared to fig-
ure 2.5. The magnitude of the current at reverse bias is linear with the incident
optical power, which agrees with equation 2.14.

Here we will look at free carrier loss. Absorption of light by free electrons
in the conduction band, or free holes in valence band do not generate EHPs,
and is therefore considered as loss (called free carrier loss). Because there are
high densities of free carriers in doped materials compared to intrinsic materials,
doped materials will give rise to far most free carriers loss. Furthermore because
there are three hole bands (heavy hole, light hole, and split off), and only one
electron band, P-type material give rise to most free carrier loss.
If we compare device PIN5 and PIN8, where the only difference is that PIN8
has a 950nm standoff layer on the P-side, and PIN5 has 150nm standoff layer.
We see that PIN8 needs more reverse bias to pull out all the generated carriers,
but PIN8 does not give more photo current than PIN5. The large standoff layer
of PIN8 will ”shield” the light mode in the waveguide so that no light is in the
P-doped layer, where free carrier absorption could occur. This suggests that
free carrier loss is negligible for these devices.
But by comparing PIN4 and PIN5, where the only difference is that PIN5 has
a 150nm standoff layer and PIN4 have none, we see that PIN5 has the highest
responsivity. The optical power in the optical mode in the waveguide is centered
in the active layer (see section 2.4), so the closer the P-doped material is to the
active material, the more free carrier loss it could induce. This can explain why
PIN5 have a higher responsivity that PIN4.
From the last section we know that it is reasonable to assume that the Zn-ions
from the P-doped layer will diffuse into the intrinsic layers, which will make the
Zn-atoms get even closer to the active layer. And the standoff layer is intended
to keep the P-doped atoms away from the active region.
On the basis of this discussion it seems that a small standoff layer, of 150nm or
less, makes free carrier loss negligible.

Another thing to note is that some of the I-V curves have one bend, and some
have two bends, in the transition from photo current in reverse to recombina-
tion current i forward. PIN number 2, 3, 4, 12, and 13 goes very abruptly from
absorbing the incoming light, to generate recombination current and have only
one bend. The other devices, PIN number 5, 6, 7, and especially 8, decrease the
photo current to zero, and then as bias is increased further the recombination
current begins.
This transition region is actually very interesting, because in this region we have
band filling. And band filling causes, as discussed above, two effects; a reduction
2.3. LIGHT ABSORPTION 21

Figure 2.9: I-V curves for all devices, with different optical input powers as
indicated on each graph.
22 CHAPTER 2. BULK PHOTODIODE

Caption: Same as previous page.


2.3. LIGHT ABSORPTION 23

Caption: Same as previous page.


24 CHAPTER 2. BULK PHOTODIODE

Figure 2.10: (a) Photocurrent at reverse bias for device PIN8, at two differ-
ent wavelengths (1530nm and 1610nm), with the same photon current Nph ,
measured at four different photon currents. (b) Magnitude of the 1530nm pho-
tocurrent minus the magnitude of the 1610nm photocurrent.

in absorption coefficient α, and an increased EHP recombination rate. These


two effects can actually be probed by making an I-V plot at two different wave-
lengths with the same photon current Nph . In figure 2.10(a) this is shown for
PIN8 at two wavelengths; 1530nm and 1610nm, each with a photon current of
Nph =1, 2, 3, and 5·1016 photons/second. In figure 2.10(b) the magnitude of the
current from the more energetic photons (1530nm), are subtracted the magni-
tude of the current from the less energetic photons (1610nm). It can be seen
that at small reverse bias, the more energetic photons (1530nm) results in most
current. Then, from a few volt reverse bias to about 8 volt reverse bias, it is the
less energetic photons that gives the most current. And finally at even higher
reverse bias it again is the more energetic photons that gives most current.
The same effect can be seen for PIN5, though the effect is much harder to see
because it all happens around ±0.5V .
The effect seen in figure 2.10(b) can be explained in the following way: As the
reverse bias is increased the absorption also increases (band filling decreases),
and from zero to a few volts not all light is absorbed (the absorption length is
longer then the length of the device ), therefore it is favorable with more ener-
getic photons that have a higher absorption coefficient (see figure 2.7). When
the curve crosses the y-axis both wavelength generate equal amount of current,
so about at this point all light is being absorbed in the waveguide. When we in-
crease the reverse bias further (∼ −4volt to ∼ −8volt)) all light will be absorbed,
but the more energetic photons (that has the highest absorption coefficient) are
just using a smaller portion of the waveguide to be absorbed, and hence the
generated carrier density will be higher. This higher carrier density generated
by the energetic photons will lead to more recombinations and therefore less
2.4. WAVEGUIDE 25

detectable current.
In the last section of the graph there is no more bandfilling i.e. the depletion
region is really depleted of carriers, and the more energetic photons again gives
the most current.
PIN5 also gives more photo current at shorter wavelengths, which we saw in
figure 2.8 for the quantum efficiency. There are a number of ways to explain
this: 1) The more energetic photons will give an extra kick to the generated
EHPs, and therefore help the EHPs over the barriers at the grading layers,
which in turn gives less recombination loss, 2) the mode overlap will depend on
wavelength, or 3) the mode in the waveguide are more confined for the shorter
wavelengths, which will lower free carrier loss in the P-side. However, we know
that free carrier loss is negligible for PIN5, so this is most likely not the reason.
There are three grading layers with four steps at both the valence and conduction
band, which gives an average barrier height of (1.35eV −0.74eV )/8 = 75meV at
each step. 75meV is three times higher then the thermal energy, so these steps
are indeed barriers for the carriers. It therefore seems likely that recombinations
of the EHPs will contribute to loss of electrons.

To summarize this section, we have seen that the material absorption co-
efficient for InGaAs is α = 0.5µm−1 at 1550nm. The best responsivity was
measured to be R = 0.75A/W , which corresponds to a quantum efficiency of
η = 60%, (or if AR was used; R = 1A/W , and η = 80%). We have also seen
that free carrier loss are negligible, when a small standoff layer of 150nm or less
are used, but that recombinations of the generated EHPs still can give loss.

2.4 Waveguide
The incident light comes from an optical fiber where it is confined in the trans-
verse plan, and has some transverse mode profile. At the end of the fiber there
is a lens that focuses the light onto the facet of the device (see figure 2.1). As
we will see later in this section, 27% of the incident light will be reflected at the
facet, but this effect can be eliminated by AR coating the device, which we will
see in section 3.2.
Inside the device the light will be guided by total internal reflection, because
the active layer has a higher refractive index than InP and BCB, forming a
so-called waveguide. The waveguide guides the light through the device with
some transverse mode profile, and keeps the light centered in the active layer.
The amount of incoming light that will be guided by the waveguide depend on
how well the mode profile of the incoming light from the optical fiber, matches
the mode profile in the waveguide.
The waveguide in the device is centered around the absorbing layer, and in fig-
ure 2.11 a simplified schematic of the different refractive indices in the device,
as seen from the facet, is shown.

The equations governing the photon mode in the waveguide is Maxwell’s


26 CHAPTER 2. BULK PHOTODIODE

Figure 2.11: The refractive indexes in the device as seen from the facet, and
the predominant directions for the electric field and magnetic field for the two
polarization modes.

equations, and by combining them the right way we reach two important wave
equations:
n2 ∂ 2 E n2 ∂ 2 H
∇2 E = 2 ∇ 2
H = (2.15)
c ∂t2 c2 ∂t2
for the electric field E, and the magnetic field H, where the refractive index n
depend on space coordinate according to figure 2.11. To solve these equations
we need approximations, and the first approximation comes from a guess on the
solution of the form:

E(x, y, z, t) = êE0 U (x, y)eikz−iωt (2.16)

and equivalently for the magnetic field. The polarization e is approximated


to be independent of the space coordinates. Then we have an amplitude E0 ,
a mode profile in the transverse plane U (x, y), and a plan wave eikz−iωt , that
makes the mode travel down the waveguide. The polarization ê of the light can
have two directions, see figure 2.11: 1) E is predominantly in the transverse
direction, and H is predominantly in the lateral direction, called Transverse
Electric (TE), or 2) H is predominantly in the transverse direction and E is
predominantly in the lateral direction, called Transverse Magnetic (TM).
By inserting equation 2.16 into equation the wave equation 2.15 we get:
µ 2 ¶
∂ ∂2 2 2 2
+ + k − n k 0 U (x, y) = 0 (2.17)
∂x2 ∂y 2

where we use k to denote the wave vector in the medium, and k0 = ω/c to
denote the free space wavevector. For the magnetic field we would get the exact
same equation, so we need only to solve this equation.
To solve equation 2.17 we need another approximation, and a widely used ap-
proximation is the effective index method [2], which first uses separation of
2.4. WAVEGUIDE 27

Figure 2.12: Effective index method calculation of the mode in PIN6

variables U (x, y) = U (x)U (y), then defines an effective refractive index nef f in
each of the three vertical regions 1, 2, 3 of figure 2.11: nef f,i ≡ k/k0 where
i = 1, 2, 3, this gives:
∂ 2 U (x)
)
1
U (x) · ∂x2 = −kx2
1 ∂ 2 U (y) 2
where − kx2 − ky2 + k02 n2ef f,i = n2 k02 (2.18)
U (y) · ∂y 2 = −ky

The next part of the effective index method is to have a thin slap, which means
the central rectangle in figure 2.11 has to be much wider then its height, so that
U (y) changes slowly compared to U (x), and we can approximate ky = 0.
The boundary conditions can to a good approximation be set to be continuous
and differentiable, (the discontinuity that arises when the electric field is perpen-
dicular to the interface will only give a small correction to to this approximation,
because the differences in refractive index are small). This approximation makes
both TE and TM modes the same.
This approximation turns equation 2.18 into a 1D problem in each of the three
lateral x-direction (1,2,3 in figure 2.11), which results in an effective refractive
index in each of the three regions; nef f,i (n = 1, 2, 3). The final part of the effec-
tive index method is to use these three effective indices to a final one-dimensional
equation in the transverse direction (y-direction), which results in a mode pro-
file and a ”total” effective refractive index nef f . This has been performed on
devices PIN6, and the result can be seen in figure 2.12.
For more accurate calculations of the mode in the waveguide, I have used a
computer program called Selene, and not the simplified structure in figure 2.11,
but the true structure from figure 2.1 and table 2.1. Selene uses a numerical
28 CHAPTER 2. BULK PHOTODIODE

method called ”multi grid finite difference” (MGFD). This program has been
used on all devices that have different modes; PIN number 2, 3, 5, 6, and 7,
with ridge widths of 1.5µm and 2µm, see table 2.4. The program is used to
calculate the confinement factor Γ, which is the fraction of optical power that
is confined within the active layer. From the confinement factor we calculate
the absorption length, which is the length at which the power of the light has
dropped to 1/e, as (αΓ)−1 . Selene is also used to find the effective refractive
index, and some characteristic field dimensions such as full width at half maxi-
mum (FWHM) waist, and the 1/e waist.

Now that we know the mode inside the waveguide, we need to find what the
incoming mode that hits the facet looks like. Because, then we can find the
overlap of the two modes, which tells us how much of the incoming light couples
into the waveguide. This mode overlap give rise to another loss term, and is
included in the quantum efficiency with a factor ηmo .
The responsivity depends strongly on what optical fiber is used, responsivity as
low as 0.14A/W for PIN5 has been measured, and all the way up to 0.75A/W
with the best fiber. In figure 2.13 measured far fields from three different optical
fibers, and the corresponding responsivities when they are used to couple light
into PIN5, are shown. A far field is optical intensity as function of a transverse
angle and a lateral angel. The best fiber, which is named A4-2, is used for the
rest of the measurements in the report.
The farfield will hit the facet of the PD, and the amount that will be transmit-
ted through the facet depends on the angle of incident according to Fresnel’s
equations[6]. For the best fiber (A4-2, figure 2.13 right) the FWHM angular
width, and 13.5% angular width is measured to be ±15deg, and ±26deg respec-
tively. Therefore the polarization of the incoming light, whether it is TE or
TM, will be polarized predominant parallel to the facet. According to Fresnel’s
equations the transmission is reduced 1.2%, and 3.7% for light polarized parallel
to the facet, at the FWHM angular width (15deg), and at the 13.5% angular
width (26deg) respectively.
We therefore, to a good approximation, say that all of the light in the mode
from the fiber will bee transmitted through the facet with the same coefficient
ηf acet = 1 − ((nef f − nair )/(nef f + nair ))2 . The factor ηf acet is part of the
quantum efficiency, and is listed in table 2.4.
With this approximation we also say that the mode that enters through the
facet, is the measured farfield projected onto the facet. So to quantify the mode
overlap, the mode inside the waveguide (calculated by Selene), and the mode
from the measurement farfield projected onto the facet, is used. Mode overlaps
from such calculations are also part of the quantum efficiency, and is called ηmo .
The factor ηmo is listed in table 2.4, and for comparison a mode overlap with a
spherical symmetrical gauss mode, is also included.

There is also measured loss in the optical fiber A4-2. With a surface PD
with known quantum efficiency, there is measured a loss of 8% through the fiber.
When we send 1mW into the fiber, only 0.92mW comes out at the end of the
2µm Ridge 1.5µm Ridge
PIN2 PIN3 PIN5 PIN6 PIN7 PIN2 PIN3 PIN5 PIN6 PIN7
Confinement Γ 65 % 23 % 6.9 % 15 % 2.0 % 56 % 22 % 4.8 % 14 % -
2.4. WAVEGUIDE

Absorption length (αΓ)−1 [µm] 3.2 8.8 28 13.6 104 3.6 9.2 42 15 -
Effective refractive Index 3.42 3.25 3.18 3.20 3.17 3.41 3.24 3.17 3.19 -
Transmission; ηf acet 70 % 72 % 73 % 73 % 73% 70% 72% 73% 73% -
FWHM transverse [µm] 1.63 1.24 1.18 1.18 1.20 1.45 1.01 0.94 0.94 -
FWHM lateral [µm] 0.40 0.33 0.47 0.38 0.62 0.40 0.33 0.49 0.38 -
1/e transverse [µm] 1.40 0.97 0.90 0.90 0.91 1.36 0.81 0.73 0.73 -
1/e lateral [µm] 0.33 0.37 0.64 0.49 0.96 0.33 0.37 0.72 0.49 -
Mode overlap with Gaussian 65 % 85 % 94 % 92 % 72 % 65 % 88 % 80 % 93 % -
1/e of gauss Gaussian∗ [µm] 0.7 0.7 0.9 0.8 1.5 0.7 0.6 1.1 0.7 -
Mode overlap with A4-2; ηmo 71 % 88 % 95 % 95 % 77 % 74 % 92 % 84 % 96 % -
Responsivity [A/W ]
Calculated ∗∗ 0.57 0.73 0.80 0.80 0.65 0.60 0.76 0.71 0.81 -
Measured with A4-2 0.51 0.66 0.75 0.71 0.64 0.51 0.63 0.72 0.68 0.62

Table 2.4: Calculations of important parameters for the mode in the waveguide, done by the computer program Selene. PIN
number 4, 8, 9, 10, 12, and 13 has the same mode as PIN5. ∗ Waist of the spherical symmetric gauss mode, that gives the
maximum mode overlap. ∗∗ calculated as 1.25A/W ηf iber ηf acet ηmo
29
30 CHAPTER 2. BULK PHOTODIODE

Figure 2.13: Farfield measurements on three different optical fibers, and the
corresponding responsivity measured when the fiber is used to couple the light
into PIN5. The farfield to the right is from fiber A4-2, and this fiber is used far
most.

fiber. This loss in the optical fiber is also included in the quantum efficiency
with a factor ηf iber = 92%.

So far we have discussed loss effects like, loss through the optical fiber,
reflection at the facet, mode overlap, free carrier loss, and recombination of
the generated carriers. Loss thorough the fiber ηf iber , transmission through the
facet ηf acet , and mode overlap ηmo are known (see table 2.4). We also know that
devices with a standoff layer has negligible free carrier loss, which all devices
in table 2.4 except PIN2 has. Therefore only loss due to recombinations of
generated carriers is left unknown.
By setting the quantum efficiency η from equation 2.12 equal to the known
loss terms: η = ηf iber ηf acet ηmo , we can calculate the responsivity as R =
η 1.25A/W . This calculated responsivity is listed table 2.4, together with the
measured responsivity for comparison.
We see that the measured responsivity matches fairly well with the calculated
responsivity, there only seems to be an additional loss raging from 0% to about
15%, with an average of 8%. Since the only loss term that are not accounted
for, is loss due to recombinations of generated carriers, these last ∼ 8% loss
could be due to that.

2.5 Dynamics
Up until now we have looked at the detector in DC measurements and with
CW light input. But, when we start to modulate the light input we also get
a modulated electrical output, and it is of course in such a setup the detector
is intended to work. In this section we will look at what happens when we
modulate the input signal, and how fast we can modulate it, while still being
2.5. DYNAMICS 31

able to retrieve the information in the output signal.


There are two ways to modulate the light input; either with a sinusoidal fre-
quency (done with a network analyzer), or with information in the form of bits
(done with a bit error rate test-set).
If we modulate the light input with a sinusoidal frequency, we can measure
the AC responsivity, as function of the this modulation frequency. The AC
responsivity is the amplitude of the modulated optical power input, divided by
the amplitude of the electrical current output. The frequency where the AC
responsivity has dropped to the half (-3dB), is defined as the bandwidth of the
PD.
If we modulate the light input signal with information in the form of bits, we
simulate a real environment for detector. In such a measurement we need to
compare each optical input bit with the electrical output bit, and measure the
rate of errors, called the bit error rate (BER). The BER is an important quan-
tity for a detector and will depend on the power of the input light, and how
we modulate of the light input. There are three important quantities connected
with how to modulate the light input; (1) The number of bits per second, called
the bitrate B, the bulk PDs are intended to work at B = 10Gb/s so all mea-
surements are performed at this bitrate. (2) The bit sequence, called pseudo
random bit sequence (PRBS), it is the number of bits that are repeatedly being
send into the PD, typically from 27 − 1 (all possible combinations of 7 bits mi-
nus one) up to 231 − 1 (all possible combinations of 31 bits minus one), where a
PRBS length of 231 − 1 is used. (3) the format of the bits can either be RZ, or
NRZ, where the NRZ format is used, because it seems to be the most commonly
used format.
With a bitrate of B = 10Gb/s, and a PRBS length of 231 −1 in the NRZ format,
we can measure the BER as function of the light input power. As we increase
the optical power, the BER will decrease, and the optical power at which the
BER is 10−9 , is defined as the sensitivity of the detector (The sensitivity is
sometimes defined at BER = 10−10 or lower).
So with a network analyzer, and a BER test-set we can measure two important
quantities for the PD, namely the bandwidth and the sensitivity.

2.5.1 Bandwidth
To determine the bandwidth of the device there are some key time parameters
which we will describe here. These key time parameters are connected with the
movement of the electrons. When an incident photon is absorbed and an EHP
is generated, the EHP is pulled out of the depletion region, and this takes a
certain time called the transit time τtr . When the EHP is outside the depletion
region there are two capacitors, the first is the parallel plate capacitor across
the depletion region, called diode capacitance Cd . The second is between the
bonding pad, and the bottom of the depletion layer, called the bonding pad
capacitance Cb . Each of these capacitors has a time constant.
An equivalent circuit diagram for the diode, together with the rest of the electri-
32 CHAPTER 2. BULK PHOTODIODE

Figure 2.14: A schematic diagram of the electrical setup for dynamical measure-
ments, with a equivalence diagram for the diode, a 50Ω load resistor Rl , and a
bias tee.

cal setup for dynamical measurements, are shown in figure 2.14. The diode can
be represented by a current source, two capacitors as just described, a resistor
Rpn originating from the doped P- and N-side, called the contact resistance,
and a large resistor accounting for the small dark current Rdark .
The two capacitors, and the resistor Rpn , can be estimated from the design of
the device. The resistance for a doped semiconductor is given by:
Ln Lp
Rn = + Rp = − (2.19)
ND eµn An NA eµp Ap
for electrons and holes respectively (holes are pulled out through the P-side,
and electron are pulled out through the N-side), where L and A are the length
and the cross section area of the region. Equation 2.19 gives about Rp ' 5Ω
and Rn ' 0.1Ω, therefore the total resistance is Rpn ' 5Ω.
The contact resistance is measured as the differential resistance at high forward
current where the slope of the I-V curve is constant. At 60mA forward current
the resistance is about ∼ 7Ω, so this matches well with the calculation.
The diode capacitance is through the depletion region, which is made up of InP
(standoff layer) with dielectric constant of 12.4, InGaAs with dielectric constant
of 13.9, and 6 layers of InGaAsP with a dielectric constant set to the average
of the two. The area of the plates A are set to the width of the ridge (2µm)
multiplied by the absorption length (see table 2.4), and the capacitance is given
by:
à !−1
X µ ²i ¶−1
C=A (2.20)
i
wi
where ² and w are the dielectric constant, and the width of each of the layers
through the depletion region.
2.5. DYNAMICS 33

Figure 2.15: Model used to calculate the bandwidth of the PDs

The bonding capacitance is calculated in the same way, where the area of the
bonding pads are 50µm × 50µm, and the dielectric materials are BCB (² = 2.7),
and the depletion layer (² is set to 13). The thickness of the BCB is equal to
the hight of the P-side from table 2.1. The calculated diode and bonding capac-
itances for all devices are shown in table 2.5. Because Rpn is much smaller than
the load resistor Rl , which is 50Ω, the discharging time of these two capacitors
can be approximated by τRC = (Cd−1 + Cb−1 )−1 · Rl , which also is written into
table 2.5.
The transit time τtr can be estimated the following way; the maximum elec-
tric field is so high that electrons and holes for all devices reaches the satu-
ration velocity1 . So the transit time is calculated as τtr = wdep /vsat , where
vsat ' 107 cm/s is the saturation velocity for both InP[12], and InGaAs[13].

Now that we have an estimate of the RC discharge time, and transit time,
we can calculate the the bandwidth of the devices, to compare with the mea-
surements.
In figure 2.15 a model to calculate the bandwidth, with three reservoirs and flow
rates between, is shown. A similar modal for a diode laser can be found in [2].
The three reservoirs are; a photon reservoir with Np photons, a carrier reservoir
in the active region with Na carriers, and a carrier reservoir of detectable car-
riers (carriers on the two capacitors) with N carriers.
The incoming photon rate is the incident power divided by the energy of each
photon, and here we have loss due to reflection at the facet, mode overlap,
and loss through the optical fiber, which collected is designated η0 (η0 =
ηf iber + ηf acet + ηmo ). From the photons in the device we could have free
carrier loss, but that is negligible as discussed, instead we have stimulated ab-
1 the velocity where the kinetic energy and the thermal energy is about the same
CHAPTER 2. BULK PHOTODIODE

PIN2 PIN3 PIN4 PIN5 PIN6 PIN7 PIN8 PIN9 PIN10 PIN12 PIN13
Emax [kV /cm] 87 224 432 226 205 242 152 152 226 235 239
Depletion region [nm] 728 303 190 302 328 285 1088 1088 302 281 273
Diode capacitance [f F ] 1 7 35 21 9 79 6 6 21 23 24
Pad capacitance [f F ] 22 24 24 24 24 24 22 17 18 24 24
RC time τRC [ps]∗ 1.2 1.6 3.0 2.3 1.7 5.2 1.4 1.2 2.0 2.4 2.4
Transit time τtr [ps] 7.3 3.0 1.9 3.0 3.3 2.9 10.9 10.9 3.0 2.8 2.7
Bandwidth [GHz]
Calculated∗∗ 18.8 34.6 32.5 30.0 31.8 19.7 12.9 13.2 31.8 30.6 31.2
Measured - - 32 34 - - 18 11 29 - -
Table 2.5: Measured and calculated bandwidths, and corresponding estimated dynamical parameters. Calculations are made
at -5 volt bias for all except for PIN8 and PIN9 at -15volt bias. ∗ Bonding pad capacitance and diode capacitance in parallel
multiplied with a 50Ω load resistor. ∗∗ Calculated bandwidth according to equation (2.24).
34
2.5. DYNAMICS 35

sorption R12 and stimulated emission R21 . This absorption takes place in the
active region, so the generated carriers are also in the active region.
The carriers generated in the active region can give spontaneous emission, but
that is negligible because the device works with about 5 volt reverse bias, and
hence the region is depleted of free carriers. The major flow rate of carriers
out from the active region is into the P- and N-side, which happens with the
rate Na /τtr . But because of the barriers at the grading layers, recombinations
could occur, which is included with a loss (Na /τtr )(1 − ηRec ) . From the P-
and N-side the flow of carriers can be seen as discharging of the two capacitors,
diode capacitor and bond capacitor, which is done with the rate N/τRC .
So to calculate the number of detectable carriers N (t), generated from a modu-
lated input power P (t), we can make a set of three coupled differential equations,
called rate equations:

dNp (t) η0 P (t)


= − (R12 − R21 )
dt hν
dNa (t) Na (t)
= (R12 − R21 ) − (2.21)
dt τtr
dN (t) Na (t) N (t)
= · ηRec −
dt τtr τRC

From equation 2.9 we have (R12 −R21 ) = Np (t)Γαvg , and by modulating the in-
put power with a specific angular frequency ω, amplitude P1 around a DC value
P0 ; P (t) = P0 + P1 eiωt and using Fourier analysis we can find the detectable
number of particles as function of the modulation frequency:
µ ¶
η0 ηRec τRC P1
N (ω) = P0 + (2.22)
hν (1 − iωτRC )(1 − iωτtr )(1 − iω(Γαvg )−1 )

We see that it has at constant DC term and a varying AC term, and for the
bandwidth we are only interested in the varying AC term. The AC responsivity
from the AC term in equation 2.22 will then be given by.
µ ¶
η0 ηRec e 1
R(ω) = (2.23)
hν (1 − iωτRC )(1 − iωτtr )(1 − iω(Γαvg )−1 )

To find the bandwidth ∆f we need to find the frequency where the responsivity
has dropped to the half (-3dB), that is to solve 0.5 · R(0) = R(∆f 2π) for ∆f
(ω in the equations above is the angular frequency). The result can to a good
approximation be estimated by:

1
∆f = (2.24)
2π(τRC + τtr )

which also is seen in textbooks[5]. Calculated values for all devices are shown
in table 2.5.
36 CHAPTER 2. BULK PHOTODIODE

Figure 2.16: Block diagram of the setup for measuring the bandwidth.

To measure the bandwidth we need to measure the responsivity as function


of the modulation frequency. This is done on five of the most interesting devices,
namely PIN number 4, 5, 8, 9 and, 10. They all have the same active layer and
same grading layers, and therefore also the high DC responsivity (∼ 0.75A/W ).
However, they differ a lot in the RC time and transit time, because they have
different standoff layer and ridge height, so therefore these five devices have
been chosen for the bandwidth measurements.
In figure 2.16 a block diagram of the measurement setup for the bandwidth mea-
surements, is shown. The network analyzer has an electrical output and input.
The output is a modulated electrical signal from 50M Hz to 40GHz, and at the
input an electrical signal at the same frequency is detected. The network ana-
lyzer sweeps the output from 50M Hz to 40GHz, measure the input, and gives
the input output ratio in decibel units as function of the modulation frequency.
I want to measure on a PD, therefore I need to convert the electrical output
from the network analyzer to an optical signal, this is done with a CW laser
and an electro absorption modulator (EAM). The measured input output ratio
as function of the modulation frequency is then the responsivity of both the
EAM and the PD, we therefore need to calibrate out the response curve of the
EAM. This is done by inserting a commercial PD with a known response curve,
and then calibrate the network analyzer to give zero decibel at all frequencies.
When we then insert my PD, the measured response curve will then be relative
to the response curve of the commercial PD, and we therefore need to subtract
the response curve of the commercial PD from the measured response curve of
my PD, to get the real response curve of my PD.

The measured (and calibrated) response curves are shown in figure 2.17,
where each curve represents a measurement on a different device. The band-
width can now be read of the curves for each device, as the point where the
curve crosses -3dB, and this is written into table 2.5.
The measured bandwidth matches quit well with the calculated bandwidth, ex-
cept for PIN8 which has a discrepancy of almost 40%. This discrepancy is the
same for both devices that are measured on, and since the limiting factor for
PIN8 is the transit time (τtr À τRC ), the discrepancy might be due to an error
2.5. DYNAMICS 37

Figure 2.17: Responsivity (in dB units) versus modulation frequency, for the
-3dB bandwidth measurement. The different curves in the same plot are mea-
surements on different devices.
38 CHAPTER 2. BULK PHOTODIODE

Caption: Same as previous page.


2.5. DYNAMICS 39

in growing the high standoff layer. I have no better explanation!

An increase in standoff layer gives lower diode capacitance and therefore


lower RC time, but a larger transit time. The change in standoff layer from
PIN4 to PIN5 that according to the calculations should make PIN4 the fastest,
can not bee seen in the measurements. Probably because the change in total
time (τtr + τRC ) according to the calculations only is 0.4ps, which must be con-
sidered small compared to the uncertainty in the estimated τtr , and τRC , and
also compared to the uncertainty in the bandwidth measurement. But the big
change in standoff layer from PIN5 to PIN8, which according to the calculations
should make PIN5 the fastest, matches well with the measurements.
An increase in ridge height gives lower bond capacitance and therefore lower
RC time. The change in ridge height from PIN5 to PIN10, and from PIN8 to
PIN9, which should make PIN10 and PIN9 the fastest can not be seen in the
measurements either. Again probably because the change in RC time (0.3ps),
is to small to be verified by the measurement.

To summarize the bandwidth measurements; the most noticeable difference


is that PIN number 4, 5, and 10 are relatively good with a bandwidth about
& 30GHz, and that PIN8 and 9 has a lower bandwidth of about 11 to ∼ 18GHz.

2.5.2 Sensitivity
To measure the sensitivity we need to measure the BER as function of the inci-
dent power, and the sensitivity is in this project defined as the power where the
BER is 10−9 . Sensitivity measurements are performed on the same 5 devices
as for the bandwidth measurements, namely PIN number 4, 5, 8, 9, and 10,
because they have the largest changes in RC time and transit time.
A block diagram of the electrical setup for the sensitivity measurements is shown
in figure 2.18. It starts with a CW laser and a erbium doped optical fiber am-
plifier (EDFA) that gives 12.5 dBm optical power at 1550nm, followed by a
polarization controller, and a Mach-Zehnder modulator. The polarization con-
troller is used because the Mach-Zehnder modulator depends on polarization.
The Mach-Zehnder modulator is controlled by the BER test-set with a PRBS
of 231 − 1, at B = 10Gb/s, and its output can be seen on an oscilloscope. The
output from the Mach-Zehnder is also send into another EDFA, because the
Mach-Zehnder modulator has a high loss (∼ 7dB), and then into an attenuator,
and a power-meter so we can measure the average optical power that we send
into the PD. After the PD we insert an electrical preamplifier, whose output is
send into the BER test-set, to measure the BER, and into the oscilloscope, to
measure an electrical eye.
The oscilloscope is triggered at a much lower frequency than the B = 10Gb/s,
but it measures a lot of data points from different bit sequences and plots them
on the same plot. So, the plot from the oscilloscope has bit1’s, and bit0’s with
all different kinds of histories (the history of a bit is the bit sequence that lays
before the bit). Such a plot from the oscilloscope looks like an eye (see figure
40 CHAPTER 2. BULK PHOTODIODE

Figure 2.18: Block diagram of the setup for measuring the sensitivity, and the
eye diagrams (optical and electrical). The Photodiode box is equivalent to the
entire figure 2.14 (including load resistor, and bias tee)

Figure 2.19: The optical eye diagram of the Mach-Zehnder modulator, and three
different electrical eye diagrams of PIN8 at different optical input powers.

2.19), which is why it is called an eye diagram.


Figure 2.19 shows the optical eye diagram, and three different electrical eye di-
agrams obtained from PIN8, at different optical input powers; Pin = −5dBm,
−7dBm, −9dBm.
As can be seen in figure 2.19 there are noise on the bit1, and bit0, and the noise
to eye-opening ratio increases as the optical power decreases. This noise makes
it difficult to determine whether a bit is a bit1, or a bit0, and there will therefore
be some error readings which is measured as the BER.

There are two terms of noise in a PD; shot noise, and thermal noise. Shot
noise is a manifestation of the fact that current is a stream of electrons that are
generated at random times, and is given by[5]:
p
σs = 2eI∆fn (2.25)

where I is the current, and ∆fn is the bandwidth of the (white) noise, which is
experimentally set by a low pass filter (in this case the bias tee; 12.5GHz). By
far the largest current in this receiver is the current in the 50Ω load resistor,
2.5. DYNAMICS 41

for PIN4, 5, and 10 the current is about 100mA (at 5volt reverse bias), and for
PIN8, and 9 it is about 300mA (at 15volt reverse bias).
The thermal noise stems from the fact that electrons in a conductor, at finite
temperature, moves randomly. In this receiver we have a 50Ω load resistor RL
that gives the following thermal noise [5]:
p
σT = 4kB T ∆fn Fn /RL (2.26)

where kB is Boltzmanns constant, T is the temperature, and Fn is a noise


figure of the electrical preamplifier. The electrical preamplifier used in these
measurements has a noise figure of 4.
A very important parameter connected with the eye-diagram is the Q-factor [5]:

I1 − I0
Q= (2.27)
σ1 + σ0

where I1 , and I0 are the currents of the bit1, and bit0, and σ1 , and σ0 are the
standard deviations of the bit1, and bit0. The standard deviations are for these
devices given by σ02 = σ12 = σs2 + σT2 . The Q-factor for an eye diagram is the
opening of the eye, divided by standard deviations for the top and bottom of the
eye. The eye opening can be related to the average optical power; by looking
at figure 2.19 (left), we can see that the eye opening is equal to ∼ 1.3 times the
average optical power. This means: I1 − I0 ' 1.3P R, where P is the average
optical input power (measured by the power meter) and R is the responsivity.
There are an approximate relation between the Q-factor and the BER [5]

exp(−Q2 /2)
BER ' √ (2.28)
Q 2π

By combining the equations above we can calculate BER as function of the aver-
age optical input power. The power at which the BER is 10−9 (the sensitivity)
is −6.5dBm for PIN4, 5, and 10, and for PIN8, and 9 it is −4dBm.

The measured BER versus the average optical input power is shown in figure
2.20, including an electrical eye diagram at the sensitivity power. We know that
PIN5 has RC- and transit-time of a few ps, and in the eye diagrams in figure
2.20 we can see the the rise and fall times are about 50ps. However, in these
measurements the time constants are not limited by the PD, but by the electrical
bias tee that cuts off at about 12.5GHz. This is also the reason why all eye
diagrams in figure 2.20 has about the same up and down times, even though we
know that the devices has different bandwidth.
The measured and calculated sensitivity are listed in table 2.6. The measured
sensitivity matches very well with the calculated sensitivity for PIN4, 5, and 10.
The measured sensitivity for PIN8, and 9 are smaller than for PIN4, 5, and 10,
though not as much as the calculations indicate.
42 CHAPTER 2. BULK PHOTODIODE

Figure 2.20: Measured BER versus optical input power, and electrical eye dia-
grams at the sensitivity power.
2.6. SUMMARY 43

PIN4 PIN5 PIN8 PIN9 PIN10 Unit


Calculated -6.5 -6.5 -4 -4 -6.5 dBm
Measured -7 -6.5 -5 -5.8 -6.5 dBm

Table 2.6: Calculated and measured sensitivity for five devices.

2.6 Summary
In this section I will compare the different PDs and try to find the best, then I
will compare this best PD with other commercial diodes, and at last I will give
some ideas on how to improve the PD.

2.6.1 Comparison of devices


Key parameters we can use to compare the PDs are; responsivity, dark current,
bandwidth, sensitivity, and to some degree also the breakdown voltage. Param-
eters like the total capacitance, and RC- and transit-time constants are included
in the bandwidth parameter. All these parameters are listed in table 2.7 under
electrical characteristics, and they are measured with the optical fiber, A4-2.
We also need to have some optical characteristics of the mode in the waveguide,
in case another optical fiber is used. Optical characteristics of the PD are there-
fore also included in table 2.7, where important parameters, calculated by the
program Selene, are listed. The most important optical parameter is the mode
overlap, and from section 2.4 the mode overlap with the mode from the good
optical fiber A4-2, and from a perfect spherical gaussian mode was found. For
a general parameter of the mode overlap, the gaussian mode overlap is listed in
table 2.7. The optical characteristic part of table 2.7 also include the material
absorption coefficient of the active layer, the 1/e waist (transverse/lateral) of
the mode in the waveguide, the confinement factor, and effective refractive in-
dex.

To decide which device is the best, I will start by comparing the responsiv-
ity, and a high responsivity of about 1A/W (with AR coating) is seen for PIN
number 4, 5, 6, 8, 9, 10, 12, and 13 (because they all have about the same
mode overlap). The dark current is < 1nA for all these devices. The Band-
width of these devices is & 30GHz except PIN8, and 9 which has a somewhat
lower bandwidth (because for the large transit time). The sensitivity of the
devices are limited by the shot noise of the current in the load resistor, and is
the same for all devices, except PIN8 and 9 that has a slightly lower sensitivity.
A commonly used reverse bias voltage is 5volt, and since this is very close the
breakdown voltage of PIN13, we have to exclude this device.
On the basis of this discussion the best devices are PIN number 4, 5, 6, 10, and
12, which only have small differences in responsivity and bandwidth. Looking
at these five devices from a production point of view PIN10 must be somewhat
CHAPTER 2. BULK PHOTODIODE

Electrical characteristics
PIN2 PIN3 PIN4 PIN5 PIN6 PIN7 PIN8 PIN9 PIN10 PIN12 PIN13
Bias voltage [volt] 2–23 2–11 2–6 2–10 2–12 2–11 12–30 12–30 2–10 2–7 2–4
Responsivity∗ [A/W ] 0.88 0.90 0.99 1.02 0.98 0.88 1.03 1.03 1.02 1.02 1.03
Dark current [nA] ∼2 <1 <1 <1 <1 <1 <1 <1 <1 <1 <1
Breakdown voltage [volt] 25 13 8 12.5 14 13.5 ∼ 35 ∼ 35 12.5 9 6
Contact resistance [Ω] ∼7 ∼7 ∼7 ∼7 ∼7 ∼7 ∼7 ∼7 ∼7 ∼5 ∼3
Capacitance [f F ] 23 31 59 45 33 103 28 23 39 47 48
RC time [ps] 1.2 1.6 3.0 2.3 1.7 5.2 1.4 1.2 2.0 2.4 2.4
Transit time [ps] 7.3 3.0 1.9 3.0 3.3 2.9 10.9 10.9 3.0 2.8 2.7
Bandwidth [GHz] - - 32 34 - - (18) 11 29 - -
Sensitivity∗∗ [dBm] - - -7 -6.5 - - -5 -5.8 -6.5 - -
Optical characteristics
PIN2 PIN3 PIN4 PIN5 PIN6 PIN7 PIN8 PIN9 PIN10 PIN12 PIN13
Mode overlap with Gauss 65 % 85 % 94 % 92 % 72% 94 %
Absorption coefficient [µm−1 ] 0.5 0.5 0.5 0.5 0.5 0.5
1/e lateral [µm] 1.40 0.97 0.90 0.90 0.91 0.90
1/e transverse [µm] 0.33 0.37 0.64 0.49 0.96 0.64
Confinement 65 % 23 % 6.9 % 15 % 2% 6.9 %
Effective refractive index 3.42 3.25 3.18 3.20 3.17 3.18
Table 2.7: Data sheet for all the bulk PDs. The table is sectioned in electrical and optical characteristics. ∗ with AR coating,
and @ 1550nm. ∗∗ With an external electrical amplifier.
44
2.6. SUMMARY 45

GiGA U2T XL Thor- metro- OEC Unit


(PIN5) labs tek
Responsivity 1 0.65 0.8 1 0.9 0.95 A/W
Dark current < 1 . 50 .1 1.5 .1 . 0.8 nA
Bandwidth & 30 > 50 > 12 >5 2.5 2.5 GHz
Capacitance 45 - . 200 300 600 500 fF
Sensitivity -6.5 - - - - - dBm
Light area Ø (∼ 1) (∼ 1) 20 80 60 60 µm
Note waveguide surface

Table 2.8: A collection of key parameters for six PDs including PIN5 from
GiGA. They are all made with bulk InGaAs active layer and a PIN junction,
GiGA and U2T are waveguide PDs whereas the rest are surface PDs.

more expensive to produce then PIN4, 5, and 6, because it has a 800nm higher
Zn-doped InP layer in the ridge. And since the three times higher P-doping of
PIN12 does not give any significant changes, the best devices are PIN4, PIN5
and PIN6.
Among these three devices the differences are negligible, but theoretically the
standoff layer should give less diffusion of the Zn-doped ions, which should give
less free carrier loss. Furthermore a longer absorption length should give less
recombinations of generated carriers.
Therefore the best devices are PIN4, PIN5 and PIN6, with PIN5 being the one
to choose.

To compare PIN5 with other commercial PDs, table 2.8 lists key parameters
for five other PDs, all of them with an InGaAs active layer, and a PIN junction,
just like PIN5. One of them (U2T) is made with a waveguide, like PIN5, the
others are surface PDs.
The responsivity of PIN5 is some what higher then the other waveguide PD
(U2T), and slightly higher then the surface PDs. This is probably because
a surface PD is not sensitive to the alignment of the optical fiber, whereas a
waveguide PD is very sensitive. The measured responsivity of 1A/W for PIN5
is with a perfectly aligned optical fiber, which has a very high mode overlap of
∼ 95%, whereas the other PDs are packed in a box with the optical fiber welded
or soldered, so the alignment for those PDs are most likely not perfect.
The Bandwidth is smaller than the other waveguide PD, but higher than the
surface PDs. A surface diode will always have a lower bandwidth, because of
the larger light sensitive area which gives a larger capacitance and RC time.
The dark current for PIN5 is among the best. And the sensitivity could not be
found for any of the other devices.
46 CHAPTER 2. BULK PHOTODIODE

Figure 2.21: New diagram of external components, which improves the sensi-
tivity. Compared with figure 2.14 there are here included a capacitor in series
with the load resistor.

2.6.2 Improvements
An obvious improvement is to use an AR coating, which, as we will see in sec-
tion 3.2, can eliminate the reflection at the facet.

The sensitivity can be improved by including a capacitor in series with the


load resistor, see figure 2.21. This will eliminate the large current in the load
resistor, which gave a large shot noise. The shot noise will in this new configu-
ration originate from photo current and dark current for a bit1, and only dark
current for a bit0. By using the equations from section 2.5.2 we can calculate
that the sensitivity should drop to about −14dBm (@ BER = 10−9 ), and that
the shot noise now should be smaller then the thermal noise.

The responsivity should be able to get as high as 1.25A/W , but we only


measure 1A/W , that is a 20% loss. We know the mode overlap gives about 5%
loss (see table 2.4), and loss through the optical fiber gives about 8% loss, so
there are still about 7% loss. As discussed this last loss seems to come from
recombinations of the generated carriers at the grading barriers. The average
barrier height at each step is ∼ 75meV , which is more then the thermal energy
(25meV ). This suggests that the four steps for electrons and holes are indeed
large barriers, which leads to recombinations. So two ways to improve the re-
sponsivity is to use a better optical fiber, that do not have loss through it, and
to make many grading layers, so that each step is small compared to the thermal
energy.

Bandwidth seems to be hard to improve, according to table 2.5 the tansit-


and RC-time are about the same. If we decrease the depletion layer to decrease
2.6. SUMMARY 47

the transit time, we also increase the diode capacitance and vice versa. A free
parameter we can optimize is the bond capacitance, but the bonding pads, which
are 50µm × 50µm, are already difficult to make a good bond on.
48 CHAPTER 2. BULK PHOTODIODE
Chapter 3

Photodiode with integrated


semiconductor optical
amplifier

In this chapter we will be dealing with a PIN WGPD with an integrated semi-
conductor optical amplifier (SOA), where the active region consists of a number
of quantum wells (QW).
The chapter starts with a description of the design of the devices, and a bit
of theory of the QW. Then there are three sections, the first is devoted to the
PD alone (section 3.2), the next to the SOA alone (section 3.3), and the third
to the PD with integrated SOA (section 3.4). Section 3.2 is much like chapter
2, and will therefore be a short treatment of the PDs. Section 3.3 is dealing
with DC and CW measurements on the SOAs, and in section 3.4 we will look at
dynamical measurements on the PD with integrated SOA. The chapter finishes
in section 3.5, with a summary that emphasizes the most important measure-
ments, and suggests some improvements.

3.1 Device design


The devices are made up of a SOA and a PIN WGPD integrated along one
continuous waveguide, which we will call a SOA-PD. The active material in the
waveguide (the same for both SOA and PD) is a heterostructure made up of a
number of QWs. A schematic illustration of the structure, as seen from one of
the facets is shown in figure 3.1.
The Ridge is made by UV-lithography and etching, the width of the ridge is
2µm, and etching of the ridge stops at the ”etch stop” layer (see table 3.1). The
SOA- and PD-section are separated by etching, a few µm wide trench, in the
ridge. This electrically isolates the two sections, so that we can set different bias

49
CHAPTER 3. PHOTODIODE WITH INTEGRATED SOA

3QW 8QW 10QW


Doping Height Doping Height Doping Height Note
[cm−3 ] [nm] [cm−3 ] [nm] [cm−3 ] [nm]
Contact layer
InP:Zn ∼ 1018 1600 InP:Zn ∼ 1018 1000 InP:Zn ∼ 1018 1000 Upper cladding
InP:Zn 1 · 1017 200 InP:Zn 9 · 1017 700 InP:Zn 9 · 1017 700 Upper cladding
Q(1.1):Zn 1 · 1017 10 Q(1.1):Zn 9 · 1017 10 Q(1.1):Zn 9 · 1017 10 Etch stop
InP:Zn 1 · 1017 40 InP:Zn 9 · 1017 200 InP:Zn 9 · 1017 80 Upper cladding
InP - 60 InP - 60 Standoff
Q(1.07) - 200 Q(1.1) - 36 Q(1.13) - 36 SCH
Q(1.29) - 18 SCH
In1−x Gax Asy P1−y In1−x Gax Asy P1−y In1−x Gax Asy P1−y
x=0.530, y=0.86 - 3.7 x=0.507, y=0.86 - 3.65 x=0.507, y=0.845 - 4.5 Upper barrier
x=0.268 , y=0.86 - 7.2 x=0.258 , y=0.86 - 7.2 x=0.258 , y=0.845 - 8 Well × 3, 8 or 10
x=0.530, y=0.86 - 7.3 x=0.507 , y=0.86 - 7.3 x=0.507 , y=0.845 - 9 Barrier × 2, 7 or 9
x=0.530, y=0.86 - 3.7 x=0.507, y=0.86 - 3.65 x=0.507, y=0.845 - 4.5 Lower barrier
Q(1.29) - 18 SCH
Q(1.07) - 200 Q(1.1) - 36 Q(1.13) - 36 SCH
InP:Si 15 · 1017 1000 InP:Si 15 · 1017 1000 InP:Si 15 · 1017 1000 Lower cladding
InP:S ∼ 7 · 1018 3.5·105 InP:S ∼ 7 · 1018 3.5·105 InP:S ∼ 7 · 1018 3.5·105 n-type substrate
Table 3.1: Growth data for the three different SOA-PD devices. Terms like Q(1.1) means a layer of InGaAsP with a bandgap
of 1.1µm, and InP:Zn means a layer of Zn-doped InP. The separate confinement heterostructure (SCH) layers are used to
optimize to optical mode in the waveguide.
50
3.1. DEVICE DESIGN 51

Figure 3.1: Schematic figure of the heterostructure used for the SOA-PD de-
vices’s, with etched ridge. To the right is a zoom-in of the QWs, where there
can be either 3, 8, or 10 QWs. All numbers are approximate. (Figure adopted
from [15])

voltages on them (there are about 2kΩ between the two sections). So the only
difference between the SOA, and the PD, is that the SOA is forward biased,
and the PD is reversed biased.
Three different heterostructures were grown, where the main difference is the
number of QWs; 3, 8, or 10, and the devices are therefore named 3QW, 8QW,
and 10QW SOA-PD. The growth data for the three different heterostructures is
listed table 3.1. The heterostructures were designed to be used as mode locked
lasers [15], thus they are not optimized for a PD with integrated SOA.

3.1.1 Quantum wells


Figure 3.2 shows the structure for the heavy hole (hh) band, light hole (lh) band,
and the electron band, through one of the QWs in a 10QW device (without any
electric field). The active material is strained with respect to the substrate,
with compressive strain in the well material, and tensile strain in the barrier
material. Strain splits up the hh band, and the lh band. Compressive strain
raises the energy of the hh band compared to the lh band, and tensile strain
raises the energy of the lh band compared to the hh band. This makes the QW
deeper for the hh, and for the lh there are no QW in the well material (see figure
3.2).
Bandgap for barrier and well material, the depth of the QWs in the conduction
band, and in the hh band, the effective masses for electrons, and hh’s in well,
and barrier materials, are listed in table 3.2. These parameters are calculated
from the growth data in table 3.1, on the basis of [17], by Kresten Yvind [15].
52 CHAPTER 3. PHOTODIODE WITH INTEGRATED SOA

Figure 3.2: Structure of bandgap through one QW in the 10QW device, without
any electric field. The lowest energy level and corresponding wavefunction for
the electron and hh is also shown. (Part of the figure is adopted from [15])

Parameter Name 3QW 8QW 10QW Unit


QW width LQW 7.2 7.2 8.0 nm
Band gap barrier Eg b 958.1 939.5 956.1 meV
Band gap well Eg w 754.3 746.8 762.4 meV
Conduction band offset ∆Ec/∆Eg 0.447 0.457 0.454
QW depth for electrons Ve 91.1 88.1 88.0 meV
QW depth for heavy holes Vhh 112.6 104.6 105.7 meV
Electron eff. mass in well mwe 0.041 0.040 0.041
hh eff. mass in well mwhh 0.372 0.372 0.373
Electron eff. mass in barrier mbe 0.050 0.049 0.050
hh eff. mass in barrier mbhh 0.374 0.375 0.375
Lowest electron QW energy Ee0 45.0 44.6 40.1 meV
0
Lowest hh QW energy Ehh 12.1 11.9 10.1 meV
Effective energy gap in QW Eg QW 811.4 803.3 812.6 meV
1528 1544 1526 nm

Table 3.2: In the first section of the table are parameters for the for the QWs,
calculated from the growth data, on the basis of [17], by Kresten Yvind [15]. In
the bottom section, the lowest energy level of the electron and hh, and effective
bandgap, are calculated.
3.2. QW PHOTODIODE 53

We now have both the width and depth of the QWs for the electrons in the
conduction band, and the hh’s in the heavy hole band. To find the energy levels
in these QWs we need to solve schrödingers equation in one dimension for a
finite depth QW, it results in the equations (see ex. [7] or [2]):
µ√ ¶
p 2mwe Ee LQW
p
w
2me Ee · tan = 2mbe (Ve − Ee )
~ 2
µ√ ¶ q (3.1)
p w 2mwhh Ehh LQW b (V
2mhh Ehh · tan ~ 2 = 2m hh hh − E hh )

where Ee , and Ehh is the energy eigenvalues of electrons and hh respectively,


~ is Planck’s constant divided by 2π, and all the other parameters are listed
in table 3.2. The first equation is for symmetric modes of an electron in the
conduction band, and the second equation is for symmetric modes of a hh in
the valence band. For antisymmetric modes we just need to include a factor
−π/2 in the tangens parenthesis. For the electrons in the conduction band there
is only one confined state, whereas for the heavy holes the are a few confined
states. But due to orthogonality between the electron and hh wavefunction in
the QWs, the transition matrix element is close to zero for all transitions, except
between states with the same level number. We are therefore only interested in
the lowest symmetric level, and numerical solutions to the two equations 3.1 for
the lowest energy levels are listed in table 3.2. The corresponding wavefunctions
of the lowest electron, and hh state, are shown in figure 3.2.
The effective bandgap of a QW material is the bandgap of the bulk material,
plus the lowest energy level of the electron, and the hh; Eg QW = Egw +Ee0 +Ehh 0
.
This parameter is listed in the bottom of table 3.2, in units of both nm and meV .

3.2 QW Photodiode
In this section we will look at the three different QW PDs without SOA. This
section is similar to chapter 2, and is therefore a short treatment of the QW
PDs, with references to chapter 2 for more thorough explanations.
An important difference compared to the bulk PD from last chapter is that all
these QW PDs are AR coated.

3.2.1 PIN Junction


The I-V curves of the PDs can be seen in figure 3.3, where the topmost solid
red curves are without incident light. We can see that these I-V curves looks
very similar to the I-V curves for the bulk PD.
Starting from the left (at high reverse bias) they have a break down, then a
low dark current region, and at small forward bias there is a sharp increase in
current.
Breakdown occurs at ∼ −30volt for the 3QW PD, and at ∼ −7volt for the 8QW
54 CHAPTER 3. PHOTODIODE WITH INTEGRATED SOA

Figure 3.3: I-V curves for the three different QW PDs, with no optical input
power and with 0.25, 0.75, 1.25, and 1.75mW optical input power. (Adopted
from [15])
3.2. QW PHOTODIODE 55

and 10QW PDs, and since the junction is a PIN junction, the breakdown must
be avalanche breakdown (see section 2.2.2).

The dark current region of the I-V curves should according to equation 2.4
from chapter 2 give about Idark ∼ 0.05nA depending on the length of the waveg-
uide. To measure so low currents are hard, and the measurement equipment
used for the bulk PDs were, at this point in the project, no longer available at
GiGA. So I have only been able to measure the dark current for these devices
with an uncertainty of about 1µA, and for the 3 and 8 QW PD the dark current
was below 1µA, probably also well below 1nA according to the calculation. For
the 10QW PD there is a large ”dark current”, but because it is relatively large
and because it is linear in the bias voltage, it must be a leak current. This leak
current was seen for all the 10QW PDs, and therefore probably originates from
defects in the epitaxial material, or a processing error.
At forward bias these PDs also increases sharply at about 0.6volt to 0.8volt and
equation 2.6 from last chapter fits fine with the measured curves, so the forward
current are also for these QW PDs, a recombination current.

3.2.2 Light absorption


In figure 3.4(a) the structure of the bandgap through the junction of the 8QW
PD with 1 volt reverse bias is shown. The absorption coefficient is given by
equation 2.10, where ρr (E21 ) now is the 2D reduced density of states, which is
a stepfunction. It is zero up to the effective bandgap of the QWs, where it steps
up to some value, and is constant up to the bandgap of the barrier. It only
steps one time because, as discussed above, only one pair of levels are allowed
for transitions.
The matrix element |MT | for a QW depends on the polarization of the light;
TE or TM, and on transition type; heavy hole-conduction (hh-c) or light hole-
conduction (lh-c). Since we only have hh-c transitions, and because the transi-
tion matrix element for the hh-c transition is |MT |2 ' 6m0 for TE polarization,
and |MT |2 = 0 for TM polarization, the devices are very polarizations depended.
If the material were not strained we would also have a confined light hole mode
in the valence band, which has a non zero transition matrix element for TM
polarization, and the PD could therefore be made less polarization dependent.
This polarization dependence is an unwanted effect because it can reduce both
absorption in the PD, but also the amplification in the SOA. We therefore insert
a polarizations controller just before the device, that changes the polarization
of all the incoming light to TE.
The absorption coefficient for QW material is shown in figure 2.7, which sug-
gests that the absorption coefficient, and thereby the photocurrent will depend
strongly on wavelength around 1550nm. In figure 3.5 a measurement of the pho-
tocurrent versus incident wavelength, for the three different devices at constant
photon current, is shown. According to this measurement the effective bandgap
for the 3QW device is at ∼ 1575nm, and for the 8QW, and 10QW devices the
56 CHAPTER 3. PHOTODIODE WITH INTEGRATED SOA

Figure 3.4: (a) Band structure of the 8QW device at 1V reverse bias. The first
few nanometers up until the first red dotted vertical line is the depleted region
in the P-side, and the last few nanometers after the second red dotted vertical
line is the depleted region in the N-side. (b) The same at 1V forward bias.

effective bandgap is located at & 1610nm.


We can also see that the effective bandgap of the 3QW PD, is shifting towards
higher wavelengths when we increase the reverse bias. This shift of bandgap as
function of reverse bias voltage is due to quantum confinement stark effect[18].

Because figure 3.5 suggests that we at 1550nm is passed the step in the ab-
sorption spectrum from figure 2.7, we approximate the absorption coefficient to
α = 1µm−1 at 1550nm, for all three devices.

3.2.3 Waveguide

The waveguide is centered around the QW material, and according to [15] the
barrier and well materials has an average refractive index of 3.45. The numer-
ical computer program Selene is again used to calculate the transverse mode
profile in the waveguide, and also to calculate the confinement factor, effective
refractive index, mode size, and mode overlap with fiber A4-2, which all is listed
in table 3.3.
Since these devices are AR coated we can calculate a responsivity as R =
1.25A/W ηf iber ηmo , which also is listed table 3.3, together with the measured
responsivity.
We see that there is a very fine agreement between the measured, and the cal-
culated responsivity.
3.2. QW PHOTODIODE 57

Figure 3.5: Photocurrent versus incident optical wavelength for the three differ-
ent PDs, at constant photon current Nph = 1016 s−1 . The bias voltage is -2V,
-5V, and -10V for the 3QW device, and -2V for 8QW, and 10QW.

3QW 8QW 10QW Unit


Confinement, Γ 7.3% 16.8% 17.1%
Absorption length 13.7 6.1 5.8 µm
Effective refractive 3.24 3.20 3.21
FWHM transverse 1.49 1.71 1.72 µm
FWHM lateral 0.46 0.41 0.40 µm
1/e transverse 1.25 1.55 1.56 µm
1/e lateral 0.45 0.50 0.49 µm
Mode overlap with Gauss 81% 79% 78%
1/e of gauss Gauss∗ 0.8 1 1 µm
Mode overlap with A4-2; ηmo 84% 83% 82%
Responsivity
Calculated ∗∗ 0.97 0.95 0.94 A/W
Measured with A4-2 0.94 0.95 0.95 A/W

Table 3.3: Calculations of important parameters for the mode in the waveg-
uide, done by the computer program Selene. ∗ Waist of the spherical sym-
metric gauss mode, that gives the maximum mode overlap. ∗∗ calculated as:
1.25A/W ηf iber ηmo
58 CHAPTER 3. PHOTODIODE WITH INTEGRATED SOA

Figure 3.6: Schematic illustration of the integrated SOA-PD device as seen from
above.

3.3 QW SOA
In this section we will look at the SOA part of the integrated SOA-PD. Some
of the measurements in subsection 3.3.1 are performed on the SOA alone, i.e.
without a PD section. The rest of the measurements in this section are per-
formed on the integrated SOA-PD devices, but focuses on the SOA section. The
SOA is made from the same material as the PD (see table 3.1), and the only
difference compared to the PD is that we set forward bias on the SOA. The total
amplification in the SOA will, as we shall see, depend strongly on its length.
Therefore the SOA will be designated with both its number of QWs, but also
with it’s length.
The SOA is an angled facet travelling wave SOA[5], (see figure 3.6). It consists
of a waveguide which is tilted 7o compared to the normal of the facet, and the
facet where the light is injected, is AR coated.

3.3.1 Optical gain in semiconductor material


At the SOA section we apply forward bias, and as we saw in the I-V curves, in
figure 3.3, we get a sharp increase in current at about 0.7volt, which is caused by
recombinations of injected minority carriers. In figure 3.4(b) the band structure
of the 8QW device with 1volt forward bias is shown, and we can see that the
quasi fermi levels are raised so that we have a lot of electrons in the conduction
band, and a lot of holes in the valence band. This situation is called population
inversion. When we send light into such a forward biased diode, we will initiate
stimulated emission, and because there are very few electrons in the valence
band, there will be very little absorption. This causes R21 > R12 from equation
2.8, and the incident light will therefore be amplified.
In equation 2.10 we have the material absorption coefficient α, and the material
gain g is simply given by g(E21 ) = −α(E21 ) [2]. In figure 3.7 the calculated
g(E21 ) as function of the incident photon wavelength, is shown, which is called
3.3. QW SOA 59

Figure 3.7: Calculated material gain versus incident photon wavelength, called
gain spectrum. It is calculated for QW material with an effective bandgap at
1585nm, and at different ∆Ef (energy difference of the two quasi fermi levels).
A Lorentzian lineshape broadening with energy broadening of the energy levels
of 13meV is included [2].

the gain spectrum. The gain spectrum is shown at different ∆Ef , which is the
energy difference between the two quasi Fermi levels, and it is shown for QW
material with an effective bandgap of 1585nm. A lorentz lineshape broadening
with energy broadening of the energy levels of 13meV , is included [2].
The gain of the mode in the waveguide is called the modal gain hgi, and is given
by hgi = Γg, where g is the material gain, and Γ is the confinement from table
3.3.
The difference in the two quasi fermi levels ∆Ef is to a good approximation
equal to the external applied bias, as long as the voltage drop across the junction
is larger then the voltage drop across the P and N side Rpn (Rpn is measured
to about Rpn ∼ 2.5Ω for a 800µm long 8QW SOA).
From figure 3.7 there are three important things to note; 1) as we increase bias
the overall gain increases, 2) as we increase bias the wavelength at which we
have the maximum gain decreases, and 3) there are only a small change from
1.05volt to 2volt applied bias, which is because we at such high ∆Ef have com-
plete inversion of carriers, which means that f1 = 0 (Fermi function for the
valence band), and f2 = 1 (Fermi function for the conduction band), and in-
creasing bias will not change this.
60 CHAPTER 3. PHOTODIODE WITH INTEGRATED SOA

It can be shown that when ∆Ff > Eg the gain spectrum starts to be positive
at some wavelength interval [2].

We have now discussed gain in the SOA, but since the reflections at the
boundaries of the SOA are finite, a photon cavity will be formed. This cavity
will have a high loss of light, both through the AR coated facet, and due to
absorption in the PD section.
At some point, as we increase bias, the gain might become as high as the loss,
and at this point the SOA will starts to lase. Therefore the SOA might turn
into a laser, if the gain becomes as high as the loss, and we shall see later that
this can happen. The point where lasing is initiated is called the threshold level.
Threshold level refers both to a threshold gain gth , a threshold current Ith , and
other parameters at threshold.
Lasing is reached the following way; as we increase bias voltage (or current) the
carrier density in the active region will increase, this will increase ∆Ef , and gain
(see e.g. figure 3.4). If the gain at some applied current reaches a level slightly
larger than the loss, an avalanche process is triggered that amplifies the light
by stimulated emission, called lasing. As the process continues the number of
photons (and optical power) inside the cavity increases, so this process cannot
continue forever. Each stimulated emission process is caused by an electron
hole recombination, and as the number of photons increases so does the rate of
recombinations. This will decrease the number of carriers, which also decreases
the gain.
If we still keep the applied current constant the gain will not decreases below
the level of loss either. Because in that case the number of photons will begin to
decrease, this will decrease the rate of stimulated emission, which will increase
the number carriers, and gain will again increase.
A steady state situation will in this manner be reached, where the gain is fixed
and equal to the loss, and gain can not increase any more. An increase in bias
will still increase the current, which increases the rate of carrier injection, but it
will not increase the carrier density and thereby ∆Ef , and gain. It will instead
increase the rate of stimulated emission, and in this manner keep the carrier
density constant. So, after the threshold level is reached, the optical output
power will be proportional to the applied current.

The threshold level can be seen in figure 3.8(c) for a 900µm long 10QW
SOA, where we increase current, and at some point the optical output power
starts to increase linear with applied current. The laser effect is shown for a
device without AR (just angeled facets), and a device with AR, and we see that
the threshold level is increased as AR is introduced, because AR gives a higher
loss through the facet.
The laser effect can also be seen in figure 3.8(a) where the optical output spec-
trum of a AR coated 10QW SOA (without the PD section) is measured at
different applied currents from 20mA to 70mA. A ∼ 20dB increase in optical
power is seen from 60mA to 70mA, which is caused by the initiation of lasing.
A closeup of the lasing peak is shown in figure 3.8(b), and we see that there
3.3. QW SOA 61

Figure 3.8: (a) Optical output spectrum from an AR coated 10QW SOA, for-
ward biased with 20, 30,...,70mA, (b) a closeup of the lasing peak showing the
Fabry Perot modes, and (c) a plot of Pout vs current showing the threshold level
with and without AR.

are many smaller peaks. Each peak is called a Fabry Perot mode and has a
wavelength λm that satisfies (λm m)/(2 nef f ) = L, where m is an integer, nef f
is the effective refractive index, and L is the length of the cavity.

Before threshold, the output spectrum in figure 3.8(a), is just amplified spon-
taneous emission (ASE), and from these curves we can demonstrate the gain
spectrum. As we increase current the overall gain should increases, which would
give more ASE at all wavelengths, and this effect can clearly be seen in figure
3.8(a). We also see the peak in the ASE is shifting towards larger photon en-
ergies as we increase the current, which matches well with the theoretical gain
spectrum in figure 3.7.
Another way to demonstrate the gain spectrum of the SOA is to send light with
different wavelengths into the integrated SOA-PD, and measure the photocur-
rent of the PD as function of the current on the SOA (ISOA ), and incident
wavelength. In this way the measured current will originate from a background
of ASE, the amplified input signal, and a small leakage current from the SOA.
Such measurements are shown in figure 3.9 for a 3QW device with length of the
SOA LSOA = 800µm, and for two 8QW devices with lengths LSOA = 250µm,
and 800µm. The 3QW SOA-PD, and the LSOA = 250µm long 8QW SOA-PD
cannot lase, they do not have enough gain to compensate for the loss. However
the LSOA = 800µm 8QW SOA-PD will begin to lase at about Ith = 70mA.
In these measurements we again see that the overall gain increases, and that the
gain peak is shifting toward higher photon energy, as ISOA is increased, but we
also see three other effects. (1) In figure 3.9(a) and (b) we have a 3QW device
and a 8QW device that can not lase, and the approximative position of the
bandgap is different. An approximative position of the bandgap is according to
62 CHAPTER 3. PHOTODIODE WITH INTEGRATED SOA

Figure 3.9: The surface plots to the left are measurements of the photocurrent
in the PD as function of the incident wavelength, and current on the SOA,
at constant photon current Nph . The graphs to the right are intersections
of the surface plots at specific SOA currents. (a) 3QW device with LSOA =
800µm, Nph = 8 · 1015 s−1 , and −2volt bias on the PD (b) 8QW device with
LSOA = 250µm, Nph = 8 · 1015 s−1 , and 0volt bias on the PD, and (c) 8QW
device with LSOA = 800µm, Nph = 1 · 1015 s−1 , and −1volt bias on the PD. The
maximum SOA currents are chosen to be as large as possible without destroying
the devices.
3.3. QW SOA 63

the gain spectrum in figure 3.7 at the middle of the increasing slope. The 3QW
device has the bandgap at a lower wavelength than the 8QW device, which we
also saw in the measurements in figure 3.5 of the photo current versus incident
wavelengths.
(2) The next effect is seen from figure 3.9(b) to (c), which both are for the 8QW
device but for two different lengths of the SOA section, the long 800µm SOA-
PD will start to lase at Ith = 70mA, the other will not. For both devices the
gain peak shifts towards higher energy as the gain current increases. However at
about 1550nm the gain peak for the long device do not shift any more, whereas
for the short device the gain peak shifts further. This is because lasing starts
in the long SOA-PD, and as discussed above, gain cannot increase beyond the
threshold level.
(3) The third effect is that the (energy) distance from the approximative po-
sition of the bandgap, to the gain peak, is different for the 3QW device, and
the LSOA = 250µm 8QW device. The distance from the bandgap to gain peak,
in the QW gain spectrum, depend solely on the lineshape broadening function.
If there were no lineshape broadening, the gain peak would coincide with the
bandgap. The lineshape broadening accounts for broadening of the energy lev-
els, which can be caused by temperature, finite lifetime of carriers [2], and in
this case where we have more then one QW it will depend on how precisely alike
the QWs are grown. So the larger lineshape broadening of the 8QW device then
the 3QW device, is probably caused by the higher number of QW, because each
QW are not precisely alike.

3.3.2 Amplifying cavity


For a SOA with length L, and modal gain hgi, we define the (modal) power
amplification as:
G = exp(hgi L) (3.2)
The SOA is intended to amplify the optical signal before it is detected in the
PD, and we would therefore like it to have a high amplification with little noise
at 1550nm. The amplification can be increased by making the SOA longer, and
just by increasing the current to the SOA. But, as discussed in the last section,
when we reach the threshold level, the device will begin to lase, and the gain can
not increase beyond this level. Furthermore, the optical power from the lasing
will also be detected in the PD, which will introduce a lot of noise. Therefore
we would like the SOA to be operated close to threshold, where it has a high
gain g, and therefore also a high amplification G, and no lasing.
The device is shown in figure 3.6, and because we apply different bias voltages
on the SOA, and the PD, we will introduce a small change in refractive index
between the SOA- and PD-section. This will give a small reflection between the
SOA and the PD, which we will denote r2 (field amplitude reflection). The field
amplitude reflection at the AR coated facet, and cleaved facet will be denoted
r1 , and r3 respectively. Therefore two cavities can be formed; in the SOA section
64 CHAPTER 3. PHOTODIODE WITH INTEGRATED SOA

Figure 3.10: I-V curves of the PD section of a 8QW SOA-PD, with LSOA =
800µm, at different currents on the SOA section, and with no incident light.
The slope of the curves at reverse bias is due the 2KΩ between the SOA and
PD section.

alone with r1 and r2 as boundaries, or in both SOA and PD with boundaries r1


and r3 .
The modal threshold gain of the SOA section is given by [2]:
µ ¶
SOA 1 1
hgith = · ln + αint (3.3)
LSOA r1 r2
and the modal threshold gain of the SOA and PD is given by [2]:
µ ¶
SOA−P D LP D 1 1
hgith = Γα + · ln + αint (3.4)
L L r1 r3
Where αint is internal loss in the SOA section, due to effects like free carrier
loss, and absorption. But SOAs has low reflection facets, and internal loss is
therefore negligible[2].
The field amplitude reflection r2 is given by r2 ' ∆n/2nef f , where ∆n is the
change in refractive index from SOA to PD. And according to [19], which con-
siders changes in refractive index due √to carrier injection, we can approximate
the field amplitude reflection by r2 ∼ 2 · 10−3 (where an injected carrier den-
sity of 1018 cm−3 , are used, which seems reasonable according √ to [2] page 170).
The field −3 [15], and
√ amplitude reflections r1 , and r3 are about r1 ∼ 10
r3 ∼ 0.01 [16]. As we will see in section 3.4.2, the devices with the best sensi-
tivity has a SOA length of about LSOA ∼ 500µm and PD length of LP D ∼ 50
to 100µm. By inserting this, and Γ = 17% (8QW or 10QW devices), the modal
SOA−P D
threshold gain of the SOA-PD is gth & 260cm−1 , and for the SOA it is
SOA −1
gth ∼ 130cm . The laser cavity must therefore be in the SOA section.
In figure 3.10 I-V curves of the PD section of an 8QW SOA-PD device (LSOA =
800µm), with different current on the SOA section (there are no incident light),
are shown.
We see that the threshold level is reached at about 60mA current on the SOA
3.3. QW SOA 65

section, and more importantly we also see that the lasing is not suppressed at
reverse bias. Which shows that the device can lase even at reverse bias, proba-
bly originating from the SOA cavity.

3.3.3 Total amplification


Since the cavity in the SOA section starts to lase first, it is equation 3.3 that
determines the threshold gain. And from that equation we can see that the
total (power) amplification G at threshold is given by:
1
Gth = (3.5)
r1 r2
Since the amplitude reflection r1 enters in this equation, and r3 do not, we only
AR coat the incident facet. √ √
As discussed we can approximate r1 ∼ 10−3 [15], and r2 ∼ 2 · 10−3 [19], which
gives a total threshold amplification of about Gth = 700 or 28dB .

Just before the threshold level we have the maximum amplification in the de-
vice without lasing, but not all SOAs can reach the threshold level. For example
a 100µm long SOA with confinement factor Γ = 17% (8QW or 10QW device),
has according to equation 3.3 a material threshold gain of gth = 0.95µm−1 . This
is just about the maximum possible gain at the right wavelength (see figure 3.7),
and it can therefore be hard to reach.
Another effect that can keep the device from reaching threshold, is if Rpn is so
high (more than a few ohms) that the voltage drop across the junction, as we
increase forward bias, becomes small compared to the voltage drop across the
P- and N-side. Because in that case an increase in bias will not increase the
voltage across the junction, but across the P- and N-side, which means that
∆Ef , and gain will not increase.
Both decreasing the threshold gain from equation 3.3, and decreasing the resis-
tance of the P- and N-side, can be done by increasing the length of the SOA.
And increasing the length of the SOA will according to equation 3.5 not affect
the threshold amplification, which we are trying to reach. So, to get close to
threshold, and thereby getting the maximum amplification without lasing, dif-
ferent devices with increasing length of the SOA section, are made.
In figure 3.11 measurements on seven 8QW SOA-PDs with different length of
the SOA, are shown. The graph shows the measured current from the PD, as
function of the applied current on the SOA (where the maximum applied cur-
rent is about the maximum we can apply, without destroying it). Since there
is no incident light in these measurements, all the photocurrent is originating
from ASE and/or lasing generated in the SOA. We see that the first five devices
cannot lase, whereas the devices with the 700µm and 760µm long SOA can.
This means that threshold can be reached for the 8QW SOAs, when it’s length
is somewhere between LSOA = 620µm and LSOA = 700µm.
Similar measurements are performed on the 10QW devices, where threshold can
66 CHAPTER 3. PHOTODIODE WITH INTEGRATED SOA

Figure 3.11: Measured photocurrent in the PD as function of applied current to


the SOA, on seven 8QW devices with different lengths of the SOA. There are
no incident light, and there are zero volt applied to the PD.

be reached when the length of the SOA is somewhere between LSOA = 530µm
and LSOA = 580µm.
The 3QW SOA is amplifying the signal very little; in figure 3.9(a) about 1mW
optical power is send into a 800µm 3QW SOA, and the measured photocurrent
in the PD, which is a combination of ASE and amplified signal, is maximum
2mA. With a responsivity of about 1A/W without the SOA, see table 3.3, this
corresponds to an amplification of less the 2. We will therefore discard the 3QW
SOA-PDs, and only look at the 8QW and 10QW SOA-PD devices. The small
amplification could be due to a low confinement factor, which in turn gives a
low modal amplification. Or if we look at figure 3.9(a) we can see that the gain
peak is relatively sharp defined around 1520nm (not 1550nm).

When we send light into the SOA-PD, the measured photocurrent in the
PD is due to a combination of ASE, and amplified incident signal, and the
responsivity of the SOA-PD is therefore defined as:
Photocurrent in PD due to amplified incident signal
RSOA−P D = (3.6)
Incident optical power
From this definition the amplification of the SOA can be approximated by:

RSOA−P D RSOA−P D
G' ' (3.7)
R 1A/W
3.3. QW SOA 67

where R is the responsivity of the PD alone (without SOA). We can make this
approximation because we still have loss through the optical fiber (ηf iber ), and
even though the mode overlap between the SOA and PD is perfect, we also still
have a mode overlap at the incident facet of the SOA, which is identical to the
mode overlap for the PD alone (except for a small change in refractive index
due to carrier injection [19]). So by measuring the responsivity of the SOA-PD,
we immediately also get the amplification of SOA.
The responsivity of the SOA-PD will of course depend on ISOA , but also on the
optical power of the incident light. Measured amplification of the SOA (respon-
sivity of the SOA-PD), as function of ISOA , and incident optical power, for the
same seven SOA-PD devices as in figure 3.11, is shown in figure 3.12.
Current in the PD due to ASE, and due to the small leakage current from SOA
to PD, is measured (figure 3.11) and subtracted, so figure 3.12 represents the
true amplification of the SOA.

We know, from figure 3.11, that the LSOA = 700µm, and LSOA = 760µm
SOA-PD devices have a current threshold at Ith ' 75mA, and Ith ' 65mA
respectively, whereas the five other devices in figure 3.12 can not lase. We
clearly see that the amplification is increasing with the length of the SOA up
till LSOA = 620µm, after which the amplification decreases.
The LSOA = 620µm device has the largest amplification of about G = 450
(26.5dB), which matches well with the approximated threshold amplification
from above; Gth = 700 (28dB). Similar measurements on the 10QW SOA-PD
devices showed a maximum gain of about 500 (27dB), for a SOA-PD with a
530µm long SOA.
We also see that the amplification depends on the incident optical power. This
is probably because the incident signal to some degree is drowning in ASE. For
the LSOA = 620µm device at ISOA = 150mA there is 3mA photocurrent in
the PD due to ASE (see figure 3.11), which corresponds to about 3mW optical
power. The incident power is only about 1 to 10µW , and at a very narrow
wavelength range around 1550nm, whereas the 3mW optical power from ASE
comes from a wide range of wavelengths, as can be seen in figure 3.8(a).
The decrease in amplification when lasing is initiated is probably because the
incident optical signal now is drowning in optical lasing power. From figure
3.8(a) we know that when lasing is initiated, the optical power at 1550nm in
the SOA is increasing much more than before lasing.

To summarize this section we have looked at the gain spectrum, to try and
locate the wavelength that has the highest gain, we have looked at the transi-
tion from ASE to lasing, to try and find the maximum length the SOA can have
without lasing, and we have looked at the total power amplification versus the
length of the SOA, to find the maximum amplification.
Ideally we would like the peak of the gain spectrum at high currents (total popu-
lation inversion) to be located at 1550nm because this would give us the highest
amplification of the incident 1550nm signal. According to figure 3.9, the gain
peak at high currents of the 8QW SOA seems to be close to the 1550nm; for
68 CHAPTER 3. PHOTODIODE WITH INTEGRATED SOA

Figure 3.12: Amplification of the SOA, or the responsivity of the SOA-PD,


as function of the SOA current, and optical input power. The measurements
are performed on seven different 8QW SOA-PDs, with different lengths of the
SOA. The two last SOA-PDs with the longest SOAs can lase, and the threshold
current are indicated on those figures.
3.3. QW SOA 69

Caption: Same as previous page.


70 CHAPTER 3. PHOTODIODE WITH INTEGRATED SOA

Caption: Same as previous page

a 250µm long 8QW SOA the gain peak is at ∼ 1520nm. We also saw that the
8QW SOA had a broad gain peak, so that the decrease in gain from 1520nm to
1550nm was small. From figure 3.8(a) we saw that a 800µm long 10QW SOA
lased at ∼ 1560nm, so a shorter 10QW SOA that can not lase will probably
also have a gain peak close to 1550nm.
If the SOA-PD starts to lase, it was found that the lasing cavity is in the SOA
section with boundary r1 and r2 , which also is why only the incident facet is
AR coated.
The 8QW SOA will start to lase when its length is somewhere between 620µm
and 700µm, and the 10QW SOA will start to lase when its length is somewhere
between 530µm and 580µm.
The maximum measured amplification is about 450 or 26.5dB for a 620µm long
8QW SOA, and about 500 or 27dB for a 530µm long 10QW SOA.

3.4 SOA-Photodiode
In this section we will look at dynamical measurements, such as bandwidth and
sensitivity, performed on the 8QW and 10QW SOA-PD devices. We have dis-
carded the 3QW devices because it had very little amplification.
For dynamical measurements the PD section are mounted just like the bulk PD,
with a 50Ω load resistor in parallel, and an external bias tee, see figure 2.14.
The SOA section is mounted with a separate lead, and do not need any external
components.
3.4. SOA-PHOTODIODE 71

3.4.1 Bandwidth
From last chapter, we know that the bandwidth is defined as the modulation
frequency, of the incident signal, that decreases the AC responsivity 3dB.
For these SOA-PD devices either the SOA or the PD will set the limit for the
bandwidth. For the PD we know, from last chapter, that the key time parame-
ters are the transit time τtr , and the RC time τRC . For the SOA there is another
key time parameter, namely the time it takes for the carriers to be injected into
the active region τin . In CW measurements τin could set a limitation; if we
send a high power CW beam into the SOA, so that the total recombination
rate becomes higher than the carrier injection rate (1/τin ), we will deplete the
active region of carriers. This will cause the amplification of the SOA to drop,
however such an effect has not been seen.
In dynamical measurements τin could also set a limitation: When we send in
an optical signal which is modulated between a high level and a low level, with
an average power that do not deplete the active region of carriers, there will be
a net decrease of carriers when the optical signal is high, and a net increase of
carriers when the optical signal is low. This is simply because the high optical
signal induces more recombinations, then the low optical signal. This effect
might degrade the signal, and therefore result in a lower bandwidth. But as we
will see in figure 3.13(a), such an effect is not seen either, which suggests that
the carrier injection time τin is smaller than τtr + τRC . The bandwidth of the
SOA-PD devices, is therefore probably limited by the PD.
From last chapter we know that the bandwidth of the PD is given by equation
2.24, from the transit time τtr , and the RC time τRC . Analog to last chapter
we can estimate the τtr , and τRC , and calculate the bandwidth, which is listed
in table 3.4.

The setup for measuring the bandwidth is the same as for the bulk PD from
last chapter (figure 2.16), where we just exchange the bulk PD with a polariza-
tion controller and the SOA-PD device. The bandwidth is measured for both
8QW and 10QW SOA-PD devices, with different lengths of the SOA. In figure
3.13 measurements of the responsivity as function of the modulation frequency
for five 10QW SOA-PD devices with different lengths of the SOA, is shown.
For each device we change one parameter; in figure 3.13(a) we change the aver-
age input power, in (b) and (d) we apply different bias voltages to the PD, and
in (c) and (e) we apply different current to the SOA.
We can see that the bandwidth depend very little on the optical input power,
current to the SOA, and length of the SOA. By applying from 2volt to 5volt
to the PD the bandwidth do not change either, and the decrease in bandwidth
when we apply less then 2volt to the PD is probably due to an increased transit
time in the PD. This all agrees with the bandwidth being limited by the PD,
and not the SOA.
Similar measurements are performed on 8QW devices, and averages of the mea-
sured bandwidths are listed in table 3.4, for comparison with the calculated
bandwidths.
72 CHAPTER 3. PHOTODIODE WITH INTEGRATED SOA

Figure 3.13: Measured responsivity in dB units as function of modulation fre-


quency, for five 10QW SOA-PD devices with different length of the SOA. In (a)
the measurement is performed with different average optical input powers, in
(b) and (d) with different applied voltages to the PD, and in (c) and (e) with
different applied current to the SOA.
3.4. SOA-PHOTODIODE 73

Caption: Same as previous page.


74 CHAPTER 3. PHOTODIODE WITH INTEGRATED SOA

8QW 10QW Unit


Emax 216 180 KV /cm
Depletion region 274 324 nm
Diode capacitance 4.7 4.0 fF
Bond capacitance 30 32 fF
RC time τRC ∗ 1.75 1.81 ps
Transit time τtr 2.74 3.24 ps
Bandwidth of PD
Calculated∗∗ 35.45 31.52 GHz
Measured 30 30 GHz

Table 3.4: Measured and calculated bandwidths, and corresponding estimated


dynamical parameters. Calculations are made at -5 volt bias. ∗ Bond capac-
itance and diode capacitance in parallel multiplied by the 50Ω load resistor.
∗∗
Calculated bandwidth according to equation 2.24.

Averages of the measured bandwidths are ∆f = 30GHz for both the 8QW and
10QW SOA-PDs, which matches quite well with the calculated bandwidths.

3.4.2 Sensitivity
The sensitivity is, as discussed in the last chapter, the optical input power
where the BER reaches some level, and in this project this level is defined to be
BER = 10−9 .
The sensitivity will increase when the eye opening increases and/or when the
noise of the bit1’s, and bit0’s decreases. Noise in the PD is due to thermal noise,
and shot noise, as discussed in section 2.5.2. But in these SOA-PD devices, the
SOA section is also contributing to the noise, where the noise originates from
ASE.
So the SOA amplifies the incident signal, which opens the eye, but it also gen-
erates ASE, which induces more noise and smears the eye. We therefore want
the SOA to have a high amplification to ASE ratio, which will depend on the
length of the SOA, and the applied SOA current.
An electrical eye diagram of a 10QW SOA-PD device is shown on the front page
of this thesis, where a short PRBS of 107 − 1 is used. Every measurement point
in such an eye diagram is from a different bit, it therefore contains bits with
many different histories, and by looking at the front page we can actually see
that the eye diagram is composed of bits with different histories. Especially the
bit0 region of the eye diagram is composed of several lines, which each stems
from a different bit-history.

The measurement setup to measure the sensitivity is very similar to the one
used for the bulk PD in figure 2.18, the only difference is that we have exchanged
3.4. SOA-PHOTODIODE 75

the PD with a polarization controller and the SOA-PD device. For all the sen-
sitivity measurements, we again use B = 10Gbit/s, a PRBS of 1031 − 1, and
NRZ format.
In figure 3.14 measured BER versus average optical input power for five 10QW
SOA-PD devices with different length of the SOA, is shown. Similar measure-
ments for five 8QW devices has been performed, and the sensitivity versus the
length of the SOA section is shown in figure 3.15(a). The sensitivity will depend
on the current to the SOA, and in figure 3.15(b) the SOA-current that gives the
highest sensitivity is plotted versus the length of the SOA.

In figure 3.15(a), we see that the SOA-PD devices has an optimum length
of the SOA; for the 10QW SOA-PD it is at LSOA ∼ 475µm, and for the 8QW
devices it is at LSOA ∼ 565µm. This optimum length of the SOA can be ex-
plained as the point where the amplification to ASE ratio is highest. Because
by making the SOA longer, the ASE would then increase more than amplifica-
tion, and by making the SOA shorter the amplification would begin to decrease
rapidly, which both corresponds to a more closed eye diagram, and in turn a
worse sensitivity.
We know from section 3.3.3 that lasing can start when the length of the SOA
is at 620µm to 700µm for the 8QW SOA-PDs, and at 530µm to 580µm for
the 10QW SOA-PDs. So, this suggests that the optimum sensitivity is reached
somewhat before threshold. This could be because ASE begins to increase more
rapidly as we approach threshold, and that the highest amplification to ASE
ratio therefore is reached somewhat before threshold.
The best 8QW SOA-PD device has a sensitivity of ∼ −24dBm, and a SOA
length of LSOA ∼ 565µm. According to 3.15(b) the best sensitivity is reached at
ISOA ∼ 130mA, and according to figure 3.12 it has a max amplification of ∼ 300
(24.8dB), and at the maximum sensitivity (Pin = −24dBm, ISOA = 130mA) it
has an amplification of ∼ 250 (24.0dB).
The best 10QW SOA-PD device has a sensitivity of ∼ −20dBm, and a SOA
length of LSOA ∼ 475µm. The best sensitivity is reached at ISOA ∼ 125mA,
and it has a max amplification of ∼ 350 (25.4dB), and at the maximum sensitiv-
ity (Pin = −20dBm, ISOA = 125mA) it has an amplification of ∼ 300 (24.8dB).

The threshold gain can be calculated by equation 3.3; the 565µm long 8QW
SOA has a modal threshold gain of about hgiSOA
th ∼ 120cm−1 , and the 475µm
SOA −1
long 10QW SOA has hgith ∼ 140cm . We know that both of them can
not lase which means that they can not reach a gain this high. The maxi-
mum modal gain can be calculated by hgimax = ln(Gmax )/L, where Gmax is
the maximum measured amplification. The 565µm long 8QW SOA has a max
modal gain of about hgimax ∼ 100cm−1 , and the 475µm long 10QW SOA has
hgimax ∼ 125cm−1 .

Figure 3.7 shows that at total population inversion we have a material gain
of about 10000cm−1 , and with a confinement factor of 17% it gives a modal
gain of 1700cm−1 , which is more that 10 times larger than the maximum modal
76 CHAPTER 3. PHOTODIODE WITH INTEGRATED SOA

Figure 3.14: Measurements of BER versus average optical input power at a


PRBS of 1031 − 1, and in the NRZ format, for five 10QW SOA-PD devices with
different length of the SOA.
3.5. SUMMARY AND DISCUSSION 77

Figure 3.15: Measurements on five 10QW SOA-PD and five 8QW SOA-PD
devices. In (a) sensitivity versus length of SOA, and in (b) current to the SOA
that gives the best sensitivity versus length of SOA.

gains that we have just calculated. This indicates that the SOAs are far from
total population inversion.

3.5 Summary and discussion


To summarize this chapter, the most important parameters for the 8QW and
10QW SOA-PD devices, are listed i table 3.5. The 3QW SOA-PD device has
been discarded because of the low amplification.
The 8QW and 10QW SOA-PD devices are very similar, which would be ex-
pected because the only difference is the number, depth, and width of the QWs
(see table 3.1). We saw in figure 3.3 that the 10QW PD had some leakage
current. Besides that, the largest difference between the two is the sensitivity,
where the 8QW SOA-PD is 4dB better. Another difference is the amplification,
where the 10QW SOA-PD has about ∼ 20% more amplification.
The amplification of the SOA is intended to give the SOA-PD a large dynamic
range of optical input powers, so that it is capable of working in many differ-
ent environments with different optical powers. If the optical power of some
bit stream in an environment is low, say −15dBm, we just apply a high SOA-
current, and if the optical power i relative high, say 0dBm, we just apply a low
SOA-current. The sensitivity is therefore a more important parameter than the
amplification, and the 8QW SOA-PD device is therefore best.

I will here suggest a few improvements: The first and most obvious improve-
ment, is to make device polarization independent. E.g. just by using a bulk
active layer, instead of QWs, because bulk material will absorb TE and TM
polarization equally.
Another improvement is to make an even better AR coat, which would increase
78 CHAPTER 3. PHOTODIODE WITH INTEGRATED SOA

Device characteristics
SOA-PD 8QW 10QW Unit
Bandwidth 30 30 GHz
Sensitivity -24 -20 dBm
@ current to SOA 130 125 mA
Max Responsivity 300 350 A/W
Responsivity @ max sensitivity 250 300 A/W
SOA
Length of SOA 565 475 µm
Max amplification 300 350
Amplification @ max sensitivity 250 300
Max modal gain 100 125 cm−1
Threshold modal gain ∼ 120 ∼ 140 cm−1
gain peak 1520 - nm
Contact resistance ∼3 ∼3 Ω
PD
Length of PD 50–100 50–100 µm
Bias voltage 2–5 2–5 volt
Responsivity 0.95 0.95 A/W
Breakdown voltage 7 7 volt
Dark current (< 1) - nA
Contact resistance ∼ 20 ∼ 20 Ω
Capacitance 35 36 fF
Absorption coefficient 1 1 µm−1

Mode characteristics
8QW 10QW Unit
Mode overlap with Gauss 79% 78%
1/e lateral 1.55 1.56 µm
1/e transverse 0.50 0.49 µm
Confinement factor 16.8% 17.1%
Effective refractive index 3.20 3.21

Table 3.5: Data sheet for the 8QW and 10QW SOA-PD devices. The table is
sectioned in device characteristics of the SOA-PD, SOA, and PD device, and
characteristics of the mode in the waveguide.
3.5. SUMMARY AND DISCUSSION 79

the threshold amplification. This could be used to make the SOA amplify more,
and in turn also give a better sensitivity.
We know that the bulk PD, from last chapter, had mode overlaps up to 95%,
therefore a ∼ 10% improvement of the mode overlap for these SOA-PDs could
probably also be obtained.
The bandwidth can not be improved much since the transit time, and the RC
time, are relative close (see table 3.4).
We have seen that the peak in the gain spectrum for a 8QW SOA with LSOA =
250µm is located at about ∼ 1520nm. By decreasing the effective bandgap a
bit, the gain peak could be shifted to 1550nm, which would increase the ampli-
fication. Since the noise originating from ASE, not would be affected much by
such a change, the sensitivity would probably also be better.

I will here suggest a few changes that might improve the SOA-PDs:
(1) An EDFA could be used as an external optical preamplifier, instead of the
SOA. Such a setup could probably give even better sensitivities, because it takes
advantages of the large gain of the EDFA (typically > 30dB). The disadvantage
of such a setup is that it is much more expensive, and complicated, than the
integrated SOA.
(2) We saw that the SOA is far from total population inversion. If the SOA
could work at total population inversion we would have a much higher modal
gain, and could therefore make the the SOA much shorter and still have the
same amplification. The advantage is that a short SOA with high gain might
have less ASE then a long SOA with low gain (when they both are operated
close to threshold, and have the same AR coating).
(3) To insert a small bandwidth optical filter between the SOA, and the PD,
that only transmits wavelengths close to 1550nm. This would reduce the noise,
because only ASE from from 1550nm would be detected. The disadvantage
with such a filter is that the reflection between the SOA, and PD (r2 ), will
be close to one, and the threshold amplification therefore will drop to about
Gth = (r1 )−0.5 . So such a filter might give a better sensitivity, but a smaller
amplification.
80 CHAPTER 3. PHOTODIODE WITH INTEGRATED SOA
Chapter 4

Conclusion

We have in this thesis characterized and optimized two different types of PDs
for use in an optical fiber communication system. The first type was a PIN
WGPD with a bulk absorbing layer of InGaAs, it was designed by GiGA-Intel
and a total of 22 different designs were made (eleven different heterostructures
all with two different ridges).
The second PD was a PIN WGPD with a SOA that was integrated along one
continues waveguide, with an active layer of InGaAsP QWs. Three different
heterostructures was characterized, each with a number of different lengths of
the SOA, for the optimization part.

The optimization part for bulk PD was to optimize its responsivity from dif-
ferent optical fibers. About ten different optical fibers, from GiGA and COM,
were used for this optimization, where three of them were shown in figure 2.13.
From measurements of the farfield of the optical fibers, and numerical calcula-
tions of the transverse mode profile in the waveguide, a maximum mode overlap
of about ∼ 95%, was achieved for the optical fiber A4-2, and PIN5. Further-
more, a transmission through this optical fiber of 92% was measured. These
coefficients seemed to match well with the measured responsivities of just above
1A/W (including AR coating).
The characterization of the bulk PDs involved measuring the responsivity, band-
width, dark current, bias voltage range, breakdown voltage, contact resistance,
and sensitivity (with an external electrical amplifier), which all were collected
in the data sheet in table 2.7. This table is for the devices with a 2µm ridge,
because we saw that the devices with a 2µm ridge generally had a higher re-
sponsivity than the corresponding devices with a 1.5µm ridge. From table 2.7,
it was concluded the best device was PIN5 (with a 2µm ridge).
We can see from table 2.7 that all the requirements GiGA-Intel made on their
PD (see section 1.2), were actually met by PIN5. Especially the bandwidth,
which is about three times larger then the requirement.
The bulk PD was, as discussed in the introduction, designed for a receiver unit
by GiGA-Intel. The best receiver unit that GiGA-Intel has made uses PIN5,

81
82 CHAPTER 4. CONCLUSION

and measurements performed by GiGA shows that this receiver has a maximum
sensitivity of about −19dBm.

The optimization part of the SOA-PDs was to optimize the length of the
SOAs, so that the SOA-PDs would give the maximum sensitivity. This was
performed for the 8QW and 10QW SOA-PDs, and the optimized length of the
SOAs were found to be LSOA = 565µm, and LSOA = 475µm respectively.
With these optimized lengths of the SOAs, the 8QW and 10QW SOA-PD gave
sensitivities of ∼ −20dBm, and ∼ −24dBm respectively.
The characterization part was performed on these two devices, and it involved
measuring quantities like sensitivity, bandwidth, responsivity, amplification, and
gain peak, which all were listed in the data sheet in table 3.5.
From table 3.5 we concluded, that the best device was the 8QW SOA-PD (with
LSOA = 565µm).
The biggest advantage of this SOA-PD is its high sensitivity of about −24dBm
(4µW ), which gives it a large dynamical range of optical input powers. This
SOA-PD actually has about 5dB better sensitivity than the PIN5-based receiver
that GiGA-Intel has designed.
Abbreviations
AC Alternating current
APD Avalanche photodiode
AR Anti reflection
ASE Amplified spontaneous emission
BCB Benzocyclobutene polymer
BER Bit error rate
CW Continuous wave
DC Direct current
EA Electro absorber
EAM Electro absorptions modulator
EDFA Erbium doped optical fiber amplifier
EHP Electron hole pair
FWHM Full width half maximum
hh heavy hole
hh-c heavy hole-conduction band (transition)
I-V Current versus voltage
lh light hole
lh-c light hole-conduction band (transition)
NRZ Non return to zero
OTDM Optical time division multiplexing
PD Photodiode
PRBS pseudo random bit sequence
QW Quantum well
rf Radio frequency
RZ Return to zero
SCH Separate confinement heterostructure
SOA Semiconductor optical amplifier
SOA-PD Photodiode with integrated semiconductor optical amplifier
TE Transverse Electric
TM Transverse magnetic
UV Ultra violet
WDM Wavelength division multiplexing
WGPD Waveguide photodiode

83
84 ABBREVIATIONS
Bibliography

[1] Charles Kittel (1996); Introduction to solid state physics.

[2] Larry A. Coldren and Scott W. Corzine (1995); Diode lasers and photonic
integrated circuits.

[3] Sze, M.S. (2001); Semiconductor devices, Physics and technology.

[4] Streetman B.G. (1995); Solid state electronic devices.

[5] Agrawal P. Govind (2002); Fiber-optic communication systems.

[6] Griffiths David J. (1999); Introduction to electrodynamics.

[7] Davis J.H, (1998); An introduction the the physics of low dimensional
semiconductors.

[8] Sajal. P, et al. J. Appl. Phys. 69 827 (1991); Empirical expressions for
the alloy composition and temperature dependence of the band gap and
intrinsic carrier density.

[9] Yater J.A, et al IEEE 2 1709 (1994); Minority carrier lifetime in InP as
function of light bias.

[10] Chelli C, et al. IEEE 127 (1999); Minority carrier lifetime in MOCVD-
grown C- and Zn-doped InGaAs.

[11] http://www.ioffe.rssi.ru/SVA/NSM/Semicond/

[12] T Gonzalez, et al. semicond. Sci. Tech 7 31 (1992); Electron transport in


InP under high electric field conditions.

[13] Thobel J.L, et al Appl. Phys. Lett. 56 346 (1990); Electron transport in
strained InGaAs.

[14] Vanhollebeke K. M, et al J. Crystal Growth 233 132-140 (2001); MOVPE


based Zn diffusion into InP and InAsP/InP heterostructures.

[15] Kresten Yvind, Phd. Thesis (2003); Semiconductor mode-locked lasers for
optical communication systems.

85
86 BIBLIOGRAPHY

[16] Assistant Professor, Kresten Yvind (COM). Personal communication.


[17] Vurgaftman I, et al, J. Appl. Phys. Rev. 89 5815 (2001); Band parameters
for III–V compound semiconductor and their alloys.
[18] Sune Højfeldt, Phd. Thesis (2002): Modeling of carrier dynamics in electro
absorption modulators.
[19] Bennett B. R, et al IEEE 26 113-122 (1990); Carrier-induced change in
refractive index of InP, GaAs, and InGaAsP.

S-ar putea să vă placă și