CS 250 VLSI Systems CMOS Imagers 2007 3 22 John Lazzaro www cs berkeley edu lazzaro www cs berkeley edu johnw courses cs250 CS 250 Silicon Imagers UC Regents Spring 2007 UCB Silicon imagers are everywhere 2006 1 9 billion units 7 billion USD 60 based on a CMOS process 2008 80 of imagers will be CMOS 90 of imager revenue will be CMOS 2010 3 6 billion units 12 8 billion USD Today s Lecture Markets Technology Source CS 250 Silicon Imagers UC Regents Spring 2007 UCB 0 45 inches Motorola Q Smart Phone Moto predicted 3M shipped Q4 2006 Source www elecdesign com CS 250 Silicon Imagers UC Regents Spring 2007 UCB Camera module cost 7 4 4 of total cost 158 CS 250 Silicon Imagers UC Regents Spring 2007 UCB Front Back CS 250 Silicon Imagers Source Portelligent Inc UC Regents Spring 2007 UCB Camera module slides into PC board cutout CS 250 Silicon Imagers UC Regents Spring 2007 UCB Typical camera module for the Micron MT9M111 Fixed focus No optical zoom 0 27 inches deep 0 37 inch x 0 37 inch square CS 250 Silicon Imagers Source www asia UC Regents Spring 2007 UCB CS 250 Silicon Imagers UC Regents Spring 2007 UCB Micron MT9M111 1 3 MPixel CMOS Imager 1280 x 1024 pixels Each pixel is R G or B So 2 3 of RGB image data is interpolated Photo a close relative MT9M011 CS 250 Silicon Imagers Pixel size 3 6 m x 3 6 m 0 2 in 5 ch es UC Regents Spring 2007 UCB Silicon Photosensitivity The physics of sensing photons CS 250 Silicon Imagers UC Regents Spring 2007 UCB Intrinsic silicon a weak conductor Band gap energy Silicon Eg 1 1 eV Electron s Conduction band Valence band Holes positive carriers Sufficiently energetic photons create electron holeh Eg pairs e n e r g y Conduction band Valence band Si is an indirect material falling e creates heat not CS 250 Silicon Imagers e l e c t r o n e l e c t r o n e n e r g y UC Regents Spring 2007 UCB Band gap bounds the sensitivity Band gap energy Silicon Eg 1 1 eV Electron s Conduction band Valence band Holes positive carriers e l e c t r o n e n e r g y Since Eg hc photons with 1000 nm can excite a valence band electron to the conduction band From infrared through the visible band Infrared 1000 nm CS 250 Silicon Imagers 700 nm 600 nm 500 nm 400 nm UC Regents Spring 2007 UCB This is how photo cells work A mainstay of 100 in 1 experimenter kits h Surface open to light Schematic symbol Semiconducto r A photosensitive resistor The semiconductor CdS is often used as its band gap falls in blue green and thus blocks infrared w o a filter CS 152 L5 Timing UC Regents Fall 2006 UCB For ICs photodiodes a better match Metal shield oxid e Reado ut circuits go h here pregion n p Metal shield oxid e Depleti on region Optically generated electrons near depletion region swept to the right de p l reg etion ion Optically generated holes near depletion region swept to the CS 250 Silicon Imagers h h n region e l e c t r o n e n e r g y UC Regents Spring 2007 UCB Photodiode I V curves I V I Io eV Vo 1 Dark current When no photons are Iph present Vo 25 60 mV Io 1 20 fA CS 250 Silicon Imagers Quadrant for photosensing Quadrant for solar cells UC Regents Spring 2007 UCB Spectral response 0 35 m n well CMOS Data shown is for a standard 0 35 CMOS logic process Quantum efficiency can be improved by modifying the process 42 of photons that fall on the photodiode are converted to electrons quantum efficiency Source A 640 512 CMOS Image Sensor with Ultrawide Dynamic Range Floating Point Pixel Level ADC David X D Yang Abbas El Gamal Boyd Fowler and Hui Tian JSSC Dec 1999 CS 250 Silicon Imagers UC Regents Spring 2007 UCB Process improvements for Q E Claim These slopes represent technology limits 75 Designed for machine vision Extended IR response is a feature CS 250 Silicon Imagers UC Regents Spring 2007 UCB Color Si photodiodes see gray scale CS 250 Silicon Imagers UC Regents Spring 2007 UCB Color filters deposited on pixel array RGB Bayer Why Source Eric Fossum IEEE Micro CSand Micron Data Sheets 250 Silicon Imagers UC Regents Spring 2007 UCB Human cone array imaged through the eye 250 Imagers 25 42 9669 9679 Source Hofer CS et al J Silicon Neuroscience UC Regents Spring 2007 UCB Pixel Scaling CS 250 Silicon Imagers UC Regents Spring 2007 UCB Camera optics limits pixel scaling 5 m 5 m pixel 5 m pixels match the optical resolving power of practical camera optical systems 1997 Fossum 2007 figure may be smaller Shrinking pixels beyond limit does not add resolution Larger die sizes are the path to higher resolution Process scaling helps imager arrays in a more subtle way CS 250 Silicon Imagers UC Regents Spring 2007 UCB Recall Photodiode design Photons that reflect off metal shielding are lost Fill factor h h oxid e Reado ut circuits go Photodiode here area n p oxid e Pixel area As process shrinks readout circuits shrink and diode grows So fill factor increases and fewer photons lost CS 250 Silicon Imagers h UC Regents Spring 2007 UCB Compound eyes of an insect microlenses CS 250 Silicon Imagers UC Regents Spring 2007 UCB Source www bioschool co uk BBC Source http micro magnet fsu edu CS 250 Silicon Imagers UC Regents Spring 2007 UCB Source Fraunhofer CS 250 ISITSilicon Imagers UC Regents Spring 2007 UCB Readout Circuits CS 250 Silicon Imagers UC Regents Spring 2007 UCB Three Transistor Active Pixel Cell Step 1 Fill Cd and sense column current Reset Row Sel Vdd Vdd Vdd Qf Cd Vdd Cd I Qf Edge circuitry samples current I Qf for later use CS 250 Silicon Imagers Parasitic photodiod e capacitanc e Colum n Sense UC Regents Spring 2007 UCB Opening the electronic shutter Step 2 Electronic shutter is open photodiode empties Cd Reset Row Sel Qd t Qf Qd t Gnd Gnd Cd Cd Too much Qd t and we empty bucket before shutter closes Not enough Qd t and we capture temporal CS 250 Silicon Imagers Limits dynamic range and signal to Colum n Sense UC Regents Spring 2007 UCB Close shutter read pixel value Step 3 Sense how empty Cd has become Reset Row Sel Gnd Vdd Qf Qd t Cd Vdd Cd I Qf Qd t Use I Qf from start of the cycle to reduce kTC reset noise CS 250 Silicon Imagers Colum n Sense Correlated double Temporal noise affects Qf value sampling UC Regents Spring 2007 UCB Improving performance of this cell Bright light performance Do several sensing cycles during an image exposure so that Cd never underflows Dim light performance Lose fewer photons improve quantum efficiency fill factor color IR filter losses …