Integrated Color Sensor in Standard CMOS
Integrated Color Sensor in Standard CMOS
net/publication/343736867
CITATIONS READS
0 1,658
1 author:
Graciele Batistell
Infineon Technologies
20 PUBLICATIONS 50 CITATIONS
SEE PROFILE
All content following this page was uploaded by Graciele Batistell on 19 August 2020.
MASTER THESIS
• the Master Thesis has been written by myself without any external unauthorized
help and that it has not been submitted to any institution to achieve an academic
grading.
• I have not used sources or means without citing them in the text; any thoughts from
others or literal quotations are clearly marked.
• the enclosed Master Thesis is the same version as the version evaluated by the
supervisors.
• one copy of the Master Thesis is deposited and made available in the CUAS library
(§8 Austrian Copyright Law [UrhG]).
I fully understand that I am responsible myself for the application for a patent, trademark
or ornamental design and that I have to prosecute any resultant claims by myself.
ii
Abstract
Optoelectronic Integrated Circuits (OEICs) have several commercial and industrial ap-
plications like data storage, communication, and color sensing. The tremendous growth
of the mass market applications like Complementary metal-oxide-semiconductor (CMOS)
cameras, Liquid-crystal display (LCD) displays, and mobile devices increased significantly
the demand of low cost integrated sensors. This project is set out to research new inte-
grated color sensor structures as a low cost alternative solution for color sensing applica-
tions. The proposed structures are designed in standard CMOS technology with no color
filters or process modifications. The proposed color sensor structures consist of vertically
and laterally arranged photodiodes, providing color separation based on lateral carrier
diffusion and wavelength-dependent absorption-depth. Furthermore, tunable spectral re-
sponses are obtained by the use of the field effect, and by applying external electrical fields.
The color sensor structures have three photodiodes with independent spectral responses
optimized to short, medium, and long wavelengths. Nevertheless the sensor responsivities
are not ideal and a linear transformation is used to match the sensor responsivities to
XYZ or Red, Green and Blue (RGB) color standards. The process simulations and device
simulations of the optical and electrical characteristics of the color sensors were performed
with Synopsys Technology Computer Aided Design (TCAD) tools. The structures were
laid out and produced in a standard 130 nm CMOS technology. Based on measurement
results a digital signal transformation is proposed to enable the color representation in
RGB and XYZ standards.
Key words: Color sensor, CMOS technology, Photodiodes, Integrated sensor, color
matching.
iii
Acknowledgement
This work was developed at the Carinthia University of Applied Sciences as part of the
COSMOS project under the supervision of Mr. Johannes Sturm. I would like to thank
all the colleagues from COSMOS project for the contributions during the development of
the project. This thesis would not have been possible without the guidance, help, and
support of my first supervisor Mr. Johannes Sturm, for which I am extremely grateful.
I also would like to thank the support of my second supervisor, Mr. Michael Köberle. and
express my gratitude to the teachers of the ISCD master program for the effort expended
and for the continuous help in our learning process.
I am very grateful to my friends, family and specially Georg for encouragement, attention
and patience during the master program and development of the thesis.
iv
Contents
List of Figures x
List of Tables xi
1 Introduction 1
2 Theoretical background 3
2.1 Light and Color . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1.1 Light Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.1.2 Transmission and reflection . . . . . . . . . . . . . . . . . . . . . . 4
2.1.3 Light absorption and optical generation . . . . . . . . . . . . . . . . 7
2.1.4 Color perception and representation . . . . . . . . . . . . . . . . . . 9
2.2 Photodiodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.1 P-n Junction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.2 Drift current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.3 Diffusion current . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.4 Drift-diffusion transport Model . . . . . . . . . . . . . . . . . . . . 19
2.2.5 Photodiodes and light interaction . . . . . . . . . . . . . . . . . . . 20
2.2.6 Quantum efficiency and responsivity . . . . . . . . . . . . . . . . . 22
2.3 CMOS process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3.1 CMOS technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.2 Fabrication Process . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.3 CMOS process-flow . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.4 Photodiodes in standard CMOS . . . . . . . . . . . . . . . . . . . . 27
2.4 Color Sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
v
2.4.1 Double photodiode . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4.2 Lateral photodiode . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.4.3 Tunable spectral response . . . . . . . . . . . . . . . . . . . . . . . 33
4 Implementation 50
4.1 Structure 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.1.1 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.1.2 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.2 Structure 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.2.1 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.2.2 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.3 Structure 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.3.1 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.3.2 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.4 Structure 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.4.1 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.4.2 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.5 Results summary and analysis . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.6 Measurement Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
vi
5 Future outlook 71
5.1 Modeling of the oxide/nitride stack interference . . . . . . . . . . . . . . . 71
5.2 Color Signal processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6 Conclusion 75
Bibliography 77
List of Abbreviations 80
List of Symbols 82
Appendix 84
A Structure 1 Layout 84
B Structure 2 Layout 85
vii
List of Figures
viii
2.23 Illustration of the lateral carrier diffusion for short- (a) and long-wavelength
light (b). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.24 Effect of the lateral carrier diffusion on the photodiodes responses . . . . . 33
2.25 Influence of electrical field on the lateral carrier diffusion. . . . . . . . . . . 34
2.26 Influence of the field effect on the lateral carrier diffusion. . . . . . . . . . . 35
ix
4.16 Measured responsivity of Structure 3 for PP voltage steps of 225 mV. . . . 61
4.17 Cross-section of Structure 4. . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.18 Electrons Current Density of Structure 4 for different electrode voltages. . 63
4.19 Simulated Responsivity of Structure 4 for different electrode voltages. . . . 64
4.20 Measured responsivity of Structure 4, in comparison to simulation, for dif-
ferent electrode voltages. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.21 Measured responsivity of Structure 4 for electrode voltage steps of 25 mV. . 65
4.22 Picture of the Testchip. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.23 Illustration of the optical measurement setup. . . . . . . . . . . . . . . . . 67
4.24 Picture of the measurement setup. . . . . . . . . . . . . . . . . . . . . . . . 70
5.1 (a) Simulated responsivity of Structure 1 with Raytracer optical solver. (b)
Measured responsivity of Structure 1 with interference caused by oxide layers. 72
5.2 Measured responsivity of Structure 1 in comparison to simulation with
BPM optical solver. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.3 Use of transformation matrix to match the output of the produced color
sensor to the RBG tristimulus for a unique input. . . . . . . . . . . . . . . 73
5.4 Use of training patterns to generate a transformation matrix suitable for
different light spectra. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
x
List of Tables
xi
Chapter 1
Introduction
Optoelectronic devices are based on the interaction between photons and electrons. Two
processes of energy conversion result form this interaction, the emission and the absorp-
tion. Light emission is observed for example in Light Emission Diodes (LEDs) when the
electrical energy is converted in optical radiation, while the absorption is the conversion
of optical radiation into electrical energy present in photodetectors and solar cells. The
behavior of optoelectronic devices is defined by the photoelectric effect, observed first time
in 1887 by Heinrich Hertz and mathematically described by Albert Einstein in 1905.
Even-though the concept of integrated circuits exist since 1950’s, only in the past few
years commercial applications of integrated optoelectronic devices are emerging. The first
image sensors, Charge-coupled devices (CCDs) as well as CMOS sensors, were developed
in parallel in the early 70’s. But due to limitations on the CMOS fabrication process the
CMOS sensor just came to the market after 25 years. The improvements on the CMOS
process enabled the production of CMOS sensors with compatible quality to the CCDs
and advantages of integrated circuits like lower production cost, miniaturization, and less
sensitivity to noise. The monolithic integration of color sensors enables the photodetector,
read out circuit and digital processing to be integrated on the same substrate. Due to this
advantages OEICs are suitable for optical communication, data storage and color sensing
applications.
As the use of OEICs increases, also new applications and requirements for the market arise.
With the growth in the use of mobile devices during past few years, an efficient method
for integrated, low-cost color sensing gets essential for the mass market application. Other
possible fields of application are the ambient light sensing for general lighting, and screen
back-light adaptive control depending on the daylight or to compensate aging effects. The
use of integrated color sensors is not restricted to commercial applications but there are
many industries in which color sensing plays an integral role. As an example, where color
is an integral part of the end product, manufacturers can utilize color sensor-based systems
1
to reduce their time-to-market, improve product quality through color consistency and
increase efficiency through reduced scrap.
The current available solutions for color sensing are normally based on Color Filter Arrays
(CFA) or stacked photodiodes which require expensive technologies or process modifica-
tion. Mass applications are basically targeted for a consumer oriented market, therefore a
highly integrated color sensing implementation based on a standard low-cost CMOS pro-
cess is a key requirement for future-oriented and cost-effective system integration. Thus,
a monolithically integrated low-cost color sensor is developed. The sensor is based on
standard CMOS technology without costly process modifications or any external color
filter structure fully compatible with mass market applications.
The color sensor research is developed as part of the COSMOS project, a founded project
supported by the Österreichische Forschungsförderungsgesellschaft mbH (FFG). The
COSMOS project consists of the research of three different blocks of the integrated color
sensor: The color detector that provides three independent sensor signals approximated
to RGB; The analog to digital converter with high dynamic range optimized for low-
power and minimum area; And the digital processing block that realizes standardized
colorimetric RGB or XYZ color space responses based on a linear transformation of the
three non-ideal color detector signals. This work deals with the research, simulations,
development, and characterization of the color detector structures and briefly describes
the used transformation algorithm.
In this thesis the theory, the development and the analysis of the produced color sensing
structures are taken into account. In chapter 2 the theory of light and its interaction
with semiconductors is introduced. Furthermore the photodiodes basic operation and its
interaction with light is discussed. Additionally the CMOS process and the photodiodes
production is presented, followed by the proposed techniques of color sensing in standard
CMOS technology. In chapter 3 the methodology used for investigation and verification
of the proposed color sensors is presented. The device production, electrical and optical
simulations are performed using the Synopsys TCAD tools, that models the semiconduc-
tor devices fabrication and operation. Information about the used tools and models is
provided. Furthermore, in chapter 4 the proposed color sensor structures, the respective
working principles, simulations and measurement results are presented. Four developed
structures based on lateral- and double photodiode arrangements are presented. The
structures provide photodiodes sensitive to short-, middle- and long wavelengths and tun-
able spectral responses according to external applied voltages. The work is finalized by
chapter 5, where future steps and further research topics related to this thesis are pro-
posed. The investigation of the oxide stack interference is suggested by using an advanced
optical solver. Additionally, the necessary signal transformation, to match the output of
the color sensors to the standard XYZ color space responses is discussed. The use of
digital signal processing based in a transformation matrix is proposed.
2
Chapter 2
Theoretical background
When light falls in an object surface, part of this light is reflected and part enters the
material. The part of the light that goes into the material can be absorbed or transmitted.
This interaction between light and material creates our perception of colorful objects and
explains how semiconductor can be used for light sensing. The semiconductor material, for
example silicon, when exposed to light will absorb the light and generate carriers. When
these carriers reach a p-n junction of a photodiode, they are separated and photocurrent is
generated. The photodiodes characteristics and the separation of the carriers are strongly
related to the CMOS technology production. Since the depth of the light absorption and
carriers generation is wavelength dependent, carriers from different wavelengths can be
collected according to the photodiodes arrangement and its p-n junction depth. This
provides some color distinction between photodiodes.
In this chapter the theory of light as wave and particle, the effects of reflection, transmis-
sion, light absorption, and carriers generation is introduced. Additionally the photodiodes
and the influence of the p-junction depth on the color detection is presented. Furthermore
the CMOS technology and the process flow is discussed. The CMOS process defines the
standard p-n junctions available for color sensing and it is an important topic on the
modeling and simulation of the photodiodes. The chapter is finalized by the presentation
of the techniques of color sensing in standard CMOS technology developed in this work.
3
2.1.1 Light Theory
A photon is the smallest unit of an optical signal. The relation between the light as a
electromagnetic wave and as a photon is given by the Planck’s constant h:
In geometrical optics the light propagation can be described as rays. This description
is a simplification used to model how light propagates in a medium. The light rays
travel in a homogeneous medium approximately in a straight line and when it reaches the
4
Figure 2.1: Electromagnetic Spectrum [6].
5
Figure 2.2 shows the light traveling trough the Material 1, for example air, until it reaches
the boundary of the Material 2, for example silicon, with an incident power PT ot and an
angle θ1 . Part of the incident light is reflected with the same angle of the incident light
Tot Refl
Trans
and with power Pref l . The material parameter that defines the light reflection is the
reflection coefficient R̃. R̃ is defined as the ratio of the reflected power and the incident
power at the surface of a material. R̃ is dependent on the refractive index and on the
extinction coefficient of the material, as shown in Equation 2.6.
(1 − ñ)2 + k 2
R̃ = (2.6)
(1 + ñ)2 + k 2
The light reflection at photodiodes surface is an unwanted effect because it decreases
the photodiode light sensitivity. Moreover the photodiode surface is not directly at the
interface with air, but is covered by a stack of oxide and nitride layers. The oxide/nitride
stack creates a complex optical system where effects as reflection, refraction and diffraction
on the different layers must be considered. The effect of the oxide/nitride layers on a color
sensor is briefly discussed in Chapter 5.1.
Since the reflection on the material is wavelength dependent, light with some specific
wavelengths will be reflected, defining the color of the material, while light with other
wavelengths penetrate the material. The portion of the light that enters the material
can be bended (refracted) at the interface of the two materials in case they have different
refractive indexes ñ. The relation between the angles of incidence and refraction is defined
by Snell’s law, as presented in Equation 2.7.
ñ1
sinθ2 = · sinθ1 , (2.7)
ñ2
where ñ1 and ñ2 are the refractive indexes of the two materials, θ1 is the angle of the
incident light and θ2 is the angle of the refraction. The refraction causes no power loss
but just reduction of velocity resulting in the bending of the rays [7].
6
The portion of the light that enters the material without being absorbed is transmitted
through the material. The ratio between the transmitted power PT rans and the incident
power PT ot is defined as transmission coefficient T . The transmission coefficient depends
on the absorption coefficient(α), on the reflection coefficient (R̃) and the depth of the
material (y) [5].
T = (1 − R̃)2 · exp(−αy) (2.8)
In transparent materials all the light that entered the material is transmitted, in opaque
material part of it is absorbed [5].
Even though several semiconductors or alloys are used in integrated circuits applications,
silicon is by far the most important. Silicon has a relatively low cost, more simple process-
ing and the wider range of operating temperature in comparison to other semiconductors.
This characteristics make silicon the first choice to the current market of electronic devices.
The optical as well as the electrical properties of semiconductors are defined by its energy-
band structure. When light is falling on a semiconductor surface, the energy of a photon
can be transferred to an electron in the valence band which is therefore excited to the
conduction band.
In case the energy of the photon is larger than the band-gap energy of the electron it
is absorbed. An electron-hole pair is generated and consequently a photocurrent flows.
Photons with energy smaller than the bandgap energy of the silicon cannot be absorbed
and the silicon is transparent for them. For the known energy bandgap of the silicon
of 1.11 eV at room temperature and from Equation 2.3 it is possible to calculate the
wavelengths absorbed and transmitted by silicon. Silicon is transparent for wavelengths
longer than approximately 1100 nm, while light with shorter wavelengths ii absorbed.
The absorption of carriers in a semiconductor is given by it absorption coefficient α.
The absorption coefficient is one important value for light sensing applications and it is
dependent on the extinction coefficient k of the material and on the wavelength of light,
as shown in Equation 2.9.
4πk
α= (2.9)
λ
Figure 2.3 presents the absorption coefficient in dependence on the wavelength for ma-
terials used in photodetectors production. The solid and the dashed line represent mea-
surements at 300 k and 77 k. For some of the materials α increases with temperature
while for others it decreases, depending on how the bandgap of the material if affected by
temperature changes.
7
Figure 2.3: Absorption coefficient of semiconductor materials at 300 K and 77 K [6].
When light enters the silicon, depending on its absorption coefficient, the photons are ab-
sorbed, consequently causing the optical intensity to decrease. According to the Lambert-
Beer’s law, the optical intensity I(y) decreases exponentially when entering the silicon,
as presented in Equation 2.10.
As well as the optical intensity I(y), the optical power P (y) decreases exponentially when
entering silicon, as shown in Figure 2.4. P (y) at a specific depth y can be calculated as
shown in Equation 2.11, where PT ot is the power of the incident light neglecting reflection
effects and α the absorption coefficient.
4πk
P (y) = PT ot · exp(−αy) = PT ot · exp(− y) (2.11)
λ
The light energy is absorbed when entering silicon, leading to the generation of electron-
hole pairs. Assuming that each absorbed photon generates one electron-hole pair, the
number of generated pairs per volume per time, called generation rate G is given by:
dP (y) 1 d(PT ot · exp(−αy)) 1
G(y) = = (2.12)
dy Ahν dy Ahν
8
Tot
Tot
Tot
Figure 2.4: Exponential decay of the optical power in dependence of silicon depth
where A is the area of light incidence and hν the electron energy. The optical power
is defined by the number of incident photons per time (photons flux) multiplied by the
photon energy. Herefrom, the number of photons per time can be calculated dividing the
power by the photon energy. As the optical power changes according to the depth the
derivative of the power depending on the depth is used.
The solution of the derivative of the power depending on the depth results in Equation
2.13, which shows the generation rate exponential decay according to the silicon depth.
αPT ot
G(y) = · exp(−αy) (2.13)
Ahν
The generation rate in semiconductor materials is simulated by optical solvers and added
to the semiconductor electrical equations as shown in Chapter 2.2.4.
The color of the light is defined by its wavelength. Most light sources, instead of a single
wavelength light, are mixtures of several wavelengths called spectrum. The sun light for
example comprises a wide range of wavelengths including ultra-violet, visible-light and
infrared, called white light, as shown in Figure 2.5.
The color of one object depends on the wavelength of the light reflected at its surface
and on the spectrum of the incident light. The color can be perceived by the human eye
by the combination of the responses of the rod and cone cells. Rods are responsible for
the perception of brightness and have no color sensitivity. Cones are the color detectors,
they are present in three groups with different spectral sensitivity behavior. One group
is sensitive to light near 570 nm, another group is sensitive to light near 540 nm, and the
third sensitive to light near 440 nm. The three types of cones are normally called long-,
9
Figure 2.5: Solar radiation spectrum [37].
Cones sensitivity
1
L-Cones(λ)
0.9 M-Cones(λ)
S-Cones(λ)
0.8
0.7
Tristimulus values
0.6
0.5
0.4
0.3
0.2
0.1
0
400 450 500 550 600 650 700 750
Wavelength (nm)
The theory of color perception based on three colors is called trichromatic theory. It
was proposed by Young in 1802 and expanded by Helmholtz in 1894. Based on the
trichromatic theory colors are formed by the combination of the responses of the three
cones. The trichromatic theory was first proved by measuring the absorption spectra of
the cones using microscopic techniques in 1964. More advanced techniques were used in
1967 when micro-electrodes were placed in the retina and presented results comparable
to the trichromatic theory [16].
Based on the fact that broad array of colors can be reproduced by mixing different ratios
10
of red, green, and blue light and due to the similarity to the human eye color mechanism,
the additive color mixing system is widely used in color detection and exhibition. Two of
the existing color standards are the RGB and XYZ, presented in the next chapter.
In 1931 the RGB additive color mixing principle was used by the International Commission
on Illumination (CIE), where the appropriate mixing of three defined primary colors was
matched to single wavelength color stimuli in order to create the RGB color matching
functions [35]. The three reference primary colors, known as tri-stimulus values, are
defined as monochromatic lights of 700 nm, 546.1 nm, and 435.8 nm respectively for red,
green, and blue light[16].
The device used to the definition of the color matching function consist of:
• A light source of specific wavelength (Stimulus). For example the 570 nm light
source shown in Figure 2.7
• The three primaries, reg, green, and blue shown in Figure 2.7.
Observer Observer
Blue Blue
Green Intensity of Green Intensity of
Red RGB light Red RGB light
(a) (b)
Figure 2.7: Illustration of the color matching experiment used on CIE 1931.
The observer sees side by side the Stimulus and the mixed result of the reference lights.
As shown in Figure 2.7 (a) initially the colors do not match. The observer adjusts the
three primary lights intensity until no difference is seen between the Stimulus and the
11
mixed light. In Figure 2.7 (b) the intensity of the red light is increased and the blue is
decreased in order to match both lights [35]. The color-matching functions are the amount
(intensity) of three references required to match the single wavelength light.
The color matching function was defined for small steps of wavelength along the visible
spectrum by several observers. Based on this experiment result the CIE 1931 RGB color
matching functions was defined depending on the wavelength, r(λ), g(λ), b(λ), as shown
in Figure 2.8.
0.2
Tristimulus values
0.15
0.1
0.05
−0.05
−0.1
400 450 500 550 600 650 700
Wavelength (nm)
For red, green, and blue primaries being R, G, and B, the tri-stimulus values represented
by the symbols TR , TG , and TB , and the Stimulus being S, we have:
S = R · TR + G · TG + B · TB (2.14)
As shown in Figure 2.8 the RGB color matching functions include negative values. That
happens because it is impossible to match the Stimulus with the sum of the three primary
colors. The saturation of the Stimulus is too high, and can be reduced by mixing the
12
Figure 2.9: Negative values on the color matching experiment.
primaries to the Stimulus. In the example presented Figure 2.9 the red primary is mixed
to the Stimulus [16].
In this case the Stimulus is defined as:
S + R · TR = G · TG + B · TB (2.18)
S = −R · TR + G · TG + B · TB (2.19)
In the 1930’s, when the RGB color matching functions were defined, the calculation of the
tri-stimulus values was done manually, and the presence on the negative values increased
the complexity of this manual calculations. To avoid the negative values of the RGB
color model, it was converted to the so called XYZ color model. The conversion was done
by a mathematical manipulation of the primary colors, therefore no experiments were
necessary for the the definition of the XYZ model.
Based on the fact that just a linear transformation in necessary to exchange between one
color model and another any transformation matrix that result in non-negative values
could be chosen to solve the previous problem. For that the Y component of the XYZ
model was chosen to match exactly to the V (λ) curve that represent the luminance of the
color.
The transformation from r(λ), g(λ), b(λ) to x(λ), y(λ), z(λ) is given by the matrix:
13
CIE 1964 XYZ
2.5
Tristimulus values
1.5
0.5
0
400 450 500 550 600 650 700
Wavelength (nm)
The tri-stimulus values of CIE XYZ color matching function obtained from the transfor-
mation is presented in Figure 2.10.
Similarly to the RGB model the tri-stimulus values for XYZ are calculated by:
Z X
TX = k x(λ)S(λ)dλ = k x(λ)S(λ)∆λ (2.20)
Z X
TY = k y(λ)S(λ)dλ = k y(λ)S(λ)∆λ (2.21)
Z X
TZ = k z(λ)S(λ)dλ = k z(λ)S(λ)∆λ (2.22)
Where x(λ), y(λ), z(λ) are the color-matching functions and k is a constant. S(λ) is the
color function of the Stimulus light and ∆λ is a variation of 1 nm wavelength [36].
The color matching functions are the standard curves which should be matched by a
color sensor in order the accurately represent color. As presented in Chapter 4 each color
sensing structure proposed on this thesis has responses of three photodiodes optimized to
match either XYZ or RGB standard.
Color temperature
The color of the light can be represented not just by the level of RGB responses but also
by the characteristics of the light spectrum. The color temperature is used to express the
appearance of the light and is measured in Kelvin (K).
The relation between temperature and electromagnetic radiation is defined by Planck’s
law, proposed in 1900 by Max Plank. It describes the electromagnetic radiation emitted
by a black body at a definite temperature.
14
Figure 2.11: Black body radiation [4].
The spectral response of the radiation provided by the black body is illustrated in the
Figure 2.11. The peak of the distribution curves shifts to longer wavelength for lower
temperature and to short wavelengths with higher temperatures.
The color temperature of the light emitted by a source is the temperature the black body
needs to be heated in order to radiate light in the same color [16]. Color temperature just
makes sense for light colors that are similar to those radiated by the black body, shown
in Figure 2.12 by the line that goes from red to bluish colors, passing by the yellow.
Figure 2.12: CIE 1931 RGB Color Space with color temperature representation [3].
The definition of cool and warm color is not related to the temperature of the black
15
body but to the visual sensation of the color. Because of that even though the black
body temperature is high for bluish colors and low for reddish colors, they are defined
respectively as cool and warm colors.
Measuring and controlling the color of the ambient light is an important topic in several
areas as for example lightning, where the color temperature is directly related to sensation
of relaxation or concentration, for that an appropriated ambient light color is required
depending on the activities to be executed on the ambient. Other important branches
are the photography, artistic applications, images edition, and product design, because
depending on the color of the light used during the production, the product can present
a different appearance when exposed to sun light. The appropriated color temperature of
illumination is in this case 5500 K due to the similarity to the sun light [3].
2.2 Photodiodes
Photodiodes are widely used as photodetectors in light sensing applications. A photodiode
by itself is sensitive to a wide spectrum of light and presents no color selectivity. However
depending on the depth of the photodiodes p-n junction and on the arrangement of certain
number of photodiodes on the same device the color selectivity can be obtained. In this
chapter the characteristics of a p-n junction and the electrical equations that describe the
photodiodes behavior are presented. Additionally the color selectivity in dependence of
the junction depth and the definition of photodiode responsivity are presented.
A photodiode consists of a p-n junction. A p-n junction contains a P-side with high
concentration of positive carriers and a N-side with negative carriers. When the p-n
junction is formed the free carriers from the P-side tend to diffuse to the N-side and the
free carriers from the N-side tend to diffuse to the P-side. The carriers that diffuse to the
opposite side recombine letting behind fixed opposite charges. This diffusion of carriers
creates a depletion region at the junction of the two materials where no free carriers exit.
A net of negative fixed charges is build on the P-side and a net of positive fixed charges
on the N-side [8]. The fixed charges create an electrical field from N-side to P-side that
opposes the diffusion of free carriers. A potential difference is formed across the junction.
When it reaches an equilibrium this potential is called built-in-potential.
A p-n junction illustration is presented on Figure 2.13 where a N-doped substrate receives
an P-implant (P-layer) in order to create the p-n junction. When a photodiode is exposed
to light the electron-hole pairs generated in the space charge region tend to be separated
by the electrical field. The holes are accelerated to the P-side and the electrons to the
16
Figure 2.13: Generation and the separation of electron-hole pairs in a photodiode [15].
N-side which results in a photocurrent. Also the carriers generated in the substrate can
contribute to the photocurrent in case they reach the region with electrical field before
they recombine. The two components responsible for the photocurrent are called drift
and diffusion current.
The drift current is created due to the electric-field, specially in the depletion region of
p-n junctions. Equations 2.23 and 2.24 present the drift current from electrons and holes.
where n and p are the electron and hole density, q is the electron charge, µn,p is electron
and hole carrier mobility, and Ẽ is the electron field. The multiplication of the carrier
mobility by the electron field is defined as drift velocity µn,p · Ẽ = vn,p
˜ .
The carrier mobility is also dependent on the electrical field and given by:
µ0n
µn = (2.25)
1 + µ0n Ẽ/vnsat
µ0p
µp = (2.26)
1 + µ0p Ẽ/vpsat
Where µ0 is the carrier mobility with no electrical field, which depends on temperature
and impurities concentration, and vsat is the saturation velocity for electrons and holes,
that is also temperature dependent.
17
As shown in Equation 2.25 and 2.26, the electron and hole mobility decreases for high
electrical fields. For that the drift velocity is proportional to the electrical field only for
small values of the electrical field.
The carrier diffusion happens in regions where no electrical field is present, generally
deep in the silicon substrate. Equations 2.27 and 2.28 present the diffusion current from
electrons and holes.
Where grad(n) and grad(p) are the gradient of the electron and hole density, q is the
electron charge, and Dn and Dp are the carriers diffusion coefficients,
The carrier diffusion coefficient for electrons and holes, presented in Equations 2.29 and
2.30, is dependent on the carrier mobility µ and on the thermal voltage.
Dn = µn VT (2.29)
Dp = µp VT (2.30)
The photogenerated carriers, generated in the substrate, can diffuse to the region with
electrical field, being accelerated to the space charge region. The diffusion length L defines
the distance these carriers can diffuse before recombining. L depends on carrier diffusion
coefficient and carriers lifetime (τn and τp ), as shown in Equations 2.31 and 2.32.
p
Ln = Dn τn (2.31)
p
Lp = Dp τp (2.32)
The impurities in the silicon, generated during the fabrication process, work as recom-
bination centers. There fore the carrier lifetime is strongly dependent on the number of
impurities.
18
2.2.4 Drift-diffusion transport Model
The investigation of photodiodes operation is done by simulating its optical and electrical
behavior. The simulation tools use optical solvers to calculate the optical generation in
a photodiode with information about the number and location of the photo-generated
electron-hole pairs. The drift-diffusion model is used to simulate the electrical behavior of
photodiodes and describes the transport of the photo-generated carriers within the silicon
and its contribution to the photocurrent.
The drift-diffusion model is defined by the basic semiconductor equations. Some of the
most important equations are the current density and continuity equation for electrons
and holes and the Poisson equation.
The continuity equation for electrons and holes are defined as:
∂n ∇Jn
= +U +G (2.33)
∂t q
∂p ∇Jp
=− +U +G (2.34)
∂t q
where U is the electron-hole recombination rate, G is the carriers generation rate (See
Section 2.1.3), n and p are the electron and hole density, and Jn and Jp are the electron
and hole current density, given by:
The current density has a drift component caused by the electrical field and a diffusion
component caused by the gradient of the carrier concentration. The total current density
j is the sum of the electron and hole current density.
During the fabrication process impurities are introduced to the device. These impurities
can work as recombination centers that increase the recombination rate. The recombi-
nation rate based on the Shockley-Read-Hall (SRH) model is shown in Equation 2.37.
pn − n2i
U= (2.37)
τp (n + n1 ) + τn (p + p1 )
for:
Etrap
n1 = ni · exp( ) (2.38)
kT
−Etrap
p1 = ni · exp( ) (2.39)
kT
19
The recombination rate U depends on the electron and hole density (n, p), on the intrinsic
carrier density (ni ), on the carriers lifetime (τ ), and on the difference of energy level on
the recombination centers and intrinsic Fermi level (Etrap = Et − Ei ) [1].
The potential within a semiconductor in dependence of the charge density is defined by
the Poisson equation. The Poisson equation is defined as:
ρ
∆Ψ = − (2.40)
ε
where ∆ is the Laplace operator, Ψ is the potential, ε is the electrical permittivity and ρ
is the charge density.
The charge density is defined by the electron and hole density n and p and the concen-
tration of donors and acceptors ND and NA , as shown in Equation 2.41.
ρ = q(p − n + ND − NA ) (2.41)
The presented electrical equations are solved by the simulation tools during the investiga-
tion of the color sensor structures. Due to the light falling on the surface of a photodiode
carriers are generated. The carriers generation rate (see Chapter 2.1.3) is calculated by an
optical solver. The connection between the electrical and optical simulations is provided
by the inclusion of generation rate in the continuity equations. Further information about
the optical and electrical simulations are presented in Chapter 3.
The light absorption and carriers generation in the silicon is depth dependent. Therefore
when light falls on the silicon, electron-hole pairs are generated close to the surface as well
as deep in the substrate, depending on the light wavelength. Figure 2.14 (a) illustrates the
wavelength dependent carrier generation. For blue color light with wavelengths of about
400 nm to 500 nm, the light absorption coefficient of silicon is very high and the carrier
generation mainly happens near the silicon surface. For red light with wavelengths above
600 nm, there is high carrier generation even deep in the silicon. Figure 2.14 (b) presents
the exponential behavior of the carrier generation, according to the silicon depth, for red
and blue light.
The generated electron-hole pairs when reaching a p-n junction are separated by the
electrical field generating a photocurrent. The most common way to operate a photodiode
is in photoconductive mode, where a reverse bias is applied across the junction and the
resulting current flow is measured [8]. Figure 2.15 shows two examples of photodiodes
and the respective spectral response expected for each one.
Figure 2.15 (a) shows a shallow p-n junction, that forms a photodiode close to the surface.
When short wavelength light falls on silicon, carriers are generated close to the surface
20
Carrier Generation
Blue
Substrate depth
Red
Figure 2.14: Light absorption and carrier generation in dependence of the silicon depth.
N+
PD1
Pwell
Response
PD1
P-substrate
Wavelength
(a) (b)
PD2
Nwell
Response
PD2
P-substrate
Wavelength
(c) (d)
Figure 2.15: Photodiodes responsivity according to the depth of the p-n junction
and mostly collected by the p-n junction. In the case where long wavelength light falls
on silicon, carriers are generated close to the surface and deep in the substrate. Just a
small part of the carriers, generated close to the surface, will be collected. The carriers
generated deep in the substrate are mostly recombined in the high doped P-well, being
prevent of reaching the p-n junction. The shallow p-n junction collects more carriers of
short than long wavelengths which results in the responsivity shown in Figure 2.15 (b). A
21
deep p-n junction, shown in Figure 2.15 (c), will collect the carriers generated close to the
surface from the blue light as well as the carriers generated deep in the substrate, from
the red light. Therefore the deep p-n junction has a wide spectral response, as presented
in Figure 2.15 (d).
P-n junction in different depths can be used for color sensing. The detailed color separa-
tion techniques are presented in Chapter 2.4, where also different configurations as double
and lateral photodiodes are discussed.
A more practical measurement of the device sensitivity is given by the responsivity [2].
The responsivity of a photodiode is given by the generated photocurrent Iph divided by
the incident optical power Ptot , as shown in Equation 2.43.
Iph q · λ0 · η
R= = (2.43)
Ptot h·c
Where h is Planck’s constant, λ0 is the wavelength of the light in vacuum, c is the speed
of light, q the electron charge, and η quantum efficiency. [1]
During the device simulation the applied optical power and the simulated photodiode
currents are used to calculate the responsivity. After the production the photocurrent
of the developed sensors are measured and the applied optical power is measured by a
reference power-sensor. The responsivities are also calculated out of the measured values.
The responsivity obtained during simulation and measurements are compared.
22
steps such as the real production. Therefore the knowledge of the CMOS process is
essential in order to create a realistic model for the color sensor, consequently obtain
optical and electrical simulation results comparable to the produced device.
In this chapter the CMOS technology, the process steps, and the possible applications in
standard CMOS color sensors are presented.
The Metal Oxide Semiconductor (MOS) technology was created in the 1930’s but just
came to the market in the 1960’s when N-type transistor were produced. Later in the
1960’s the CMOS technology was created, when N-type and P-type transistors became
available. The use of CMOS grew fast specially in digital electronics and later also in ana-
log circuits. The low production costs and the possibility of down scaling are some of the
advantages that contribute to the wide use of CMOS technology. The persistent improve-
ments in dimension and speed made CMOS the dominant technology in the fabrication
of integrated circuits [10].
The basic CMOS structure consist of an N-type and a P-type transistor in the same
substrate, as presented in Figure 2.16.
Metal 4
Metal 3
Metal 2
Metal 1
N+ STI P+ P+ STI N+
P+ N+ STI P+ STI
N-well P-well
P-Substrate
Figure 2.16: Cross-section a CMOS structure, with a PMOS and an NMOS transistors.
The N-channel MOS (NMOS) is created inside a P-well which is externally connected by
a highly doped P-type implant (P+). The drain and source of the NMOS are two highly
doped N-type implants (N+) and the polysilicon gate G is isolated by a thin layer of
silicon oxide. The P-channel MOS (PMOS) structure is complementary to the NMOS,
23
it contains an N-well and two P+ implants. A Shallow Trench Isolation (STI) is used to
isolate the two structures.
The starting material for the production of CMOS devices is a pure crystalline silicon
wafer obtained by Czochralski method. The silicon is normally lightly P-doped silicon,
with a doping concentration in between 8 · 1014 cm−3 and 1.2 · 1015 cm−3 . The Czochralski
method results in a silicon ingot that is sliced and passed through several steps of polishing
and etching to remove all imperfections at its surface before the CMOS process starts.
The fabrication of Integrated Circuits (ICs) includes several mechanical and chemical pro-
cesses used to the generate the source/drain implants, wells, metal contacts and isolators.
The main processes used in a CMOS production are described according S. Wolf [11].
• Photolithography
The patterns drawn during the circuit layout are transferred to the silicon by lithog-
raphy process using a mask. The wafer is covered by a photoresist layer, and it is
exposed to the light through the mask. According the the mask, part of the photore-
sist is exposed to light. In case of a positive photoresist the part of the photoresist
exposed to light is removed, allowing the process step to be performed just on this
opened area, while the covered area is protected (the opposite occurs for a nega-
tive photoresist). Using negative and positive photoresists enables the use of the
same mask for complementary process avoiding the production of expensive extra
masks. The lithography is used to define the areas to be submitted to etching and
ion implantation steps.
• Oxidation
There are two possible kinds of oxidation: Dry oxidation where the silicon is exposed
to oxygen and wet-oxidation where silicon is exposed to water vapor. Even though
the silicon oxidation happens naturally at room temperature, the layer created is
generally too thin for applications in integrated circuits. To improve the oxidation
the processes is carried out at high temperatures (800 to 1200 o C). One of the main
applications of oxidation is the growth of gate oxide. It can also be used as isolation
material and protective layer in other process steps.
• Ion Implantation
The ion-implantation is used to add dopants to the silicon, to create the Source/Drain
regions and wells. A thin oxide layer is grown on the silicon surface as protection
during the ion implantation steps. Atoms are accelerated with high energy to the
24
silicon surface and penetrate the exposed areas. The penetration depth is defined
by the implantation energy and the doping concentration by the intensity and du-
ration of the implantation precess. The well-implants are created with an energy
of approximately 2 MeV while the Source/Drain implants need less energy since
they are shallower implants (approximately 0.2 keV). The implantation steps are
followed by temperature steps (annealing) to correct the damage created by the ion
implantation. During this step also the diffusion of the ions in the silicon takes
place.
• Deposition
Some materials as polysilicon, dielectric, and metal are added to the silicon surface
by deposition. One common method of deposition is the Chemical Vapor Deposition
(CVD).
• Etching
The etching is used to remove completely or partially material from the silicon
surface. Different etching technologies are used depending on the material to be
attached and on the required precision.
• STI
The STI is created by etching trenches of 400 µm to 500 µm depth by an anisotropic
dry-etch step as shown in Figure 2.17 (a). Later a thin thermal oxide is grown on
the walls of the trench and a CVD dielectric film is used to fill the trenches. In
Figure 2.17 (b) the structure with completely filled trenches is presented.
• Well-formation
To form twin-wells two steps are necessary. To create the P-well the lithography step
is performed with the P-well mask followed by a P-implantation as shown in Figure
2.17 (c). Figure 2.17 (d) shows the P-well already formed and the lithography with
the N-well mask performed while the ion implantation for the N-well is in progress.
After the implantation a diffusion temperature step is performed.
25
STI STI STI STI STI
STI STI STI STI STI STI STI STI STI STI
P-well
STI STI STI STI STI STI STI STI STI STI
Metal 1 Metal 1
26
• Polysilicon Deposition
A 400 µm to 500 µm thick polysilicon layer is deposited by Low-Pressure Chemical
Vapor Deposition (LPCVD) over the whole structure. Figure 2.17 (e) presents both
wells formed, and the gate oxide and gate polysilicon already deposited over the
surface. The lithography step is performed with the gate-mask. The gate region
remains protected, while the polysilicon and gate oxide outside are precisely etched.
• Contact Formation
Titanium salicide is formed over the Source/Drain implants and gate to decrease the
contacts resistance. An Inter-level Dielectric (ILD) layer is deposited, and according
to the contact-mask the holes are etched as shown in Figure 2.17 (h). Later the
metal is deposited to form the contact.
Modification on the CMOS process or the use of advanced CMOS technologies can signif-
icantly increase the production costs of color sensors. In this case the device may become
unsuitable for mass market application, since low production cost is a basic requirement
on this field.
Photodiodes can be created in Standard CMOS technology by using existing p-n junctions,
where no modification and additional cost are required. The available p-n junctions are
shown in Figure 2.18 and listed bel ow.
27
STI N+ STI P+ STI P+ N+ STI P+ STI STI N+ STI P+ STI STI N+ STI P+ STI
N-well P-well N-well P-well P-well
28
a wide spectral range, while the color sensitivity is provided by filters added on top of the
sensor.
In Figure 2.19 (a) a Bayer array is shown, where 50% of the photodetectors are covered by
green filters and the rest is equally divided between red and blue. In the photodetectors
Figure 2.19: (a) Bayer Filter Array. (b) Light absorption and reflection on the photodiode
covered by blue filter; (c) Green filter; (d) Red filter [18].
covered by blue filter, shown in Figure 2.19 (b), the blue light passes through the filter
and absorbed by the photodiode. Light with other wavelengths is completely absorbed by
the filter before reaching the photodiode. The same effect happens for the red and green
filters (See Figure 2.19 (c) and (d)).
The use of CFA has some drawbacks. Due to the use of filters, just part of the incident
light is transmitted to the photodiode, decreasing significantly the overall sensitivity of
the sensor. Another factor to be considered is that at each pixel the information of a
single color is read, while the other two must be obtained by interpolation algorithms.
Furthermore these sensors require complex filter assembly on top of the sensor which
significantly increases the production costs.
Alternatively to external filter based sensors, also filter-less sensor concepts are used,
mainly based on wavelength dependent light absorption properties of silicon, as explained
in Section 2.2.5. An implementation is proposed in the Foveon X3 Technology [13]. This
technology has a vertical stack of three optimized photodiodes (See Figure 2.20), where
each one absorbs the generated carriers related to different light colors [14]. In this color
sensor the red, green, and blue light intensity can be measured, avoiding absorption losses
29
Figure 2.20: Stacked-photodiodes color sensor [9].
and therefore better quantum efficiency than sensors with color filters. Even though the
triple-junctions are usually available in modern CMOS technologies, the spectral respon-
sivity of the 3 stacked photodiodes is directly defined by technology parameters like doping
concentrations and profiles and there is almost no room for optimization without expen-
sive technology modifications. Also for older CMOS technologies a triple-well option is
not always available.
In order to search for a low cost solution for color sensing application, differently from the
presented solutions, modifications of standard CMOS technology should be avoided. For
this reason the innovative methods of the newly developed color separation are entirely
based on standard CMOS technology.
In the next sections solutions are presented where the depth dependent carrier generation
and the lateral carrier diffusion can be used to produce color selectivity. Additionally to
these solutions, the possibility of tuning the spectral responses by using the field effect
and electrical field is introduced.
The theory of the separation methods are presented in the next sections. The application
on complete color sensor structures, with the purpose of obtaining responsivities similar
to the color matching functions are presented in Chapter 4.
A double photodiode is formed by stacked photodiodes. The different depth of the two
p-n junctions are responsible for the achieved color separation. A structure of a double-
photodiode is shown in Figure 2.21 (a).
The first p-n junction (P D1) is formed between a P+ implant and the N-well (PPNW).
The second p-n junction (P D2), which is vertically stacked below the first photodiode, is
created between the N-well (NW) and the P-substrate. This configuration can be realized
in any standard CMOS process [2].
The Figure 2.21 (b) shows the wavelength dependent carrier generation rate. Light with
short wavelength (blue-light) is absorbed and consequently carriers are generated close to
30
Carrier
PPNW NW P-sub Generation
P+ N+ P+
Blue
N-well
Red
(a) (b)
Figure 2.21: (a) Cross-section of a double-photodiode. (b) Carrier generation rate versus
silicon depth.
the surface. Therefore most of the carriers diffuse or drift to the upper p-n junction P D1
which result in an increase of the photocurrent at P D1. For light with long wavelengths
(red-light) the absorption and generation occurs even deep in the substrate, so the carriers
are mainly collected by junction P D2 [2]. The two photodiodes P D1 and P D2 have
PD1
0.6 PD2
0.5
Responsivity
0.4
0.3
0.2
0.1
0
400 500 600 700 800 900 1000
Wavelength
(c)
different spectral responses, with a maximum of about 500 nm for P D1 and a very broad
response with a maximum at about 750 nm for P D2 as shown in Figure 2.22.
31
The current at the terminal PPNW is equal to the current of P D1 while the current at
the terminal NW is equal to the currents of P D1 + P D2.
Even though the two diodes have quite different spectral responses, it is not enough to
measure the accurate color of the light. A color measurement requires at least three
spectral independent sensor signals to represent an RGB color space [16].
The third spectral response necessary for the color reconstruction can be obtained by
lateral carrier diffusion effects. The carrier diffusion within the proposed color sensor for
blue and red light is illustrated in Figure 2.23. The structure contains two photodiodes
placed side-by-side. The photodiode formed by the N-well Dark (NWD) is completely
shielded from the incident light by a metal layer. And the other photodiode consists of
an N-well under Light (NWL).
-
- - - -
--
- -
- - - - - - - - - - - -
- - - - - -
- - - - - - - - -
- - -
- - - - - - - - -
- - -
- - -
- - - - - - -
- - -
- - -
- -
- -
- -
- - -
- -
P-substrate P-substrate
Figure 2.23: Illustration of the lateral carrier diffusion for short- (a) and long-wavelength
light (b).
As already shown in Figure 2.21 (b), photons are absorbed and carriers are generated in
different depths depending on the light’s wavelength. The carrier diffusion in semicon-
ductors is defined by a local electrical field and a carrier density gradient.
When blue light is applied to the sensor (Figure 2.23 (a)), the carriers are generated
near the surface and are mainly collected by the photodiode NWL. When red light is
applied (Figure 2.23 (b)), a considerably high number of carriers diffuse also to the covered
photodiode NWD.
The expected response for the two photodiodes is presented in Figure 2.24. The NWL
32
Figure 2.24: Effect of the lateral carrier diffusion on the photodiodes responses
is sensitive to a wide spectral range and NWD is sensitive exclusively to the long wave-
lengths.
The implementation of a color sensor based on the combination of a double-photodiode
and lateral arranged photodiodes is presented in Chapter 4.1. A second implementation
entirely with lateral photodiodes is presented in Chapter 4.2.
According to the trichromatic theory, three spectral responses are necessary for the color
representation. But in case more independent spectral responses are available more ac-
curate results can be obtained. Two mechanisms are investigated where the spectral
response can be tuned according to an applied voltage. These mechanisms are based on
substrate current and on the field effect.
Substrate current
An external electrical field applied to the substrate of the color sensor can redirect the
generated carriers and consequently change the spectral response of the photodiodes. As
presented in Figure 2.25 the structure consists of photodiode covered by metal NWD and
an uncovered diode formed by the N+ implant under Light (NPL). The P-substrate is
externally connected at two points (PP and Psub).
A voltage is applied to the contact PP while Psub is connected to 0 V. The potential
difference creates a substrate current. For PP > 0 the substrate current (created by
majority carriers) flows from PP to Psub and an electrical field is created.
33
Light Light
- - - - - - - - - - - - - -
- - - -
-
- - - - - - - -
- - - --
- - - -
- - -
- - - - - - - - -
- - - -
- -- - - - - -
- - - - - - -
- - - - - Substrate Current Substrate Current - - - -
- - -
- - - -
- - - - - - -
- - - - -
- - - - -
- -
E-Field E-Field
P-substrate P-substrate
(a) (b)
The minority carriers (that represent the photogenerated current) are redirected to PP
as shown in Figure 2.25 (a). Before reaching the contact PP the minority carriers can
be collected by the Space Charge Region (SCR) of NWD, thus increasing its current.
With voltage in PP < 0 the carriers are redirected to NPL as shown in Figure 2.25 (b),
increasing its current.
The photogenerated carriers corresponding to the short wavelengths light are generated
close to the surface and immediately collected by a p-n junction, for that suffering no
effect of the electrical field. On the other hand the carriers generated by long wavelengths
light take more time to reach a SCR and therefore can be redirected by the electrical field.
An implementation of a color sensor with tunable spectral response based on this principle
is presented in Chapter 4.3.
Field Effect
In this structure the spectral responses are tuned by using three polysilicon electrodes
over STI between the uncovered N+ implant (NPL) and the dark N+ implant (NPD), as
presented in Figure 2.26.
Different voltages are applied to the electrodes in order to create a potential gradient on
the region under the electrodes, causing the generated carriers to flow from the region of
lower potential to the region with higher potential. Figure 2.26 (a) shows that when the
electrodes (E1, E2, and E3) are biases respectively with 3 V, 1.5 V, and 0 V, the minority
carriers are directed to NPD. The opposite configuration with E1, E2, and E3 biased at
0 V, 1.5 V, and 3 V is shown in Figure 2.26 (b), where the carriers are redirected to NPL.
A color sensor structure, based on the field-effect, which presents tunable spectral response
depending on the applied voltages is designed and presented in Chapter 4.4.
34
Light Light
E1 E2 E3 E1 E2 E3
3V 1.5V 0V 0V 1.5V 3V
NPD NPL NPD NPL
- - - - - - - - -
- - - -
-
-
- - - -
- - - - - - - - - - - -
- - - - - -
- - - - - -
- - - - - - - - - - - -
- - - - - - -
- - - - - - - -
- - - - - -
- - - - - - -
-
- -
- - - - -
- - -
- -
- - -
-
-
-
P-substrate P-substrate
(a) (b)
Figure 2.26: Influence of the field effect on the lateral carrier diffusion.
35
Chapter 3
The methods of color separation presented in this thesis are innovative ideas which should
be investigated and studied before a practical application. The understanding and verifi-
cation of color separation mechanisms presented in Chapter 2.4 was done by simulations.
The use of actual technology process parameters and proper models for optical and elec-
trical simulation can bring simulation results closer to the real structure behavior, thus
increasing the chances of a working silicon prototype.
The simulations are performed using the Synopsys TCAD tools. TCAD tools model
the semiconductor devices fabrication and operation. An illustration of the simulation
flow is shown in Figure 3.1, presenting the following four steps: the device design, mesh
generation, device simulation, and visualization and analysis.
The device design can be done using two tools, Sentaurus Structure Editor SSE and
Sentaurus Process Sprocess. Both tools have as output structures with defined dimen-
sions, materials doping profiles, and electrical contacts (Figure 3.1 (a)). Nevertheless the
methods for structure generation are really distinct from each other. Sprocess models
and simulates all the steps of the technology process flow. SSE is a simplified structures
generator tool, which creates structure as ideal geometrical objects of defined material
and doping.
The output file of the device design is used as input for the mesh generation. Snmesh is
used to create the discretization mesh according to the optical solver used on the device
simulation (see Chapter 3.3.2). The output of Snmesh is presented in Figure 3.1 (b).
The device simulation is performed by the tool Sentaurus Device Sdevice and comprises
the optical and electrical simulations. The optical simulation consists of solving the optical
generation rate on the device. This can be performed by different optical solvers. In this
thesis three optical solvers are used, Raytracer, Beam Propagation Method (BPM) and
Sentaurus Electromagnetic Wave Solver (EMW). The electrical simulation is performed
based on the drift-diffusion transport model. Sdevice generates one output file with
36
Device Design
(a)
Mesh generation
(b)
Device simulation
Figure 3.1: Simulation flow. (a)Designed structure. (b)Designed structure with generated
mesh. (c)Light rays used for optical simulation. (d)Simulated electron current
density. (e)Simulated photodiodes responsivities.
37
simulation results spatially distributed on the device and a second file which contains the
simulated output on the electrical contacts.
The output of SSE, Sprocess, and Sdevice can be visualized and analyzed with Sentaurus
Visual (Svisual) as shown in Figure 3.1 (c) and (d), where respectively the light rays used
for the device simulation and the electron current density on the device are presented.
The current simulated on the electrical contacts is plot in Matlab. Normally the contact
currents in dependence of the wavelength are used to calculate and plot the responsivity
in Matlab
Sentaurus Workbench is used to collect several tools on the same environment.It also
allows the parametrization of the simulation values and automation of the simulation
steps.
Detailed information about the four simulation steps and the simulation tools is presented
in the next chapters.
38
Command file
Sequence of commands
that describes initial
structure and each
Sentaurus process
process step.
Run all process steps Sprocess output-file
according command and Contains information of
mask files. structure dimensions,
Mask file Creates an output file with doping, grid, etc.
Define size and position the structure data.
of implants (N+, Nw,
P+), STI and contacts on
the initial structure.
The initial structure is defined as a piece of silicon of given dimensions, with orientation
100 and a initial P-type doping concentration. The masks dimensions are defined for each
fabrication step. The implantation, oxidation, diffusion, etching, and deposition steps are
simulated to create the required structure.
Several commands, that represent process steps, allow the selection of physical models
and parameters. The process steps are simulated according to the real process flow used
in the CMOS production and the used models are specified by technology specialists in
order to provide the optimum matching between simulated and produced structures. The
process steps modeled by Sprocess and how they are handled in this work are introduced
below.
• Photolithography
For the lithography process the mask dimensions are defined, and the photo com-
mand is used to make a resist spin-on step, followed by an exposure bake and etch
steps. The thickness of the photoresist can be defined. Positive and negative pho-
toresist can be simulated [23].
• Oxidation
The oxidation is defined during a diffusion or temperature ramp step. The temper-
ature, time, pressure, and ambient are defined. The ambient and gas flow can be
defined during the oxidation. For example: Oxide growth with H2 O or O2 ambi-
ent and Oxynitride growth with ambient N2 O. A more complex oxidation can be
defined with several temperature ramps. Mechanical stresses caused by the volume
increase resulting from the conversion from Si to SiO2 are considered during the
simulation, the mobility of charged carriers and leakage currents are dependent of
the stress in the system [26].
• Ion Implantation
Ion implantation is one of the most widely used processing techniques to introduce
39
doping atoms into semiconductor materials. In Sprocess, either analytic functions or
the Monte Carlo method is used to compute the distribution of implanted ions and
the implantation damage. Here the Monte Carlo models are used with Sentaurus
MC simulator. Alternatively the Crystal-TRIM simulator is available for Monte
Carlo implantation [24].
• Diffusion
After the implantation a temperature step is performed for the correct diffusion of
dopants. During the fabrication process, dopants are introduced into the substrate
with different concentration profiles. The process goes through various thermal an-
nealing cycles, the dopants diffuse and redistribute through the structure. Sprocess
models several important effects for the diffusion, like: Dopant (de)activation,
dopant-defect interaction, chemical reactions at interfaces and in bulk materials,
material flow, moving material interfaces and internal electrical fields [25].
• Deposition
The deposition methods available in Sprocess are: Isotropic, fill, anisotropic, Fourier,
directional, crystallographic, and polygonal deposition. Also selective deposition can
be performed [27].
• Etching
The etching methods available in Sprocess are: Isotropic, trapezoidal, anisotropic,
Fourier, directional, crystallographic and polygonal etching, and Chemical-Mechanical
Polishing (CMP) [27].
The process simulation used during this work is a description of the real process steps.
Real implantation energies, oxidation time, diffusion, time and oxide/nitride layers thick-
ness are used. Due to the high complexity of the process and models involved, a process
simulation can take hours to be completed.
Sentaurus Structure Editor is used to create structures as geometric objects. The initial
structure is defined as a piece of silicon with certain depth, width, and with an initial
P-type doping concentration. Scripts are used to create the structure. Those are added
to a command file and used as input for SSE. The STI is created as an ideal geometric
object where the material is defined as oxide. Windows are defined where implants will
be located and the vertical doping profile is loaded from a file [19].
The vertical doping profile is the information of doping type and concentration, depending
on the silicon depth. The doping profile for each implant (N+, N-well, P-well, etc..) are
40
extracted from a process simulation to be used as input in SSE, while the lateral dopant
diffusion is an estimated value. The dielectric stack on the top of the structure is created
as an ideal geometric object where the materials are defined as oxide and nitride. The
material and dimensions are manually extracted from a previous process simulation.
As shown in the flowchart in Figure 3.3, SSE needs as input a command file with the
description of the structure size, position and dimensions of the implants, and STI and
oxide/nitride layers. A second input file contains the doping profile to be imported. The
output of SSE is a .tdr file with information about the used materials, dimensions, doping
(no stress and point defect are modeled in SSE).
Command file
Contains information and
commands for
generation of the Sentaurus Structure
geometric object,
Editor Sse output-file
implants and grid.
Run the commands Contains information of
according to command structure dimensions,
Doping profile file. Creates an output file doping, grid etc.
Contains information of with the structure data.
doping type and
concentration depending
on the silicon depth.
Simulations show that there are no significant differences on the results of a structure
generated in Sprocess or in SSE, while the required time can vary from approximately
5 minutes in SSE up to 4 hours in Sprocess. For that the first simulation necessary for
extraction of the technology data is performed in Sprocess but during the investigation of
the structures operation just SSE is used.
41
Figure 3.4: Device mesh created for discretization of optical and electrical simulation.
Different meshing strategies are used closer to the surface, where more accuracy is required
and deeper on the substrate. A fine mesh depending on the doping concentration is applied
to the first 10 µm, this mesh is adapted depending on doping profiles around p-n junctions.
Deeper in the substrate a coarser mesh is used in order to decease the simulation time.
42
Command file Sdevice output-file
Contains information about Contain information of
electrical and optical structure dimensions, doping,
equations to be solved, grid in addition to spatial and
solvers, methods and quantitative data of simulation
models. Bias voltage and Sentaurus Device results (optical generation,
data to be saved. Run command file, solve electrical field).
optical and electrical problems
Sprocess/Sse output- according to specified models
file and solvers . Sdevice output-file
Creates two output files. Contain the current plot data
Contains information of
structure dimensions, doping, and the corresponding
grid, etc. wavelength.
The electrical simulations are performed using the drift-diffusion transport model, that
is defined by the basic semiconductor equations where the electrostatic Poisson and the
carrier continuity equations for electrons and holes are solved (See Chapter 2.2.4). The
Newton’s method is used to solve the equations [29].
The optical simulation consists of solving the optical generation rate, resulting from a
monochromatic optical source. The angle of the incident light is defined as perpendicular
to the silicon surface, the light intensity is fixed in 1 W/cm2 and the wavelength of the
monochromatic light is ramped from 400 nm to 1000 nm.
The refractive index, absorption, and recombination models as well as the optical solvers
used during the simulations are summarized below.
Optical Solver
Several solvers are available in Sentaurus Device for computation of the optical generation
rate. The choice of the proper solver depends on the required combination of accuracy
and computation time. Three of these solvers, Raytracer, EMW, and BPM were used
during the color sensors simulation.
• Raytracer
Raytracer is implemented based on linear polarization and geometric optics where
the light propagation is described as rays. The geometrical optics is a simplification
of the optics where the refraction, transmission, and reflection (see 2.1.2) are taken
into consideration while effects such as interferences and diffraction cannot be mod-
eled. Raytracer is optimized for speed and is a good model to be used when the size
43
of the structure to be simulated is much bigger than the wavelength of the incident
light. Raytracer needs to be used together with the complex refractive index model.
After the optical generation rate has been computed (see 2.1.3), the values are
divided by the volume to obtain the final units of cm−3 /s. This value is added to
the carrier continuity equation as the generation rate G [30].
• EMW
Sentaurus Electromagnetic Wave Solver (EMW) is a simulation module for the nu-
meric analysis of electromagnetic waves. EMW is a full-wave time-domain simulator
based on the Finite-Difference Time-Domain (FDTD) solver. The simulations per-
formed with EMW are the most accurate ones but on the other hand they can
become very time-consuming [21].
• BPM
The Beam Propagation Method ’s (BPM) efficiency and sufficient accuracy make it
an attractive solver for filling the gap in between Raytracer and EMW. The BPM
implemented in Sentaurus Device is based on the Fast Fourier Transform (FFT)
and is a variation of the FFT BPM, which was developed by Feit and Fleck [30].
Due to the long simulation time required to solve the optical generation with EMW and
BPM, most of the simulations in this thesis are performed with Raytracer.
Simulations with EMW and BPM are performed for comparison, showing that the ac-
curacy loss by using Raytracer is not critical. Despite the effects of diffraction and in-
terferences not modeled by Raytracer the results are satisfactory for the characterization
and investigation of the color sensor. Additionally simulations with EMW and BPM
frequently result in convergence problems.
The complex refractive index model allows the definition of the refractive index and
the extinction coefficient depending on mole fraction, wavelength, temperature, carrier
density, and local material gain.
The only correction factor used for the simulations is the wavelength dependence. It can
be done by using an analytic formula that considers a linear and a square dependency on
the wavelength, or by loading the complex refractive index from the parameter file of the
correspondent material or external tabulated file [32].
Sentaurus has the complex refractive index added to the parameter file of several mate-
rials, that are used in the simulations. The complex refractive indexes not provided by
Sentaurus are obtained on a on-line Refractive index database [17].
44
Absorption models
Depending on the optical generation model and used optical solver, the absorption models
are defined. For Raytracer, BPM, and EMW the absorption is defined as a function of
the complex refractive index [31]. (See 2.1.3).
Recombination models
Sdevice provides several models for carrier recombination. The recombination model used
during this work is the Shockley-Read-Hall (SRH) model (see Chapter 2.2.4). In Sdevice
Etrap is defined in the parameter file of each material.
The carriers lifetime in the SRH model can be defined as doping, temperature, and elec-
trical field dependent. The temperature is not considered during the simulations and
the electrical field on the simulated devices in not high enough to create a considerable
influence on the lifetime. Strong electrical fields, that exceed 3 · 105 V /cm, should not
be neglected [34]. Comparative simulations are performed with and without the field
dependence factor on the carriers lifetime, that result in an insignificant change on the
photodiode responsivity. For that just the doping dependency of the lifetime is enabled.
Mobility
Sentaurus Visual is used in three situations as shown in the flowchart of Figure 3.6.
After a structure is generated by Sprocess, Figure 3.6 (a), the .tdr file can be opened in
Svisual to check the used dimensions and materials and to plot the doping profile. The
45
doping profile can also be exported to a text file and the dimensions of the oxide/nitride
layers can be measured for later use in SSE.
Sdevice output-file
Contain information of
structure dimensions, Sentaurus Visual
doping, grid in addition to Visual check electrical and
spatial and quantitative data optical simulation results.
of simulation results (optical
generation, electrical field). (c)
If the structure is generated with SSE the .tdr file can also be checked in Svisual (Figure
3.6 (b)). The structure dimensions, positions of the implants, and mesh are generated by
command line in batch mode. Possible typing or dimension mistakes on the code will just
be recognized by a visual check of the output file.
After a device simulation, Sdevice also creates a .tdr file for each simulated wavelength.
This file, in addition to the dimension, doping and grid information, includes the results
of the electrical and optical simulations, such as current density, optical generation, and
electrical field. As presented in the flowchart of Figure 3.6 (c), the results can be analyzed
in Svisual .
3.4.2 Inspect
Inspect is used to load the .plt file created by Sdevice that contains the photocurrent
of the simulated photodiodes and information of wavelength. The data can be plotted
directly in Inspect or can be exported in a proper format to Matlab.
3.4.3 Matlab
Even though Sentaurus provides a tool for plotting the results, Matlab is more convenient
when further calculation or comparison between several projects needs to be performed.
The data file generated by Inspect and the dimensions of the structures are used as input
for Matlab. Then the responsivity of each photodiode is calculated and plotted.
46
The dimensions of the 2D structure is calculated considering that Sentaurus device defines
the 3rd dimension as 1 µm. For example a structure with width of 25 µm and length of
1 µm long, has a total area of 25 µm2 .
The incident light is applied to the photodiode with wavelengths between 400 nm and
1000 nm and with a light-power of 1 W/cm2 . In order to obtain the responsivity of the
simulated photodiodes the current of each photodiode is divided by the total power. The
total power is obtained by the multiplication of the area of the photodiode and applied
power per area, as shown in Equation 3.1.
Ipd
R= (3.1)
P ower · W idth · Length
As the length of the structures is fixed to 1 µm and the applied power per area in 1 W/cm2
the responsivity is defined as shown in Equation 3.2.
Ipd [A]
R= (3.2)
W idth[µm] · 10−8
Due to the long time required for the process simulation it is not feasible to create all the
structures by simulating the complete process steps. Otherwise each time the dimension of
an implant or the ratio of opened and covered area are changed a whole process simulation
should be done. Therefore one process simulation is done where all the possible implants
47
are created. The .tdr file created by Sprocess is opened in Svisual, the created structure
is visually checked and the doping profile is extracted. Also the depth of the STI and the
material and the thickness of the oxide/nitride layers are manually extracted.
A shown in Figure 3.7 Sprocess and Svisual are used for this task. The simulation for
characterizing the technology is run just once.
During investigation and optimization each color separation technique is studied depend-
ing on the modifications of the layout. For each color sensor structure several simulations
are performed changing the size of the wells, source drain implants, changing the distance
or the position of a specific photodiode or even changing the dimensions of the dark and
opened area of the sensor.
The structures are created with SSE and the optical simulation is performed with the
Raytracer solver, as shown in the flowchart of Figure 3.8.
To ensure the accuracy between Sprocess and SSE the technology data previously ex-
tracted from the process simulation is used to create the sensor in SSE.
The oxide/nitride stack can reduce the sensor responsivity due to reflection and absorp-
tion. Therefore the stack is implemented in all simulations even though the diffraction
and interference are not modeled.
48
Visual check on the generated
With Raytracer optical solver. structure.
49
Chapter 4
Implementation
Based on the techniques of color separation presented in Chapter 2.4 four different struc-
tures are created. The first structure is based on lateral and double photodiode arrange-
ment while the second structure contains just lateral photodiodes. In both structures,
photodiodes sensitive to short-, middle-, and long wavelengths are obtained. The struc-
ture 3 and 4 provide tunable spectral responses according to an external applied voltages.
The created structures were simulated, optimized, and produced on a 130 nm CMOS test-
chip. The implemented color sensors were measured and the responsivity calculated as
presented in Chapter 4.6. The responsivities calculated out of simulation and measure-
ments are compared. In this chapter the proposed color sensor structures, the respective
working principles, simulations, and measurement results are presented.
4.1 Structure 1
The structure 1 is designed in order to provide three different spectral responses. A
combination of a double-photodiode and a laterally arranged covered photodiode is used.
It is expected that the shallow photodiode on the double-photodiode presents higher
responsivity to short wavelengths, the deeper photodiode a wide spectral response, while
the covered photodiode should respond to long wavelength light.
The cross-section of Structure 1 is presented in Figure 4.1. It consists of three photodiodes
NWD, NWL, and PPNW. NWD is the diode formed between the N-well and the substrate
and is shielded from the incident light by a metal layer. NWL is also formed between the
N-well and the substrate and is completely uncovered. PPNW has a shallow p-n junction
which is formed between the P+ and the NWL. The substrate is contacted by the P+
implant (Psub).
The structure is laid out as square cell of 25 µm x 25 µm. The top view of the sensor unit
cells is shown in Figure 4.2.
50
NWD Psub NWL PPNW NWD Psub NWD
P-Substrate
In order to increase the current provided by the photodiodes the complete sensor is formed
by an array of 10 x 10 of the unit-cells placed side-by-side and connected in parallel,
resulting in a 250 µm x 250 µm sensor. The top-view of the sensor is shown in Figure A.1,
where the gray represents the area covered by metal.
4.1.1 Simulations
Simulations are performed as described in the Section 3.5.2, where the structure is im-
proved in order to provide the highest possible light sensitive area that still allows the
color separation. All simulations are done with ideal electrical contacts biased at 0 V
DC-voltage.
When the NWD is completely covered by metal its current is formed by carriers generated
in the substrate that diffuse to the p-n junction. The grater the metal coverage the smaller
51
is the influence of the shorter wavelengths light. In other words a bigger metal coverage
results in a narrower spectral response and improves color selectivity. For this reason
it would be interesting to stretch the metal coverage, ensuring that just the carriers
generated by red light create a current in NWD. But the increase of the covered area
decreases the responsivity of the sensor. On the other hand if the opened area is extended
over the NWD, to increase the responsivity, the color separation is negatively affected.
Figure 4.3 shows the example where the structure is optimized by change of metal shield
dimensions.
Responsivity Structure 1
0.12
Optimized
Improved selectivity
0.1
Improved responsivity
0.08
Responsivity [A/W]
0.06
0.04
0.02
−0.02
400 500 600 700 800 900 1000
Wavelength [nm]
The red line is the responsivity of the proposed structure when NWD is completely cov-
ered, the blue line is the responsivity in case the metal shield is increased (covering also
Psub). This results in a narrower spectral response and a lower responsivity. The green
line represents the diode’s responsivity for a smaller metal shield (half of NWD is under
light). The responsivity is improved but the color selectivity is degraded. As a final
solution the NWD is maintained completely covered.
One disadvantage seen in maintaining the NWD completely covered is that, if the current
of NWD must be increased in relation to the current of the other photodiodes, a wider
NWD is necessary. That also means a wider covered area must be used, resulting in a
smaller overall responsivity of the sensor. The final version has, in 2D, an opened area of
79%, but due to the used layout the opened area is reduced in 3D to 62%.
The electrons current density on the structure for a light wavelength of 900 nm is presented
in Figure 4.4. Even though the light is falling just on the middle of the structure the
generated carriers reach the NWD at the borders.
The simulated responsivity of the three photodiodes is presented in Figure 4.5. The
NWD is sensitive to long wavelengths, the PPNW is sensitive to short wavelengths and
52
Figure 4.4: Electrons Current Density of Structure 1.
the NWL has a wide spectral response. The green curve, sensitive to middle wavelengths,
is obtained by a scaled subtraction of NWD and PPNW from NWL. The scaling is done
by the factor a and b shown in the legend.
Responsivity Structure 1
0.12
PPNW
NWD
NWL
0.1
NWL−a.NWD−b.PPNW
0.08
Responsivity [A/W]
0.06
0.04
0.02
0
400 500 600 700 800 900 1000
Wavelength [nm]
The performed 2D simulation presents some differences of the real 3D structure. For
example the area of the NWD in 2D represents 20% of the sensor’s total area, while in
3D it is 36%. The correction of the simulation values is estimated by simple geometric
comparisons, where the increase of the photodiode area are taken as linear increase of
the responsivity. The area of the 3 photodiodes as well as the metal shield are corrected
53
according to the 3D values.
4.1.2 Measurements
Responsivity Standard 1
0.14
PPNW
NWD
0.12 NWL
NWL−a.NWD−b.PPNW
0.1
Responsivity [A/W]
0.08
0.06
0.04
0.02
−0.02
400 500 600 700 800 900 1000
Wavelength [nm]
Three independent responsivities with peak in short-, middle-, and long wavelengths are
obtained with the subtraction of the curves. The obtained curves are much wider and do
not match exactly to the color matching functions. Nevertheless the implementation of a
color signal processing algorithm can be used in order to produce a color sensor based on
RGB or XYZ color models.
One important characteristic of this structure is that the current of PPNW is on the
opposite direction than the other currents. This should be handled by the analog to
digital converter, responsible for the read out of the sensor’s currents. Also due to the
use of the double-photodiode the current on the contact NWL is the sum of the current
on both photodiodes (NWL and PPNW).
54
4.2 Structure 2
A similar structure is created where the double-photodiode is replaced by two laterally
arranged photodiodes. This structure eliminates the problem of Structure 1 of having
current with different directions and the sum of the currents in NWD.
Instead of PPNW for the shallow p-n junction, an N+ implant in P-well (NPPW) is used.
Since the P-well under the N+ is highly doped the carriers generated on the substrate
will recombine before reaching the p-n junction. Therefore the NPPW is just sensitive
to short wavelengths similar to PPNW. The wide response previously provided by the
N-well under light can be replaced a NPL straight in the substrate. The responsivity of
the NPL is expected to be lower for blue light in comparison to Structure 1. That happens
because the shallow photodiode response is not anymore added to the wide response. The
NWD, that provides the response to long wavelengths, is maintained and the substrate is
contacted by Psub.
This structure consists of only laterally arranged diodes and is also compatible to standard
CMOS technology. The structure is shown in Figure 4.7.
P-Substrate
55
Figure 4.8: Top view of the Structure 2.
2D - Simulation 3D - Implemented
25 µm x 25 µm 79% 62%
50 µm x 250 µm 80% 77%
The 250 µm x 250 µm sensor is shown in Figure B.1, where the area covered by metal is
shown in gray.
4.2.1 Simulations
Similarly to before this structure is improved for the optimum light sensitivity and color
selectivity. The simulations are performed as described in the Section 3.5.2. The simula-
tions are done with ideal electrical contacts biased at 0 V DC-voltage.
The electron current density on the structure for a light wavelength of 900 nm is presented
in Figure 4.9, and the simulated responsivity of the three photodiodes in Figure 4.10.
The NWD is sensitive to long wavelengths, the NPPW is sensitive to short wavelengths
and the NPL has a wide spectral response with low responsivity on low wavelengths. The
green curve, sensitive to middle wavelengths, is obtained by a scaled subtraction of NWD
and NPPW from NPL. The scaling is done by the factor a and b shown in the legend.
4.2.2 Measurements
The measured spectral responsivities of the 3 photodiodes are presented in Figure 4.11,
where the dashed lines represent the simulation results and the solid lines the measure-
56
Figure 4.9: Electrons Current Density of Structure 2.
Responsivity Structure 2
0.14
NPPW
NWD
0.12 NPL
NPL−a.NWD−b.NPPW
0.1
Responsivity [A/W]
0.08
0.06
0.04
0.02
0
400 500 600 700 800 900 1000
Wavelength [nm]
ment. The spectral response of NWD is much wider than the previous solution. A
redesign, including a larger metal coverage, could be used to ensure that just the carriers
generated by red light create a current in NWD, creating a narrower and more selective
curve. Nevertheless three distinct and independent spectral responses are obtained, that
allow the transformation to a color standard.
The simulation and measurement values have an amplitude difference of approximately
20%. This difference can be caused by measurement uncertainties. The sensitivity of the
setup when tuning the focus point of the light on the surface of the measured color sensor
or the reference light-power sensor can cause variation on the responsivity amplitudes.
Nevertheless this mismatch does not affect the shape of the spectral response neither the
57
Responsivity Structure 2
0.14
NPPW
NWD
0.12 NPL
NPL−a.NWD−b.NPPW
0.1
Responsivity [A/W]
0.08
0.06
0.04
0.02
0
400 500 600 700 800 900 1000
Wavelength [nm]
ratio between the photodiodes responses. More details about the measurement setup are
presented in Chapter 4.6.
4.3 Structure 3
The Structure 3 contains the same photodiodes as Structure 2. The photodiodes are
arranged in order to improve the effect of the electrical field and to increase the opened
area. And the dimensions are adjusted according to simulation results.
The Structure 3 consists of three photodiodes NWD, NPL and NPPW. NWD is formed
between the N-well and the substrate and is shielded from the incident light by a metal
layer. NPL is formed between the N-plus and the substrate and is completely uncovered.
NPPW is a shallow photodiode formed between the N+ implant and the P-Well. The
substrate is connected by the Substrate contact (Psub). A second substrate contact (PP)
is used for tuning the spectral response. The cross-section of Structure 3 is presented in
Figure 4.12
Structure 3 allows the variation of the spectral response of NWD and NPL by applying a
voltage to PP. The voltage difference between PP and Psub generates a substrate current
and an electrical field. The electrical field redirects the photo-generated carriers according
to its direction.
The spectral response in Structure 3, without applying an electrical field, is expected to
be the same as Structure 2. When a positive voltage is applied to PP the current in NWD
should increase. When a negative voltage is applied to PP the current in NPL should
increase. The current in NPPW should not be affected, because it is shielded from the
substrate by the highly doped P-well.
58
Psub NPPW NPL NWD PP NWD NPL NPPW Psub
P-Substrate
4.3.1 Simulations
The electron current density of the structure for a light wavelength of 900 nm is presented
in Figure 4.13. Figure 4.13a presents the current density when PP is biased with 0 V,
the current density of minority carriers is distributed between NWD and NPL. In Figure
4.13b, when a negative voltage is applied to PP, an electrical field is built up from Psub to
PP, causing the photo-generated carriers to be redirected from the center to the borders
of the structure. When a positive voltage is applied to PP the electrical field is built up
from PP to Psub causing the currents to be redirected from the borders to the center of
the structure, as shown in Figure 4.13c.
The simulated responsivity of the three photodiodes is presented in Figure 4.14 and shows
the expected behavior. The NWD is sensitive to long wavelengths, the NPPW is sensitive
to short wavelengths and the NPL has a wide spectral response. Depending on the voltage
applied to PP the responsivities of NWD and NPL can be tuned. The simulation also
shows that when the responsivity of NWD increases the one of NPL decreases and vice-
versa.
4.3.2 Measurements
The measured spectral responsivity of the 3 photodiodes, in comparison to the simulated
values is presented in Figure 4.15. Where the dashed lines represent the simulation results
and the solid lines the measurement.
The biggest responsivity variation during the measurements was obtained by changing
PP voltage from −1.5 V to 3 V. Voltages lower than −1.5 V did not present any further
changed on the responsivities, voltages higher than 3 V were not used to avoid damaging
the structure.
The responsivities tuning range from the measurement needs lower voltages in the simula-
tion. This difference between simulation and measurements can be explained by different
59
(a) PP=0. (b) PP Negative.
(c) PP Positive.
Figure 4.13: Electrons Current Density of Structure 3 for different voltages applied to PP.
Responsivity Structure 3
0.25
NPPW PP Positive
NPL
NWD PP Negative
0.2
Responsivity [A/W]
0.15
0.1
PP Positive
0.05
PP Negative
0
400 500 600 700 800 900 1000
Wavelength [nm]
Figure 4.14: Simulated Responsivity of Structure 3 for different voltages applied to PP.
substrate doping. The substrate doping and consequently the substrate resistance can
change due to process variations.
The comparison between simulation and measurement, for approximately the same re-
sponsivity variation, shows that the measured responsivities of NWD and NPPW match
with the simulations. For NPL the simulation shows a big variation depending on PP
voltage that is not seen in the measurement. Figure 4.15 (a) shows the spectral response
for PP biased at 0 V, Figure 4.15 (b) for negative voltage on PP, and 4.15 (c) for a positive
voltage on PP.
60
Responsivity Structure 3 Responsivity Structure 3 Responsivity Structure 3
0.12 0.12 0.12
NPPW NPPW NPPW
NPL NPL NPL
NWD NWD NWD
0.1 0.1 0.1
Responsivity [A/W]
Responsivity [A/W]
0.06 0.06 0.06
0 0 0
400 500 600 700 800 900 1000 400 500 600 700 800 900 1000 400 500 600 700 800 900 1000
Wavelength [nm] Wavelength [nm] Wavelength [nm]
Responsivity Structure 3
0.12
NPPW
NPL
0.1 NWD
0.08
Responsivity [A/W]
0.06
0.04
0.02
−0.02
400 500 600 700 800 900 1000
Wavelength [nm]
Figure 4.16: Measured responsivity of Structure 3 for PP voltage steps of 225 mV.
The fact that the responsivity of NPL is not decreasing on the same level NWD increases
shows that the created electrical field is not simply redistributing the photo-generated
carriers between the two p-n junctions. But it also contributes to the increase or decrease
of the number of carriers collected by the p-n junctions.
When no electrical field is applied, the carriers generated deep in the substrate and far
away from the NWD, recombine before reaching the space charge region. When a positive
61
voltage is applied to PP the carriers are moved to the center of the structure, close to
the deep p-n junction, thus increasing the chances of reaching a SCR to contribute to the
photocurrent.
On the other hand, when a negative voltage is applied to PP the carriers generated in
the substrate under the NWD are moved away to the border of the structure where
just shallow p-n junctions are available. The carriers diffusing from the substrate will
recombine before reaching the SCR of the shallow p-n junctions due to the highly doped
P-well.
Even though this structure can be used to improve the color signal transformation it
requires a more complex digital processing. Additionally to the color signal process,
the digital block should control the applied voltage and read the responsivity for several
voltages. Another disadvantage of this solution is the produced substrate current on the
color sensor. The measured substrate currents reach up to 10 mA. It increases the current
consumption and can disturb the circuitry near the color sensor, as for example the analog
to digital converter.
4.4 Structure 4
A second structure with tunable spectral responsivity, Structure 4, is build by applying
external bias voltages to three polysilicon electrodes. Due to the field effect the voltages
applied to the electrodes create a voltage gradient within the silicon. The generated
carriers should move from the region with lower voltage to the region with higher voltage.
The proposed structure consists of a shallow photodiode NPPW, a lighted photodiode
NPL, and a third photodiode shielded from the light NPD. Additionally three electrodes
E1, E2, and E3 are implemented in between NPL and NPD as shown in Figure 4.17.
P-well
P-Substrate
NPPW in sensitive to short wavelengths, NPL has a wider response over the whole simu-
lated range, and NPD is sensitive to long wavelengths. The three electrodes can be used
62
to tune the responsivities of NPL and NPD. When a higher voltage is applied to E1 the
current in NPD should increase. When a higher voltage is applied to E3 the current in
NPL should increase. The use of the electrodes does not cause any increase or decrease of
the total responsivity, but changes the spectral response by sharing the generated carrier
between NPL and NPD.
4.4.1 Simulations
The electron current density on the structure for a light wavelength of 900 nm is presented
in Figure 4.18 (a). Figure 4.18 (b) presents the current density on the highlighted region
NPL E1 E2 E3 NPD
(b)
NPL E1 E2 E3 NPD
a) (c)
Figure 4.18: Electrons Current Density of Structure 4 for different electrode voltages.
when the electrodes E1, E2, and E3 are biased respectively with 0 V, 1.5 V, and 3 V.
A high electron current density can be observed under E2 and E3 that represent the
inversion channel created under the STI. The inverted region works as an extension of the
NPL, therefore collects more carriers that NPD.
Figure 4.18 (c) presents the current density when the electrodes E1, E2, and E3 are
biased respectively with 3 V, 1.5 V and, 0 V. A high electron current density can be
observed under E1 and E2 that represent the inversion channel created under the STI.
The inverted region works as an extension of the NPD, that in this case collects more
carriers that NPL.
The simulated responsivity of the three photodiodes is presented in Figure 4.19.
Depending on the voltage applied to the electrodes the responsivities of NPD and NPL
can be tuned. In the three simulation results presented, the electrode voltages are set as
shown in Table 4.2.
63
Responsivity Structure 4
0.12
PPNW
NPL
NPD
0.1
0.08
0.04
0.02
0
400 500 600 700 800 900 1000
Wavelength [nm]
Table 4.2: Applied voltages for tuning the spectral response in Structure 3.
E1 E2 E3
Voltage set 1 3V 1.5 V 0V
Voltage set 2 1.5 V 1.5 V 1.5 V
Voltage set 3 0V 1.5 V 3V
A higher voltage in E1 increases the current in NPL and a higher voltage in E3 increases
the current in NPD. The simulation also shows that when the responsivity of NPD
increases the one of NPL decreases and vice-versa.
The use of the electrodes to tune of the responsivities of NPD and NPL just creates on
these two curves a scaling adjustment. But if part of NPD is subtracted from NPL the
obtained green curve has the responsivity peak changing from 850 nm to 630 nm.
4.4.2 Measurements
The measurement responsivities of Structure 4 are plotted for the same voltage sets pre-
sented in Table 4.2. Figure 4.20 shows the measurement results compared to the simula-
tions where the measurement results are shown by the solid lines and the simulations by
the dashed lines. In Figure 4.20 (a) the measurement and simulation for voltage set 2 is
presented, in Figure 4.20 (b) for voltage set 1, and in Figure 4.20 (c) voltage set 3.
Several spectral response curves can be obtained if the voltage at the electrodes are
changed by small steps, as presented in Figure 4.21. The initial measurement is defined
as voltage set 1 (See Table 4.2) and the final voltages as set 3. The voltage on E2 is
64
Responsivity Structure 4 Responsivity Structure 4 Responsivity Structure 4
0.14 0.14 0.14
PPNW PPNW PPNW
NPL NPL NPL
0.12 NPD 0.12 NPD 0.12
NPD
Responsivity [A/W]
Responsivity [A/W]
0.08 0.08 0.08
0 0 0
400 500 600 700 800 900 1000 400 500 600 700 800 900 1000 400 500 600 700 800 900 1000
Wavelength [nm] Wavelength [nm] Wavelength [nm]
Responsivity
0 0 0
400 500 600 700 800 900 1000 400 500 600 700 800 900 1000 400 500 600 700 800 900 1000
Wavelength [nm] Wavelength [nm] Wavelength [nm]
Figure 4.21: Measured responsivity of Structure 4 for electrode voltage steps of 25 mV.
On the last Figure 4.21 can be seen that the responsivity tuning saturates when the
voltage of NPD decreases and NPL increases. Nevertheless the maximum and minimum
values, that determine the tuning range of the responsivities, are the same in simulation
and measurement.
The responsivity shift of the subtracted curve (NPL-NPD) is shown in Figure 4.21 (c).
The shifting of the responsivity peak from 850 nm to 630 nm is also observed.
In comparison to Structure 2 a more representative tuning of the spectral response is
obtained, since the responsivity peak has a large variation. Structure 4 provides a mech-
anism of tuning the spectral response that does not require extra current consumption.
On the other hand the control of the applied voltage is still required.
65
4.5 Results summary and analysis
The four structures presented are completely compatible with standard CMOS technolo-
gies, suitable for low-cost color sensing applications.
Structure 1 presents, amongst all the structures, the best color selectivity to long wave-
lengths. Additionally a simple subtraction can be used to obtain response to middle
wavelengths with peak around 600 nm. On the other hand it has one current in the
opposite direction, which can be a disadvantage for the analog to digital converter.
Structure 2 does not contain double photodiodes, consequently all of the current have
the same direction. However the photodiode sensitivity to long wavelengths presents a
wider spectral response. Also the peak of the middle wavelength response is shifted to
approximately 700 nm. This can have a negative effect the color transformation, resulting
on a less accurate sensor.
The third structure presents as standard the same responsivities of structure 2, but an
additional tuning of the NWD responsivity is implemented. A great variation on the
responsivity can be obtained in addition to a precise control depending on the applied
voltage. The additional color information obtained from the several responsivities con-
tribute to color transformation accuracy, at the expense of a significantly high substrate
current.
Structure 4 presents even better characteristics of the responsivity tuning. The subtracted
curve presents a clear shift on the responsivity peak, besides the variation on the spectral
response breadth, thus improving the color transformation. In contrast with structure 3,
no substrate current is required to provide the responsivity tuning.
66
Figure 4.22: Picture of the Testchip.
Light source passes through the monochromator and is focused by the lenses on the surface
of the color sensor for the first measurement and to the surface of the power-sensor for
the second measurement.
Powersensor
or
Lense 150mm Lense 250mm
Color sensor
Illuminator
Light source
Monochromator
1500µm slit
For the measurement of the photodiode currents the color sensor is placed in front of the
light beam and the light is focused on the surface of the color sensor. The photodiodes are
connected to three sourcemeters. The light beam is adjusted to cover the whole area of
the color sensor (250 µm x 250 µm). For the measurement of the light power, the reference
power-sensor is placed in front of the light beam, which is focused on the surface of the
power-sensor. The beam passes trough a pinhole of 1 mm of diameter.
The sourcemeters, the monochromator, and the power-sensor are connected to a computer
and the measurement is automated and controlled by Matlab. The Monochromator is
configured to select light with center wavelengths from 400 nm to 1000 nm in steps of
67
10 nm. For each wavelength step the photocurrent or the light power, depending on the
measurement being performed, are stored in the computer.
After the two measurements are concluded the currents of the three photodiodes and
the light power in dependence on the light wavelength are available. That enables the
calculation of the responsivity as shown in Equation 4.1.
Iph
R= (4.1)
Ptot
Since the area of the color sensor and the lighted area of the power-sensor are not the
same, the power read from the sensor cannot be directly used. Assuming that the light
intensity is uniformly distributed over the beam, just the ratio of the two areas must be
calculated. The area of the color sensor is 0.062 mm2 (0.25 mm · 0.25 mm) and the lighted
area of the power-sensor is 0.78 mm2 (π · 0.5 mm2 ), resulting in an area ratio of 12.56. The
responsivity of the three photodiodes is calculated as shown in Equation 4.2.
Iph
R= · 12.56 (4.2)
Ptot
Since two measurements are performed, the power measured by the power-sensor (Ptot )
is not exactly the same used to create the photocurrents (Iph ). Time and temperature
variations of the light power, between the first and second measurement, can lead to
errors on the calculation of the responsivity. In order to minimize these errors the power
measurement is performed immediately after the current measurement.
Another factor than can introduce errors in the measurements is the need of moving the
color sensor away and place the power-sensor in front of the light beam. To ensure the
same power in both measurements, the surface of both sensors should be at the same
position. Also the color sensor as well as the pinhole in front of the power-sensor should
be completely under light. If the adjustment is not precise an error will be obtained on
the amplitude of the responsivities. This does not affect the shape, but just the amplitude
of the spectral response.
A picture of the measurement setup is shown in Figure 4.24 and the list of equipment
according to the picture are presented in Table 4.3.
68
Table 4.3: Measurement setup description.
69
Figure 4.24: Picture of the measurement setup.
70
Chapter 5
Future outlook
The development of color sensing structures was carried out as proposed. Nevertheless
there is still some room for improvement of the optical simulations as well as for the
development of the signal processing that is in charge of matching the obtained sensor
responses to the XYZ standard. In the following chapters a description of both topics
that can be used as initial point for upcoming research works is presented.
71
Responsivity Standard 1 Responsivity Standard 1
0.12 0.14
PPNW PPNW
NWD NWD
NWL 0.12 NWL
0.1
0.1
0.08
Responsivity [A/W]
Responsivity [A/W]
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0 0
400 500 600 700 800 900 1000 400 500 600 700 800 900 1000
Wavelength [nm] Wavelength [nm]
(a) (b)
Figure 5.1: (a) Simulated responsivity of Structure 1 with Raytracer optical solver. (b)
Measured responsivity of Structure 1 with interference caused by oxide layers.
Responsivity Standard 1
0.16
PPNW
NWD
0.14 NWL
0.12
Responsivity [A/W]
0.1
0.08
0.06
0.04
0.02
0
400 500 600 700 800 900 1000
Wavelength [nm]
The simulation using BPM as optical solver requires complex definition of the structure
boundaries and often due to the several layers and different materials used the simulation
does not converge. Furthermore small changes in the thickness of the oxide/nitride layers
cause big changes in the simulation, for that a process simulation is required for more
accurate results.
The initial simulation of the oxide/nitride stack interference let some open points that can
be taken as an extension of the current research. Further investigation of the BPM optical
solver and understanding of the optical simulation would be required for the improvement
of the simulation results and a satisfactory matching to the real structure.
72
5.2 Color Signal processing
Ideally the responsivity of the produced color sensor should match to either XYZ or RGB
color-match function in order to be able to sense color accurately. The matching than
can be provided by a transformation matrix.
As shown in Figure 5.3 (a) the multiplication of the input Stimuli S by the RGB color
matching function generates the tristimulus output (TR , TG , TB ). The same Stimuli input
when multiplied by the responsivity of the three photodiodes of the produced sensor also
generate three outputs (TR0 , TG0 , TB0 ), different from the previous as shown in Figure 5.3
(b).
RGB Color-
matching function
S(λ) r(λ) TR(λ) TR(λ)=∫S(λ)r(λ)dλ
Transformation
Matrix
Non-ideal Responsivities of the
Produced sensor
S(λ) R1(λ) T’R(λ) T’R(λ)=∫S(λ)R1(λ)dλ
Figure 5.3: Use of transformation matrix to match the output of the produced color sensor
to the RBG tristimulus for a unique input.
A transformation matrix can be used to match TR0 , TG0 , TB0 to TR , TG , TB , where the matrix
M is a well-defined unique matrix that presents zero error.
If the same transformation matrix is used to convert the tristimulus data from an input
light with different spectra a wrong result is obtained.
73
To allow the sensor to represent light color of different spectra a statistical training can
be used as shown in Figure 5.4. Several different input light sources (S1 , S2, S3 ...Sn )
are used to create a set of outputs for the RGB color matching function (TR1 , TG1 , TB1
0 0 0
to TRn , TGn , TBn ) and a set of outputs to the produced color sensor (TR1 , TG1 , TB1 to
0 0 0
TRn , TGn , TBn ). Since this set of linear equations is over defined, there is no unique solution
for the transformation matrix. Therefore an estimation of the matrix with minimized error
can be calculated using the pseudo-inverse transformation.
RGB Color-
matching function
Non-ideal Responsivities
of the Produced sensor
1 R1(λ) T’R1(λ) T’R2(λ) T’R3(λ) ... T’Rn(λ)
2
R2(λ) T’G1(λ) T’G2(λ) T’G3(λ) ... T’Gn(λ)
3
n
R3(λ) T’B1(λ) T’B2(λ) T’B3(λ) ... T’Bn(λ)
Figure 5.4: Use of training patterns to generate a transformation matrix suitable for dif-
ferent light spectra.
The signal processing of the color sensor is a topic to be further developed. The appropri-
ate training pattern, the generation of an optimized transformation matrix, as well as the
restrictions caused by the limited number of bits available for the transformation matrix
can be investigated in continuity to the work already developed.
74
Chapter 6
Conclusion
Integrated color sensors are widely used the industrial and mass market applications.
Consumer market-related applications normally require low cost solutions and small area.
A CMOS sensor can meet these needs due to the possibility of a monolithic integration
of the color detector, the read-out circuit, and the digital signal processing. In order
to develop a low cost solution for color sensing application, modifications of the CMOS
technology and color filters should be avoided. Therefore the innovative methods of color
separation investigated and developed in this thesis are entirely based on standard CMOS
technology.
This project was undertaken to design color sensing structures based in standard CMOS
technology. Three important steps are taken for the development of this work. The first,
includes the modeling of the process flow and of the optical and electrical characteristics
of semiconductors. The second step is the investigation of the proposed color separation
techniques. And as a last step, color sensing structures are designed based on the proposed
techniques, simulated and optimized using the TCAD tools, and produced in 130 nm
CMOS technology.
The most significant finding that emerges from this study is that, color sensors can be
developed in fully standard CMOS technology by making use of silicon depth-dependent
carriers generation and carriers lateral diffusion. It was also shown that the lateral carriers
diffusion can be enhanced and controlled by a substrate current or by voltage-biased
electrodes.
Finally, some limitations of the proposed solutions need to be considered. Firstly, a
covered area is required for the long wavelengths detection. This coverage decreases the
sensor efficiency. A compromise between color selectivity or sensor efficiency must be
found. Secondly, the color sensor responsivities do not match exactly the color matching
function of the RGB or XYZ standard. Therefore a color signal transformation is required.
The basic idea of color transformation based on a pseudo-inverse matrix is presented.
75
Nevertheless further work needs to be done to establish the appropriate training sets and
study the accuracy of the color transformation. In addition to the color signal processing,
the modeling of the optical behavior of the structures is still to be improved. The used
optical solver does not model the diffraction and interferences, caused mostly by the
oxide/nitride stack on the top of the sensor. Future research should therefore concentrate
on the investigation of the stack on the sensor’s surface and its modeling with adequate
optical solvers.
In summary, four color sensing structures, based on different color separation techniques
are presented. The structures were designed and produced in standard CMOS technology
with no process modifications. The simulation and measurement of the four structures
demonstrate photodiodes sensitivity to short-, middle-, and long wavelengths, required for
color recognition. Therefore the proposed solutions are competitive solution for low-cost
integrated color sensors.
76
Bibliography
[2] J. Sturm. ”Photodiode Modeling and Optoelectronic Integrated Circuits for Optical-
Data-Storage Applications” Dissertation, Technischen Universitt Wien, Vienna,
2006.
[6] S. M. Sze. Physics of Semiconductor Devices. 2nd Edition. New York: John Wiley &
Sons, 1981.
[8] D.A. Johns, K. Martin. Analog Integrated Circuits Design.John Wiley & Sons, 1997.
[9] Y. Audet and S. S. Aboutorabi. ” CMOS Process Compatible Color Sensor using
Wavelength Dependent Absorption Depth,” presented at the 2nd Circuits and Sys-
tems and TAISA Conference, Montreal, 2008.
[13] R.B.Merrill, Color Separation in an Active Pixel Cell Imaging Array Using a Triple-
Well-Structure, US Patent 5,965,875, 1999
77
[14] R.F.Lyon and P.M.Hubel, Eyeing the Camera: into the Next Century, Tenth Color
Imaging Conference: Color Science and Engineering Systems, Technologies, Appli-
cations, Scottsdale, Arizona; November, 2002.
[21] Synopsys. Sentaurus Device Electromagnetic Wave Solver User Guide, Jun. 2009.
[23] Synopsys. Sentaurus Process User Guide, Sep. 2011, pp. 715-721.
[24] Synopsys. Sentaurus Process User Guide, Sep. 2011, pp. 77-178.
[25] Synopsys. Sentaurus Process User Guide, Sep. 2011, pp. 179-340.
[26] Synopsys. Sentaurus Process User Guide, Sep. 2011, pp. 575-620.
[27] Synopsys. Sentaurus Process User Guide, Sep. 2011, pp. 688-709.
[29] Synopsys. Sentaurus Device User Guide, Jun. 2009, pp. 178-179.
[30] Synopsys. Sentaurus Device User Guide, Jun. 2009, pp. 393-472.
[31] Synopsys. Sentaurus Device User Guide, Jun. 2009, pp. 473-477.
[32] Synopsys. Sentaurus Device User Guide, Jun. 2009, pp. 477-483.
[33] Synopsys. Sentaurus Device User Guide, Jun. 2009, pp. 267-305.
[34] Synopsys. Sentaurus Device User Guide, Jun. 2009, pp. 309-320.
78
[35] s. Westland, C. Ripamonti Computational Colour Science using MATLAB. Chich-
ester: John Wiley & Sons Ltd, 2004, pp 4-6.
[36] J. Schanda Colorimetry undertanding the CIE system. Hoboken,NJ: John Wiley &
Sons, 2007, pp 29-32.
[39] ”CIE 1931 color space”. Internet: http://en.wikipedia.org/wiki/CIE 1931 color space.
[Jun 10, 2013].
79
List of Abbreviations
80
NWD N-well Dark
NWL N-well under Light
NPL N+ implant under Light
PP second substrate contact
Psub Substrate contact
NPD Dark N+ implant
SCR Space Charge Region
NPPW N+ implant in P-well
SSE Sentaurus Structure Editor
Sprocess Sentaurus Process
Sdevice Sentaurus Device
BPM Beam Propagation Method
EMW Sentaurus Electromagnetic Wave Solver
Svisual Sentaurus Visual
2D Two-Dimensional
3D Three-Dimensional
Tcl Tool command language
H2 O Water
O2 Dioxygen
N2 O Dinitrogen oxide
SiO2 Silicon dioxide
Si Silicon
CMP Chemical-Mechanical Polishing
FDTD Finite-Difference Time-Domain
FFT Fast Fourier Transform
SRH Shockley-Read-Hall
DC Direct Current
81
List of Symbols
82
Symbol Description Unit
P (y) Optical power in dependence of depth W
PT ot Total incident power W
PRef l Reflected optical power W
PT rans Transmitted optical power W
q electron charge As
R̃ Reflection coefficient
R Responsivity A/W
T Transmission coefficient
t time s
U Recombination Rate cm−3 /s
v˜n Drift velocity of electrons cm/s
v˜p Drift velocity of holes cm/s
vnsat Saturation velocity of electrons cm/s
vpsat Saturation velocity of holes cm/s
VT Thermal Voltage ( kT q
) V
x x direction or or width µm
y y direction or thickness µm
α Absorption coefficient µm− 1
∆ Laplace operator
∇ Divergence operator
r Relative dielectric constant
ε Electrical permittivity F/cm
η Quantum efficiency %
λ light wavelength in a medium nm
λ0 light wavelength on vacuum nm
ν frequency of light in a medium Hz
µn Electrons mobility cm2 /V s
µp Holes mobility cm2 /V s
µ0n Electrons mobility without electrical field cm2 /V s
µ0p Holes mobility without electrical field cm2 /V s
Ψ Potential V
ρ Charge density As/cm−3
τn Electrons lifetime s
τp Holes lifetime s
83
Appendix A
Structure 1 Layout
PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW
PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW
NWL NWL NWL NWL NWL NWL
PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW
NWL NWL NWL NWL NWL NWL
PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW
PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW
250um
PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW
NWL NWL NWL NWL NWL NWL
PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW
NWL NWL NWL NWL NWL NWL
PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW
NWL NWL NWL NWL NWL NWL
PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW
NWL NWL NWL NWL NWL NWL
PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW PPNW
NWL NWL NWL NWL NWL NWL
250um
84
Appendix B
Structure 2 Layout
250um
250um
85