DESIGN OF PIXEL LEVEL CMOS READOUT CIRCUITRY FOR CONTINUOUS BIAS UNCOOLED BOLOMETRIC LWIR FOCAL PLANE ARRAYS

READOUT CIRCUITRY FOR CONTINUOUS BIAS UNCOOLED BOLOMETRIC LWIR FOCAL PLANE ARRAYS Troy Alexander Chesler, Master of Science, 2004 Thesis Directed By: Professor Martin Peckerar, Departm

READOUT CIRCUITRY FOR CONTINUOUS BIAS UNCOOLED BOLOMETRIC LWIR FOCAL PLANE ARRAYS

Troy Alexander Chesler, Master of Science,

2004

Thesis Directed By: Professor Martin Peckerar, Department of

Electrical and Computer Engineering

Modern IC foundries do not provide large analog storage capacitors and therefore limit the charge well capacity for modern LWIR infrared readout integrated

circuits A technique for increasing the effective capacity and integration times while

maintaining linearity for a continuous biased bolometric LWIR focal plane arrays is presented Increasing integration times in the pixel reduces noise bandwidth and effectively increases dynamic range without saturating the integration capacitor Background suppression techniques, such as current skimming and charge subtractionreduce the overall temporal noise by subtracting the DC input current This thesis addresses the following: a) the practicality of continuous bias uncooled pixel designs b) optimum skimming implementation and c) the impact of background suppression

on temporal and spatial noise

ii

DESIGN OF PIXEL LEVEL CMOS READOUT CIRCUITRY FOR CONTINUOSBIAS UNCOOLED BOLOMETRIC LWIR FOCAL PLANE ARRAYS

By

Troy Alexander Chesler

Thesis submitted to the Faculty of the Graduate School of the

University of Maryland, College Park, in partial fulfillment

of the requirements for the degree of

[Master of Science]

[2004]

Advisory Committee:

Professor Martin Peckerar, Chair

Professor Pamela Abshire

Professor Timmer Horiuchi

iii

© Copyright by[Troy Alexander Chesler]

[2004]

Acknowledgements

1) Family

iv

Table of Contents

Acknowledgements Table of Contents

vi

Chapter 1: Uncooled Infrared Imaging

Thesis Contributions:

Background

Origins of Uncooled Thermal Imaging

Military & Commercial Developments

Bolometer Operation

Pulse Biased

DC Biased

Temperature Sensitive Resistive Materials xvii

Vanadium Oxide (VO2) xviii

Amorphous Silicon (α-Si) xviii

Uncooled Systems: Figures of Merit

Responsivity

NEP

D*

NE∆T & TCR

Thermal Time Constant xxii

Chapter 2: Fundamentals of Infrared Imaging xxiii

Introduction xxiii

Wavelength Spectrum & Atmospherics xxvi

Readout Systems xxviii

Vdet Readout Process Electronics xxviii

Chapter 3: Background Suppression in the LWIR xxxiii

xxxiii

Justification for Current Skimming & Charge Subtraction xxxiii

Spatial & Temporal Noise xxxiv

Sensitivity & Dynamic Range xxxvii

DC Suppression Concepts xxxix

Current Skimming

Fundamentals

Self-Biased Cascode Origin & Fundamentals xlvi Analysis of Pixel Circuitry

DC Bias Point of DI FET

Conductances of DI FET

DC Bias Point of SCFET

Conductances of SCFET

Single MOS Skimming Pixel

Self-Cascode Skimming Pixel

Self-Cascode with MOS Capacitor

Pixel Comparisons

Charge Subtraction Concept with Non-Linear MOSCAP

MOS Capacitor System

Circuit Description

Chapter 4: Associative Issues & Limitations with Current Skimming lxiii Subthreshold Transistor Mismatch lxiii Subthreshold Region Operation

vii

Current Mismatch vs Voltage Mismatch lxviiiBody Effect in Self-Cascode lxixEffects of MOS Scaling lxxiPower Supply lxxiiGate Oxide Thickness lxxiiiLayout &Geometry of SCFET lxxiv Noise lxxviiTemporal Noise lxxviii Spatial Noise lxxviiiSensitivity lxxixNE∆T lxxxBolometer Bias Variations lxxxiChapter 5: Signal & Noise lxxxiiSignal lxxxiiBolometer Transfer Function lxxxiiPixel Transfer Function lxxxiiNoise Sources lxxxiiFlicker Noise lxxxiiiShot Noise lxxxivReset Noise lxxxvThermal Noise lxxxviii Theoretical Noise Analysis lxxxviiiSingle MOS lxxxviiiSelf-Cascode Current Mode Skimmer lxxxviiiChapter 6: IC Test CHIP Design lxxxixIRCHIP1 lxxxixChip Description & Function lxxxix Simulation & Test Results xciiIRCHIP2 xcixChip Description & Function xcix Test Results xcixIRCHIP3 xcixChip Desription & Function xcixTest Results xcixFuture Considerations 100Conclusion 101References 102

viii

Chapter 1: Uncooled Infrared Imaging

Origins of Uncooled Thermal Imaging

Feeling heat waves emitted from hot objects yet seeing no visible light gave clues to early scientists that light is a form of radiation and exists beyond oureye’s response range These heat waves emanating from a recently distinguished fire, a hot rock, or molten iron ore elucidated the emissive nature of

“invisible”(infrared) light

Military & Commercial Developments

Before World War II, there had been considerable research for uncooled imaging systems including sensing materials, electrical readout, and system packaging The two most prominent areas of uncooled materials research are 1)

ferroelectric & pyroelectric bolometry and 2) resistive bolometry [1] The

pyroelectric effect (not covered in this study) is a result of electrical polarization

of opposing faces on certain types of polarized crystals that can have a “transient”electrical charge induced by means of a change in temperature [2] The resistive bolometer, which is the detector of choice for this study, is essentially a

ix

temperature sensitive resistor As the resistor absorbs electromagnetic radiation it heats up and changes its electrical properties thus changing the electrical

resistance Usually, the material will be a thin metal or semiconductor film suspended over a air gap to provide adequate thermal isolation and minimize thermal conductance to its surroundings The thin film temperature rises when radiations impinges on the detector element With a rise in detector temperature, ifthe sensing film is a metal, the resistance increases while semiconductor film’s resistance decreases In the absence of radiation, the detector temperature

decreases which causes the resistance to decrease (metal) and increases

(semiconductor) [2] For the remainder of this thesis, characterization of the CMOS readout process will focus only the semiconductor resistive bolometric detector

Efforts began in the mid 1970’s to create room temperature 2-D imaging arrays for both military and commercial applications At the time, cooled arrays had much better performance However, for most staring applications NE∆T’s (discussed later) of 0.01 K° to 0.1 K° with f/1 optics would be adequate for most military applications [4] 1983 marked the year for Honeywell’s vanadium oxide (VOx) bolometric arrays implementing a solid-state readout This was one of the

first attempts of a silicon MEMS device, (Micro-Electro-Mechanical System)

which promised to offer a new era of low cost monolithic imaging arrays By implementing silicon micro-machining, the highest thermal isolation could be attained thus leading to high array performance [4] Throughout the 70’s and 80’s,most of this research was classified under DOD guidelines for the military The

x

US Army Night Vision Laboratory (now NVESD) and DARPA saw a great advantage for developing potentially low-cost monolithic uncooled infrared focal plane arrays Moving from bump bonding the detector array to the readout to a monolithic approach improved the cost and yield of the array

An important point that allowed uncooled infrared imaging to flourish from the 80’s, 90’s, and up until 2004 was the rapid improvement of the silicon

process technologies (e.g Moore’s Law) The move to using CMOS over mature

bipolar readouts provided better cost effective arrays with higher overall

performance advantages (e.g lower power consumption) BiCMOS processes are

expensive and require several additional mask steps Chief reasons for CMOS over bipolar and BiCMOS for uncooled applications are: 1) effective cost 2) availability 3) yield and 4) well defined and easy access to process parameters The industry standard is currently CMOS, however this does not discount bipolar

or BiCMOS readouts for future applications Moore’s Law has scaled the device dimensions to sub-micron levels and now 1000 x 1000 element arrays or greater are realizable today There are have been considerable advancements in MEMS and material science providing extreme precision and high performance out of uncooled infrared focal plane arrays

Bolometer Operation

The mathematical expressions describing the physics of bolometer

operation are not the focus of this thesis The process is governed by several parameters and constants while being controlled by several partial differential

xi

equations Therefore, the mathematics will be kept to minimum and only used to

emphasize a point If the reader is interested please consult sources [1,2,4]

The transduction of (long wave) infrared radiation into electrical signals is described by the following two circuits The following two circuits considered explain different methods of bolometer

operation Figure 1.1a shows a constant bias voltage VB and Figure 1.1b shows an ideal constant bias current IB

The detector resistance is a function of temperature Rbol (T) In figure 1.1b

the potential drop appearing across the bolometer is V b =I R b bol (Ohm’s Law)

Therefore a change in detector voltage due to a tiny change in bolometer

α = − is called the (TCR) temperature coefficient of resistance

(discussed in figures of merit section) From figure 1.1a, the current through the

bolometer is b b

bol

V I

R

= A tiny change in bolometer current due to a tiny change in

bolometer temperature results in:

• G mech is the total thermal conductance due to the bolometer support legs

• W s is power absorbed by the bolometer due to any radiation that

represents the interesting signal component In infrared imaging systems,

Ws is dependent on camera optics, infrared passband, detector area,

absorption efficiency, and scene temperature variation

• T bol is the bolometer temperature suspended over the readout and Tsub is the underlying substrate temperature

• C th is the total heat capacity of the bolometer The bolometer is composed

of layers of individual material each with its own heat capacity

• V b I b is the power dissipated by the bolometer due to the bias current This

is the unwanted signal and has no practical information content for the readout to process

Equation (1.3) has the well known solution which consists of the

homogeneous solution Ae−τt where the time constant is th

mech

C G

particular solution caused by G mech sub T is T The temperature response due to sub

absorbed infrared radiation is governed by a first order low pass filter with a

xiii

3dB cutoff frequency of ω3dB 1

τ

= The temperature response as a function of

frequency caused by a small amount of absorbed radiation is:

( ) 1

( )

1

s mech

W

T T

ωω

ωτ

+ (1.4)where the absorbed infrared radiation signal W s( )ω is sufficiently small so that it induces small temperature changes in the bolometer Also, these small changes are such that the current response of the bolometer is able to reach steady state and establish an equilibrium point of operation The current response of figure 1.1a and b respectively is the following:

perfect black body) and the pass band is 8 um to 12 um, this causes

approximately a 10 mK change in bolometer temperature If the TCR = 2.5% and our desired resolution of scene temperatures is 20 mK we see a 200 uK change in bolometer temperature Therefore the ratio of changes in bias voltage or current to either a bias voltage or current is:

xiv

B ; b

I V

(1.6)

which says that α∆ =T (0.025)(2 10 )x − 4 =5 parts per million Take the

situation for an uncooled bolometer operation at 300 K where the bias current

≈ 20nA Given the above information and using equation (1.2) we find that the change in current as a result of a 200 uK change in bolometer temperature with a 20 mK resolution the

variation in there resistances (process induced mismatch) This leads to a term called fixed patter noise The second current is the most important signal and contains the relevant scene changes an actual infrared absorption Therefore, this thesis introduces a way to subtract the unwanted DC bias current and

integrate only the signal of interest This method is known has current

In pulse biased systems, integration time is determined by the line read time defined as:

frame line

rows

τ

τ =

Σ (2.1)For example, in a 640 x 480 array, at 60 Hz frame rate the line read time is34.7 us As a consequence, pulse biased integration times are not very long In pulse bias operation the detector resistances are in the range of 20-500 kΩ The electrical bias (1 V or more) causes joule heating of the sensor as well as heat from the incoming infrared radiation The bolometer’s temperature rises several degrees within the pulse duration Thus in pulse bias operation there is no stable equilibrium point The infrared signal is several orders of magnitude lower than

the generated bias signal Power dissipation for pulse bias operation is on the order of 10 -6 W.????????

DC Biased

Detector bias is continuously on so that a stable operating point is reached Small fluctuations in the observed scene induce the same fluctations around the

bolometer operating point set by DC bias across the bolometer Detector

resistances are on the order of 10-60 MΩ We assume that the substrate which houses the readout circuitry is thermally stabilized to some known temperature reference we will call Tsub (≈ 300 K) We can view a series resistor from the battery as its own internal resistance and will call this Rload [BOLOMETER CIRCUIT]With no radiation present, the initial dc current that flows through the

bolometer will heat up the resistor thus altering the resistance and raising the

xvi

bolometer temperature to Tbol So now, without any infrared radiation we have a temperature difference ∆T1 = Tbol - Tsub Adding IR radiation will further raise bolometer temperature to ∆T3 = ∆T2 - ∆T1 Bias currents for these detectors can be

on the order of 10-50nA with power dissipation in the range of 1-100nW [for what size-a pixel or array…120x160 & 480x460]

Temperature Sensitive Resistive Materials

One desirable property of a good quality detector is to have a high

temperature coefficient of resistance, which is described in detail in the next section Vanadium oxide and amorphous silicon are the current industry standards for the uncooled market Vanadium oxide is the more mature material whereas

amorphous silicon is gaining wide support for very low power applications The

materials are both semiconductor resistive materials The differences between them lie in the growth process and final resistance values Vanadium oxide

bolometers tend to have lower resistances and higher power dissipation

Amorphous silicon bolometers have much larger resistances and have very low power dissipation 1/f noise seems to be at higher concentrations in amorphous silicon bolometers The non-uniformities responsible for fixed pattern noise are lower in amorphous silicon This thesis addresses the associative issues of

reducing spatial and temporal noise by employing background suppression techniques CMOS readouts in this thesis were designed based upon the

amorphous silicon detector

xvii

Vanadium Oxide (VO 2 )

In 1982 R Andrew Wood and his team at Honeywell Technology Center

originally developed the micro-bolometer composed of sputtered thin films of oxides onto a silicon nitride bridge structure Chemical vapor deposition is the preferred processing mechanism This material has a high room temperature coefficient of resistance (TCR) around 2-2.2% and is a more mature technology as

well as the industry choice.[Kruse book]

Amorphous Silicon (α -Si)

[Kent cit]One advantage this detector has is the compatibility with

conventional silicon processes and CMOS foundries while offering low-cost compatible solutions This thermally sensitive resistor has a room temperature coefficient of resistance of 2.5-6 %/K Material resistances are several orders of magnitude larger (order of 106 Ω) and thus find applications in continuous bias designs aimed at low power designs The processing mechanism used is RF sputtering Problems with this detector are the 1/f noise

xviiiFigure 2.1: 25u x 25u micro-bolometer array photomicrograph

from Dr Paul Norton at NVESD

Uncooled Systems: Figures of Merit

The following are commonly used benchmarks to compare the uncooled

camera systems on the market today These definitions were taken from [Kruse 9]

7-xix

Absorbing Resistive Element (VOx or α-Si)

Figure 2.2: Single uncooled pixel element from Dr

Paul Norton at NVESD

Via Contact to Voltage supply

Thermal Isolation Support Legs

Via Contact to Readout Input

Defined as the signal output from a single pixel of an array divided by the

incident radiation falling onto that pixel:

2

int

14

wavelength dependence due to the spectral response of the absorbing layers[].

xx

NE∆ T & TCR

The first is noise equivalent temperature difference and can be expressed for a single pixel or the average of the pixels in the array NE∆T has two formal definitions: 1) “The NETD is the change in temperature of a blackbody of infinite lateral extent which, when viewed by a thermal imaging system, causes a change

in signal-to-noise ratio of unity in the electrical output of the pixels of a focal plane array, or else of the readout electronics which receives an input signal from the pixels of the array” and 2) “The NETD is the difference in temperature

between two side-by-side blackbodies of large lateral extent which, when viewed

by a thermal imaging system, gives rise to a difference in signal-to-noise ratio of unity in electrical outputs of the two halves of the array viewing the two

blackbodies.” It is the measurement of the smallest temperature difference of target that produces a gain of unity at the pixel output

τo is the optical transmission, AD is the pixel area and 1

2sin

F

θ

= where θ is the

angle which the marginal ray from the system optics makes with the optical axis

at the focal point of the image P

T λ

∆

∆ is the change in power per unit area radiated

by a blackbody at temperature T, with respect to T in a particular waveband

The TCR is defined as the temperature coefficient of resistance

xxi

( )

det det

[Kent] Figure 2.3 below was taken from a course on detectors given by Paul

Norton from NVESD The graph shows various bolometer material resistance vs temperature

Thermal Time Constant

The bolometer has a response time or relaxation time defined by the following:

( )

th thermal

eff

C

s G

τ = (2.7)

xxiiFigure 2.3: The semiconductor has negative TCR inferred from the

graph

Cth is the total heat capacity of the bolometer The detector can be considered a composite of several layers each contributing some thermal capacitance Geff is thesum of the thermal conductances containing all the heat loss mechanisms Units

of C are J

K and the units of thermal conductance G are

W

K The reader may

consult the following sources for a more thourough explanation; [Kruse and Skatrud/Kruse].

Chapter 2: Fundamentals of Infrared Imaging

Introduction

Maximum performance in today’s and future uncooled infrared staring focal plane arrays are limited by available well capacity storage Terrestrial IR imaging applications contain high levels of background radiation and impose highdynamic range requirements on the readout electronics Achieving maximum signal dynamic range at the pixel level is paramount However, these imaging arrays are limited on what the semiconductor foundry has to offer in terms of active or passive storage devices and process parameters This thesis focuses on a

specific infrared band called the long wavelength infrared region or LWIR This

region spans from 8 µm – 14 µm and contains a very high flux concentration throughout the day’s diurnal cycle; heating during the day and cooling at night Therefore to accommodate such high flux applications, appropriate electrical readouts must be designed The direct injection readout design was chosen for this

xxiii

thesis based on the excellent injection efficiency, minimal area, and low power dissipation in the LWIR The direct injection readout utilizes a common gate p-channel current buffer to isolate any variations from the detector to the integrationnode.

Because of the lack of increased charge storage capacity in modern day sub-micron CMOS geometries, saturation of the integrating well occurs

frequently This precludes attaining long integration times in the readout

Currently, integration times are a few hundred micro-seconds to a few seconds at best with today’s pulse-biased readouts and some DC bias readouts This thesis explores innovative low-power uncooled DC bias readout designs focused on increasing the integration times up to and exceeding uncooled

milli-bolometer time constants Pixel time constants are governed by this simple

mathematical relationship:

int max int

det

C V I

integration times The 1/f or pink noise is proportional to integration time At 1Hz

or 10Hz, the noise is indistinguishable from a low signal Therefore it seems

xxiv

increasing integration times may not increase the signal-to-noise ratio However,recent and future advancements are reducing the 1/f noise to acceptable levels to

achieve johnson noise limited performance The benchmark for sensitivity used

by state-of-the art military and commercial markets is called Noise Equivalent

Temperature Difference or NE∆T, which are on the order of 20-100 mK changes

in a 300 K background Two methods reported in this thesis of helping to achieve

these sensitivities require background suppression schemes called current

skimming and charge subtraction Current skimming subtracts the input DC

current signal to some reference point; usually the substrate This allows a muchsmaller signal representative of the tiny infrared temperature scene change toaccumulate onto the integrating capacitor Charge subtraction, utilized by a MOScapacitor, is another method of extending the integration times by siphoningcharge from the integration node The active storage element is the gate oxide of aMOSFET where the source and drain are electrically connected Ironically, theMOS capacitor has the highest capacitance per unit area but has an inherent non-

linear nature The non-linearity is not a problem as long as the capacitor is operated in its linear regime The nice feature of the MOS capacitor is its

dynamic operation Effective storage capacity may be increased or decreased bydriving the MOS capacitor between depletion and inversion The goal of this work

is to design, fabricate, and test pixels that achieve successful backgroundsuppression techniques Suppression mechanisms investigated are:

• Single MOS transistor

• MOS cascode transistor

xxv

• Self-biased cascode transistor

• MOS capacitor

Optimized layout and design in the above suppression schemes arediscussed The study includes discussion of the practicality of per-pixel skimmingand the related biasing mechanisms To date, there is no viable per-pixelskimming CMOS readout in the military or commercial uncooled markets.Chapter 4 explains best the present situation on the many design challenges andhurdles for per-pixel skimming in uncooled focal plane arrays

Wavelength Spectrum & Atmospherics

Figure 1.1 shows the atmospheric transmission of the infrared spectrum and respective bands Absorption occurs when molecules such as water vapor, ozone, carbon dioxide, carbon monoxide, and nitrous oxide collide and exchange energy with the incident photon thus causing a temperature rise in the molecule

[driggers] The band gap energy of the photon must be greater than or equal to

the molecular band gap in order for absorption to occur Band gap energy is defined from Einstein’s energy relation, which is proportional to Planck’s constantmultiplied by the frequency of the wave

Scattering is when large particles such as pollution, dust, smoke, precipitation, and various man-made or natural aerosols collide with incoming radiation are

xxvi

redirected with a different energy The extent to which the infrared radiation is

redirected has to do with the mass of the molecule and wavelength of the

incoming photon [cit] Of course since energy is conserved in these collisions

whether it is absorption or scattering, these mechanisms are important functions

of wavelength Various factors influence the atmospheric transmission of the

incoming infrared radiation These include ambient temperature, altitude,

humidity, atmospheric pressure, and refractive index fluctuations (turbulence)

[Driggers, Cox, 127] As seen by the figure the long wave infrared has a high

transmission factor and is also the largest band (but not evident from a log scale inwavelength)

xxvii

Figure 1.1: Reprinted from Introduction to Infrared and Electro-Optical Systems

The spectrum breakdown shows the different infrared bands

extending from 8 m to >30 m

Readout Systems

Imaging systems employ a readout integrated circuit that translates the infrared detector output to an electrical signal that represents the observed scene

A primary function of the readout is to provide pre-amplification (detector signal

conversion), gain, and some cases A/D conversion Below Figure 1.2 shows an

example of a typical single readout signal path from pre-amplification to video

output

xxviii

Pixel Column Multiplexer Video Amplifier & A/D and D/A

A/D D/A Video Amplifier Output Pad

Vdet Readout Process Electronics

Figure 1.2: ROIC & image system level description

The function of the pixel (sometimes referred to as the unit cell) is a

current-to-voltage conversion or voltage-to-current conversion Two methods of pixel readout are voltage and charge mode output The first method usually

employs a current-to-voltage conversion on the integration capacitor and

subsequently buffers the voltage signal from the integration node to the output

through a source follower Doping concentration of the source & drain diffusions

and the substrate determine source follower gain levels for a given CMOS

foundry To achieve maximum dynamic range, designers should utilize native

transistors that yield approximately 98% gain for n-channel transistors (due to lowthreshold voltage) If native transistors are unavailable, the next best option is

using p-channel transistors, which have signal gains of > 90% The

current-to-voltage conversion is performed by the direct injection transistor (common gate

current amplifier) and capacitor combination The direct injection transistor has a

current gain of ≈ unity

xxix

0

Row_Sel

Vdibias Vdd

Voltage ModeReadout withCurrent Skimming

Cint Rdet

Col_Amp

Column Bias

Figure 1.3: Voltage mode readout For lower noise, designs should employ p-channel

(well transistors) where possible.

One immediate draw back using a voltage buffer is the body effect In essence, the threshold raises and gate overdrive reduces which translates to lower source follower gain and loss of integration signal Using a p-channel buffer will

eliminate this problem (by connecting the well and source) as well as being a better candidate for lower noise Another method of readout is to operate in current or charge mode Some reasons for implementing this architecture is to 1) minimize power consumption by eliminating the source follower, 2) lower noise, and 3) decreased pixel area so as to increase the integration capacitor size The above reasons are valid in choosing a charge mode approach over the voltage mode, but the voltage mode achieves higher dynamic range than the charge mode readout In charge mode readout, when the row select switches are open and closed, the integration capacitors are exposed to the column integrators where charge injection could occur (loss of signal) Also, the integration capacitor contains a noise floor and therefore does reset down to ground Usually, the power

xxx

supply swings in the pixels do not cover the full range Therefore when using charge mode readout, designers have a decreased voltage swing due to the integration node sharing its capacity with the column capacitance when being

read Therefore

int_

column node

Q V

Figure 1.4: Charge mode (also called current mode) Output signal is charge

rather than voltage The reset of the integration capacitor is part

of the readout operation The advantage of charge mode is less pixel area, lower power consumption, and lower noise The dis- advantage is the decrease in dynamic range on the column as a result of integration node capacitance shared with column

capacitance

0

Column Amplfier

Pad

0

0

Video Buffer

We have discussed the system architecture and will now explain the two

modes of the readout process; rolling and snapshot Rolling mode is a sequential

read where the system software clocks the first row to be read then moves onto the next row So as we cycle to a new row for readout, the previous row has begun

to integrate again This method reads one row at a time, and integration is

staggered over the frame for individual rows Snapshot mode is where each row integrates in parallel and then is readout after all the pixels have integrated

simultaneously Snapshot mode requires a hold capacitor per unit cell and more

sophisticated clocking electronics [Calist.MSEE].

Chapter 3: Background Suppression in the LWIR

Justification for Current Skimming & Charge Subtraction

Background suppression is implemented to improve signal-to-noise by extending integration times and hence dynamic range The difficulty of

background suppression is extracting the low contrast IR signal from a high background photon concentration without introducing significant additional noise

xxxiii

to the pixel One solution for background suppression involves current mode

skimming implemented by MOS transistor current sources Another form of suppression is charge subtraction, which is implemented by connecting a MOS capacitor’s source and drain to the integration node while biasing the gate Charge subtraction augments the pixel signal sensitivity by maintaining high dynamic range from a smaller effective capacitance

Due to response non-uniformity of the detectors, the background pedestal varies from pixel to pixel and requires (off focal plane) non-uniformity correction methods to keep track of pixel bias levels to interpret the appropriate injection levels Usually off-chip DSP circuitry is implemented for non-uniformity

correction Associative issues of current skimming are investigated (Chapter 4) to better explain the advantages and disadvantages for improving detector non-uniformity Current skimming should improve pixel performance by reducing the

overall temporal (white or broadband) noise by longer integration times leading

to higher signal-to-noise, while charge subtraction enhances signal sensitivity and achieves higher signal-to-noise as well This chapter explores the details of background suppression methods for continuous bias uncooled infrared focal plane arrays, which aim to improve pixel signal-to noise

Spatial & Temporal Noise

Without careful design, adding background suppression circuitry to thepixel runs the risk of not improving the performance The reality of per-pixelskimming (discussed in CH 4) raises questions as to whether this will be effective

in reducing the fixed pattern noise from the detector array Per-pixel skimming is

xxxiv

possible with smart analog functions, adaptive skimming pixels, and accurate

subthreshold design and modeling, [Philippe’s explanation].

A limiting factor in current uncooled focal plane arrays is the detector

resistance variations These micro-bridge structures are not fabricated with equal

resistances Inherent in the growing phase of these sensors are process variationsthat inevitably may cause a 2-3% variation in detector resistance across the staring

array [Kent McCormack] The ideal situation is to have all the pixels in the

staring array to be subtracting the same dc background pedestal Capturing thefiner details of the re-created scene image requires a great deal of sensitivity in thedetector and CMOS readout electronics For resistive bolometers to operate with20-100 mK sensitivities within a 300 K background, uniformity is paramount.Therefore it is imperative that detectors maintain similar performance.Neighboring pixels as a result of different resistance values absorb differentphoton flux intensities, which translates to a variable spectral response across thearray contributing to fixed pattern noise Fixed pattern noise is considered aspatial noise component, which includes the detector, pixel, column, row, andoutput CMOS readout electronics noise [4] The following are some physicalmechanisms that result in responsivity non-uniformity

Detector bias non-uniformities due to transistor matching impact the

image quality by adding to the spatial noise component For example, if the detector bias in pixel11 is 1 V and pixel12 bias is 980 mV, the injection currents willdiffer The results are different magnitudes of charge storage in the well

Threshold voltage non-uniformity causes certain spatial fixed pattern noise within

xxxv

the pixel elements Threshold variations for a given modern CMOS process on a given 8″ wafer typically are less than ±5 mV Threshold variations may vary from process to process and wafer to wafer From a camera system point of view, integrating as much of the low-contrast IR signal translates to increased

sensitivity and image quality, therefore a significant array variation in well integration negatively affects dynamic range There are non-uniformity correctiontechniques, which alleviate fixed pattern noise in the detector array Correction voltages in the form of a digital code are clocked at the output pixel rate and introduced back into the pixel to adjust the non-uniform injection currents and attempt to make them equal [4]

Some pixels in the array may need to skim more or less than others as a result of the following:

• Detector resistance non-uniformity

• Detector bias variations

• Pixel location on the array

• Transistor threshold voltage non-uniformity

Since the skimmers will operate in subthreshold (high gain and low power), current mismatch may be as great as 10% between neighboring pixels in

the worst case [Vittoz stat] Maintaining individual skimming current level

relative to the level of injection current is critical for successful operation

Design for continuous bias arrays with extended integration times makes the design more flexible for reducing the temporal noise as opposed to the spatial

xxxvi

noise requirement because of the inverse relationship of noise bandwidth to integration time Noise bandwidth for the uncooled integrator pixel is defined as:

( )

int

12

BW

τ

= (3.1)

This thesis describes pixel-level concepts that achieved integration times

on the order of bolometer time constants (thermal time constant in equation 2.7)

Sensitivity & Dynamic Range

These two concepts are very important referring to the ability of thedetector and processing electronics to attain high signal-to-noise Chapter 2covered sensitivity figures of merit measured by ℜ, NEP, D*, NE∆T, and thermalresponse time Envision the signal path from the scene to the column amplifierelectronics As an infrared photon is collected by the detector, noise accumulates

on (summed in quadrature) the detector, pixel, and column electronics A goal forthe readout designs is to have the rms noise levels at the output of the column

amplifiers to be less than the noise contributed from the detector (detector

limited) Sensitivity cannot fully be explained by one or two parameters, but

rather a combination of system parameters that result in a particular cameraperformance level However, a good figure of merit that states a direct sensitivitymeasurement referring to the pixel is NE∆T, which relates signal to temporalnoise NE∆T is defined as the smallest temperature difference from a target thatproduces a signal-to noise ratio of unity at the pixel output Detecting minutetemperature changes may be expressed as NE∆T for a single pixel or the average

of pixels in the array [5]

xxxvii

Dynamic range can have several meanings depending on the area of engineering Essentially it is the ratio of some maximum parameter (power, voltage, or current) to some minimal detected level of that parameter In terms of uncooled thermal systems and for the remainder of this thesis the dynamic range of an uncooled bolometer pixel or array will be:

of the pixel dynamic range the well capacity can be envisioned as a storage bucketwhereby charge is accumulated as a voltage signal The goal is to integrate as much charge as possible without saturating the integration capacitor In the case of

a voltage mode style readout operation the well signal is buffered out by a source follower, which typically has 80-97% gains depending on the process and type of FET used In a sub-micron process such as a 0.25 um mixed-signal, a native MOSFET provided by the foundry, which enables a lower Vt., have 97% gain Vendors which do not supply native transistors typically provide devices that will give gains ~ 80-95% In terms of dynamic range and signal-to-noise, a p-channel transistor would be a better choice because the body effect can be eliminated, has

a higher gain, and lower noise (as a result of lower mobility) An n-channel transistor suffers from the body effect and usually drops the signal by an amount equal to its threshold Process parameters from a given foundry directly impact the performance of pixel dynamic range and sensitivity

xxxviii

DC Suppression Concepts

This section offers details of the background suppression schemes and highlightsthe preferred methods An important aspect of implementing a current source inMOS technology is to maintain precision matching in layout, which reduceschannel conductance fluctuations Maintaining a stable constant current source

requires non-minimum channel lengths to increase the Early voltage 1 [6] which

essentially flattens the I-V curves Analog design usually requires minimumlengths to be roughly 2-5 times the size of Lmin for a given process technology.However designs may have area constraints If the design is not constrained byarea requirements making the channel longer than required increases the outputresistance further

The first concept is the single MOS transistor skimmer Figure 3.1 below

illustrates a single n-channel transistor current skimmer connected in a commongate configuration Since these skimmers are operating in subthreshold, thechannel currents are exponentially related to the gate voltage and are quitesensitive to gate bias Fluctuations in drain-to-source voltage will cause theskimmer to have an incremental output current where the slope in the I-V curve isthe output resistance Figures 3.2 and 3.3 highlight the effects of the outputresistance as channel length is increased A major drawback of this configuration

is that all signal changes on the integration node will be reflected through theskimmer’s channel therefore reducing signal sensitivity by having a non-linearcurrent

base region This mimics the effect of channel length modulation in above and below threshold

xxxix

Simulated single MOS transistor current skimmer

Subthreshold Operation Vgs = 700mV

Figure 3.2: As the channel lengths increase, the I-V curves have lower area

conductance; hence the increased area decreases the percentage of

the channel affected by channel length modulation The area most affected will be the area closest to the drain where the electric field is

Cint Vskim

Figure 3.1: Description of a direct injection pixel with a single MOSFET current skimmer This skimming suffers from channel length modulation at

the integration node and therefore has a non-linear current Also, any time varying capacitances will adversely affect current skimming No matter how long the channel is, it is exposed to the integration node

Transistor width would have to be kept at a minimum and the lengthincreased to where the I-V curves are flat Also, the single transistor must remain

in subthreshold The skimming transistor’s intended region of operation is utilized

to achieve very low power dissipation and the highest transistor gain Since thedrain current is exponentially related to gate voltage, controlling the gate bias isthe most difficult task in the per-pixel skimming concept

Leff (um)

Conductance(gds)pA/V

Outputresistance(GΩ)