`
`P.B. DENYER, D. RENSHAW, W. GUOYU and L. MINGYING,
`
`“CMOS image sensors for multimedia applications,”
`
`Proceedings of the IEEE 1993 CICC, pp. 11.5.1-11.5.4 (1993)
`
`TRW Automotive U.S. LLC: EXHIBIT 1051
`PETITION FOR INTER PARTES REVIEW
`OF U.S. PATENT NUMBER 8,599,001
`IPR2015-00436
`
`
`
`CMOS IMAGE SENSORS FOR MULTIMEDIA APPLICATIONS
`
`P. B. Denyer, D. Renshaw, Wang Guoyu, Lu Mingying
`
`VLSI Vision Ltd. & University of Edinburgh
`The King's Buildings,
`Edinburgh, UK
`
`sensor is improved by maximising the fraction of each pixel
`occupied by the exposed diode. At the same time we have an
`interest in keeping the pixel pitch as small as possible to
`maximise the sensor resolution achievable within a given
`cost (silicon area). Typically we achieve a fill factor better
`than 50% at a pixel pitch of around 10 microns using a 1
`micron double—metal CMOS technology.
`
`
`
`horizontal shift regi st
`
`0/pstage
`
`
`
`§
`
` vertical
`
`
`
`
`
`shiftregister
`
`Figure 1. CMOS Sensor Architecture
`
`low-noise
`The major circuit problem is to achieve rapid,
`linear detection and output of the charge detected in each
`pixel. For typical sensor responsivity and operation down to
`gloomy lighting conditions
`at normal video rates
`the
`minimum detectable charge may need to be less than 1 fC. We
`have found that a useful performance may be obtained by the
`architecture shown in Fig 1. We provide charge—integrating
`amplifiers at
`the top of each column of pixels.
`Reading
`whole rows of pixels in parallel removes much of the speed
`constraint on these arnplfiers and allows them to be designed
`with regard to sensitivity and size (the need to realize this
`function within the pixel pitch is a major layout challenge).
`For this reason we prefer
`to use a simple single-ended
`amplifier,
`and for extreme
`sensitivity
`the
`integrating
`capacitor may be purely parasitic (around 10 fF). Once a row
`has been sensed, it is stored on an array of capacitors before a
`second, fast charge sensing operation is used to scan the row
`of values out through a single output amplifier. As there is
`
`ABSTRACT
`
`Building on earlier work We confirm the feasibility of
`integrating image sensors with other system functions on
`single chips, fabricated on standard CMOS ASIC processes.
`We report novel circuit techniques and present results from
`successful implementations.
`
`1.
`
`INTRODUCTION
`
`We have previously reported [1] that quality images may be
`obtained by carefully designed sensors
`implemented in
`standard ASIC CMOS processes.
`This leads to several
`powerful advantages;
`
`through a very high level of
`greatly reduced size
`-
`integration. This may extend to encompass other elements of
`the imaging system on the same chip as the sensor (e.g. A/D
`conversion, compression logic).
`
`— greatly reduced power consumption, and operation from a
`single 5v rail.
`
`than comparable systems constructed with
`lower cost
`-
`discrete camera technologies. System cost reductions of 90%
`are often feasible.
`
`These advantages are relevant to many applications in the
`vision marketplace, but nowhere so much as in personal
`computing
`and
`telecommunications, where multi-media
`applications are growing but the advantages conferred by the
`capability of sight are yet barely recognised.
`Integrated
`CMOS-based sensors and sighted peripherals are made
`feasible at economic cost by the technology reviewed and
`presented in this paper.
`
`2. FUNDAMENTALS
`
`Figure 1 shows an architecture for an array image sensor
`which is realisable in a conventional ASIC CMOS process.
`The entire function may be likened to a single-transistor
`DRAM module, except data is written optically,
`the cell
`(pixel) array is constructed in a regular
`two-dimensional
`format, and the sensing operation is analogue.
`
`contains a single access transistor, as in the
`The pixel
`DRAM analogy, but the source region is physically extended
`to form a reversebiased photodiode. The efficiency of the
`
`IEEE 1993 CUSTOM INTEGRATED CIRCUITS CONFERENCE
`
`0-7803-0826-3/93 $3.00 © 1993 IEEE
`1051-001
`
`11.5.1
`
`1051-001
`
`
`
`only one of these, a more complex design is permitted to
`achieve the necessary speed (up to 14 MHz for full resolution
`BIA video).
`
`3. OUTPUT STAGE
`
`It is most useful to augment the output stage with some form
`of digital gain control.
`This assists
`in implementing
`automatic gain control (AGC), and may be adapted for A/D
`conversion.
`Ideally, this stage should be free of any analogue
`offset (or else the applied gain value will also cause variation
`of the offset).
`Fig. 2 shows a simple self-calibrating
`digitally-controlled gain stage.
`
`
`
`Figure 2. Self—calibratt'ng output stage including digiial gain
`control.
`
`The video signal is first converted to a current through M1.
`M2 provides a constant current source Imax, of value equal to
`the greatest current expected from MI. For an arbitrary value
`of Vin, which converts to Iin through M1, the residual current
`Iv = (Imax - Iin) flows out of this stage, through the common
`gate device M4 and to ground through load devices M5, M6,
`etc. These load devices operate in their linear regions and
`form a binary-ratioed series. The total conductance of the
`load is G.D, where G is the conductance of the smallest device
`in the binary series (say M5) and D is the digital word formed
`by the control bits applied to the series of gates. 80 the
`output becomes;
`
`Vout = Iv/(G.D)
`
`Now, if the input to this stage is inverted video (Vin = Vblack
`- Vv) and we approximate M1 as a linear transconductor of
`value Gt and Imax = Gt.Vblack, then;
`
`Vout = VV/G'.D.
`
`G‘ = G/Gt is a scaling constant and D is the digital input word.
`This stage is therefore a Dividing DAC (DDAC).
`This is
`useful enough to control gain incrementally within a feedback
`loop that uses comparator compl
`to monitor
`the
`image
`quality (judged by the fraction pixels with values above Va).
`
`Although this stage is far from perfectly linear, it is suitable
`for most video applications.
`Indeed where the resulting video
`is to be displayed directly on a monitor some nonlinearity is
`called for to provide gamma correction.
`
`An accurate calibration of Imax is obtained by turning M3 on
`and M4 off during a known period of video corresponding to
`black, for example during the readout of a reference line at the
`bottom of the array which is deliberately shielded from light.
`The resulting value is held via C1 for one video field until
`the
`calibration process can be repeated.
`
`4. OPTICAL RESPONSE
`
`The overall device efficiency is a function of the optical path,
`the physical structure of the photodiode, and the gain of the
`detection path between the photodiode and the output.
`Optoelectronic conversion in the photodiode plays a key role
`in determining the response of
`the sensor.
`Photons are
`converted to charge by splitting electron—hole pairs; where
`this occurs within the depletion region beneath the diode the
`resulting free carriers form a photocurrent under the influence
`of
`the
`depletion region
`field.
`Unfortunately
`some
`photocharge is lost by recombination if
`the conversion
`occurs
`in the heavily
`doped
`surface diffusion
`layer.
`Additionally, charge created below the depletion region may
`still be collected by diffusion as the recombination time
`Within the lightly doped substrate is relatively long. The
`depth of
`conversion is
`a
`statistical
`function of
`the
`penetrating wavelength, and therefore the spectral
`response
`is not uniform.
`Fig. 3 shows typical photodiode responses
`from two commercial CMOS processes. The response to blue
`light (shorter wavelengths) is attenuated by the surface effect,
`and the response to red light by the bulk effect. There is good
`coverage of the visible spectrum in both cases and significant
`response at near-infra-red wavelengths above the visible
`spectrum. The attenuated blue response compromises colour
`camera
`realisation,
`but
`the
`technology
`trend
`towards
`The near-IR
`shallower diffusion mitigates this problem.
`enable
`covert
`camera
`response
`is
`good
`enough
`to
`applications using IR illumination.
`
`
`
`responses of two CMOS
`Predicted spectral
`Figure 3.
`processes. (arbitrary vertical scale).
`(a) 2 micron single tub epitaxial
`(b) 1 micron twin tub nonepitaxial
`
`11.5.2
`
`1051-002
`
`1051-002
`
`
`
`S. SECOND ORDER EFFECTS
`
`image
`acceptable
`a visually
`for
`rule-of-thumb,
`a
`As
`imperfections must be contained to better than 1% . A sensor
`of
`the
`above architecture, built
`in a
`standard CMOS
`technology, may easily suffer fixed-pattem noise effects
`worse than this criterion.
`Specific problems induced by
`threshold and transconductance mismatches are spatially
`random pixel offsets, caused by threshold variation in the
`pixel access transistors, giving a measels-like effect, and
`random column offsets, giving vertical stripes, caused by
`threshold offsets in the column sense amplifiers. Both of
`these effects may be cured or compensated however.
`The
`pixel effect is easily cured by ensuring that the common reset
`potential
`is
`lower
`than the gate—voltage-minus-threshold
`limited maximum. This removes the primary Vt-dependence
`and hence the source of noise. As for the column sense
`
`several common compensation
`amplifier offsets, any of
`Thus
`the desired analogue
`schemes may be adopted.
`performance may be achieved by careful circuit design whilst
`retaining the primary advantage of working within a standard
`CMOS ASIC process. Figure4shows examples of the effects
`of these sources, and their elimination, taken from the fully-
`integrated CMOS camera chip shown in Figure 5.
`
`
`
`(£1)
`
`(19)
`
`(C)
`
`Figure 4. Typical image artefacts and their elimination.
`(a) measels effect of Vt variation in pixel reset potential
`(b) vertical striping caused by offsets
`in column sense
`amplifiers
`(c) defect-free image after compensation
`
`6. EXAMPLE CAMERA
`
`The device of Fig. 5 includes a 312 x 287 pixel array,
`formatted to accord with the 4:3 aspect ratio which is a TV
`standard, timing logic to achieve the CCIR video standard,
`exposure control logic covering a range of 40,000 :
`1, and
`AGC for up to 20 dB of gain in poor lighting conditions. The
`chip integrates all of the camera function except the lens, a
`Sv regulator, a clock crystal and a few decoupling capacitors.
`All of these may be provided within a space of approximately
`1 cubic inch, and the resulting camera module consumes less
`than 40 mA whilst continuously driving a 50 ohm video load
`(e.g. a TV monitor).
`
`
`
`Figure 5. An example CCIR camera fully integrated using
`ASIC CMOS technology
`
`The resulting camera provides subjectively excellent video at
`a fraction of
`the cost,
`size and power consumption of
`contemporary technologies, exemplifying the
`advantages
`cited in the introduction.
`
`7. DIGITAL APPLICATIONS
`
`require video
`Many multimedia and related applications
`signals in digital form.
`It is clearly attractive to provide A/D
`conversion on—chip, and again the use of CMOS technology
`makes this feasible. One approach, using the same output
`stage as in Fig. 2, is to implement a sucessive—approximation
`technique.
`For this purpose we apply a suitable refrence
`voltage Va,
`to comparator compl.
`The output of
`the
`dividing-DAC is then successively compared to this value
`whilst a binary search proceeds for
`the closest digital
`representation. At
`the conclusion of this process, Vout
`is
`close to Va and so;
`
`D = IV/(Va.G)
`
`where D is the digital word applied to‘the DAC. Thus the
`output stage can be used to form a linear digital conversion of
`the video current.
`The logic overhead to implement
`the
`sucessive approximation process is about 400 gates only,
`and it is ideal for applications requiring digital pixel rates up
`to 1 MHz. This encompasses the CIF image standard for
`video compression at frame rates up to approximately 12
`frames per second.
`For
`faster digital
`rates, other A/D
`techniques may be equally well applied in CMOS on the same
`chip as the sensor.
`
`8. MINIATURE OPTICS
`
`The camera function is always completed by the addition of a
`lens, and this threatens to reduce the cost and size advantages
`given above if conventional
`lenses are used. We have
`experimented with two miniature lens forms. Fig. 6 shows in
`
`11.5.?)
`
`1051-003
`
`1051-003
`
`
`
`cross section a miniature two-part glass lens which is bonded
`to the sensor surface.
`
`9. EXAMPLE APPLICATIONS
`
`We conclude by examining two multimedia case studies; a PC
`stills
`camera
`peripheral,
`and
`a
`fu1ly—integrated
`sensor/compressor Videophone device. These both illustrate
`the potential for all-CMOS video systems.
`In addition to
`integrating the camera function, most of the remainder of the
`system may in principle be included onchip. The only off
`chip components need be those that are not physically
`realisable in an ASIC CMOS process (e.g. clock crystal) or
`quantities of RAM which are more economically provided by
`commodity parts.
`
`System
`
`Potential Onchip fns.
`
`Ofichip fns.
`
`PC camera
`peripheral
`
`Videophone
`
`5V Regulator
`Crystal
`RAM
`
`Decoupling
`RS232 Driver
`
`5v Regulator
`Crystal
`(RAM)
`Decoupling
`Comm. Driver
`Disp. Driver
`Display
`
`Sensor
`Exposure Control
`A/D
`
`RAM Control
`RS232 Format
`
`Sensor
`Exposure Control
`A/D
`Block Format
`Compress/Decompress
`Display Format
`Comm. Format
`
`10. CONCLUSION
`
`A range of multimedia applications which require adequate
`performance at minimum power,
`size and cost may be
`satisfied by combining the sensor function on-chip with the
`remaining system functions, using standard CMOS ASIC
`technology. The required perfonnance may be achieved by
`careful analogue design, obviating the expense and bulk of
`solutions using discrete camera technologies.
`
`ACKNOWLEDGEMENTS
`
`We gratefully acknowledge the support of the UK Science and
`Engineering Research Council who supported this work under
`Grant GR/F 36538.
`
`REFERENCES
`
`[1] Renshavv, D., Denyer P.B., Wang G.,and Lu M., "ASIC
`Vision", CICC 90
`
`package substrat-
`
`silicon die
`
`high index bloc
`
`I
`
`1mm U
`
`
`
`low index hemisphen
`V
`
`
`g\\V
`
`
`
`Figure 6. Cross section of a miniature lens assembly directly
`bonded to the silicon surface.
`
`Its dimensions and mass are comparable with the sensor die
`itself;
`the example shown comprises a glass block with a
`high refractive index aproximately 1600 microns thick,
`to
`which is cemented a hemisphere of lower index and radius
`approximately 650 microns.
`Its
`low mass
`and solid
`construction make this optical
`system extremely robust
`against physical shock. Lenses of this type are also capable
`of awide field of view. ‘Fig. 7 shows an image resulting from
`the use of this lens to provide a 90 degree field of View over a
`CMOS array of 100 x 156 pixels.
`
`
`
`Figure 7. Image obtained with chip-mounted lens and on-chip
`digitisation
`
`A somewhat more conventional approach is to use a single
`element aspheric lens moulded in plastic (typically acrylic).
`This is also small (although larger than the lens of Fig. 6) and
`of low mass, but is only good for fields-of—view up to around
`65 degrees.
`Either
`lens
`technique
`can yield camera
`implementations which are remarkably small and economical
`compared with existing camera technologies.
`
`11Ji4
`
`1051-004
`
`1051-004