How we wanted to revolutionize X-ray radiography, and how we then "accidentally" discovered single-photon CMOS imaging Stanford University EE Computer Systems Colloquium February 23 rd, 2011 EE380 Peter Seitz, Ph.D. Vice President Nanomedicine, CSEM SA Adjunct Professor, EPFL Institute of Microengineering
X-Ray Imaging Information Content: Photography and X-Ray Imagery First (black-and-white) photograph: 1826 Early X-ray image: 1896 Photograph today X-ray image today Stanford EE380 Peter Seitz Page 1
X-Ray Imaging Absorption Properties of X-Rays in Matter I 0 α I I I 0 e x x I 0 α 1 α 2 α 3 I n i1 I I e 0 i x i x 1 x 2 x 3 α : linear attenuation coefficient (1/cm) μ : mass attenuation coefficient (cm 2 /g) ρ : density (g/cm 3 ) tot n n w i i i1 i1 i Stanford EE380 Peter Seitz Page 2
X-Ray Imaging Photon Energy Dependent X-Ray Absorption Spectra Medical X-rays Source: NIST XCOM Database Stanford EE380 Peter Seitz Page 3
X-Ray Imaging Functional Dependence of the Photoelectric Effect Textbooks: Bragg-Pierce Law for the photoelectric absorption of a homogeneous piece of elemental matter as a function of X-ray photon energy E and atomic number Z : ( Z, E) Z b E a b 4.0 ; a -3.0 Stanford EE380 Peter Seitz Page 4
End of slide show, click to exit. X-Ray Imaging Absorption Properties of X-Rays Stanford EE380 Peter Seitz Page 5
X-Ray Imaging Photoelectric Absorption Revisited Assume cross dependence b(e) and a(z) : ( E) c Z Z b( E) E a( Z ) NIST Database : Tables of X-Ray Mass Attenuation Coefficients and Mass- Energy Absorption Coefficients http://www.physics.nist.gov/physrefdata/xraymasscoef/cover.html log Z (E) Absorption due to photo-effect E Stanford EE380 Peter Seitz Page 6
X-Ray Imaging Photoelectric Absorption : Monotonous Functions! ( E) c Z Z b( E) E a( Z) Stanford EE380 Peter Seitz Page 7
Color X-Ray Imaging! Stanford EE380 Peter Seitz Page 8 Color X-Ray Imaging i = 1, 2, n constituting elements ) ( ) ( 1 ) ( i Z a E b i n i i E Z c E ) ( ) ( 1 j Z i a j E b i n i i j E Z c j = 1, 2, m energy sampling j a i j b i i j E c Z, Simple linear problem: Find the elemental composition vector ρ i, given the (tabulated/measured) matrix elements B i,j and the measured attenuation vector data α j i i j j x x
Color X-Ray Imaging Color X-Ray Imaging In Practice j n c i Z i1 b( E i ) E a( j j Z i ) j = 1, 2, m energy sampling 1. Measure reference absorption spectra for pure elements (basis) or use tabulated data 2. Measure absorption spectrum of unknown sample 3. Solve for linear combination of basis spectra which best fit the measured attenuation results / (m 2 /kg) 18 16 14 12 10 8 6 4 2 Al (550m) Si (390m) Ti (46m) NIST Al NIST Si NIST Ti 0 0 2 4 6 8 10 12 14 16 18 20 energy (kev) Stanford EE380 Peter Seitz Page 9
Color X-Ray Imaging Experimental Verification X-ray spectrometer (Amptek X-123) Microfocus X-ray source (Hamamatsu L1010101) Sample Stanford EE380 Peter Seitz Page 10
Color X-Ray Imaging Element-Sensitive X-Ray Imaging Demonstrated! Element-sensitive X-ray Conventional X-ray E = 11.6 13.1 kev Stanford EE380 Peter Seitz Page 11
Color X-Ray Imaging Color X-Ray Imaging Around the Corner? j, i c Z b i E j a i j Ill-conditioned inversion problem! Example: cond(b) ~ 200 600 for the combination Al and Si with the limited single energy interval of around 11 14 kev Possible way out: Multiple energy intervals for reduced cond(b) Large, expensive, power-hungry, highresolution X-ray spectrometer Wanted: Affordable Megapixel 2D array of <100 100 μm X-ray pixels with ΔE<50 ev Stanford EE380 Peter Seitz Page 12
X-Ray Pixel Fundamental Noise Source: Johnson Noise in Resistor V 4kT R B σ V : noise voltage; k : Boltzmann s constant; T : temperature; R : resistance; B : bandwidth Stanford EE380 Peter Seitz Page 13
X-Ray Pixel Energy-Selective Single X-Ray-Photon Detector Pixel R r C sensing q noise ktc sensing C detector C C load sensing C detector C detector C load V 4kT R B ; B 1 RC Problem: Large X-ray pixels (area of several 1000 μm 2 ) can have capacitances of pf and more Stanford EE380 Peter Seitz Page 14
X-Ray Pixel Lateral Drift-Field Pixels! K. Hoffmann: Surface charge transport with an MOS-transmission line, Solid State Electronics Vol. 20, 177 (1977) Note: Lateral drift-field pixels have recently been adopted by industry (Hamamatsu, Mesa Imaging, Espros Photonics, etc.) Stanford EE380 Peter Seitz Page 15
X-Ray Pixel Fundamental Noise Limit : Recharge Resistor! R r C sensing qnoise ktc sensing C detector C load Note: Johnson (resistor) noise is RC-filtered: Independent of R! Typical value: C sensing = 50 ff, T = 300 K : q noise = 90 electrons Stanford EE380 Peter Seitz Page 16
X-Ray Pixel Energy-Selective Single-Particle (X-Ray) Detection Integration on (small) capacitance on sensor side Continuous reset on sensor side Continuous-time high-pass filtering of reset noise Narrow bandwidth shaping of recharge noise: High R r (GΩ) implementation difficult when connected as feedback resistor R r sense node C s Stanford EE380 Peter Seitz Page 17
X-Ray Pixel Energy-Selective Single-Particle (X-Ray) Detection Parameter Value detected pulse width 150 ns 1.5 µs conversion factor 27 µv/e - recharge time constant 10 µs high-pass time constant 2 µs pixel area 30 x 20 µm fill factor 56 % Hi-pass filter capacitance 200 ff Overall noise (r.m.s.) 13.5 e - Stanford EE380 Peter Seitz Page 18
Low-Noise Sensing Low-Noise Charge Detection : Noise Sources Noise contribution Value Buffer (first transistor) 1.6 e - High-pass filter resistor 3.9 e - Active low-pass filter 1.4 e - Reset resistor (R r ) 12.7 e - Overall noise (r.m.s.) 13.5 e - Reduce noise substantially (to less than 5 electrons) by changing from continuous (asynchronous) reset to switched (synchronous) reset! Input stage resembles a CMOS active pixel (APS). Is it possible to employ the same ideas (bandwidth engineering, in-pixel amp, input capacitance reduction, synchronous reset) to ultra-low-noise CMOS image sensing? Stanford EE380 Peter Seitz Page 19
Low-Noise Sensing The Holy Grail : Single-Electron/Photon Detection! Stanford EE380 Peter Seitz Page 20
Single electron detection CMOS/APS Image Sensing Conventional CMOS pixel reset transfer V R V Reset V V R sense node select reset reset t column line bias Stanford EE380 Peter Seitz Page 21
Single electron detection Noise Sources in CMOS/APS Pixels Reset noise (ktc noise) reset transfer MOS-FET channel noise (input-referred Johnson noise) Solution: Correlated Double Sampling (CDS) select sense node column line bias Q C S 4kTB g m Stanford EE380 Peter Seitz Page 22
Single electron detection State of the Art: MOS-FET Channel Noise Q C S 4kTB g m C S = 10 ff ; T = 300 K ; B = 20 MHz ; α = 1 ; g m = 50 μs σ Q = 5.1 electrons Stanford EE380 Peter Seitz Page 23
Single electron detection The Long Quest for Single-Electron/Photon Detection S G D reset gate p + p + n p-well n-substrate CCD sensing channel output gate summing gate P1 P2 P3 out dump gate dump drain V DD V reset V DD reset V diff select out Stanford EE380 Peter Seitz Page 24
Single electron detection Novel CMOS/APS Pixel With In-Pixel Gain In-pixel amplification for reduced bandwidth and reduced impact of downstream circuit noise very low readout noise Amplifying pixel (common-source connected p-mos amplifier pixel) transfer select_n sense node reset_n CF pixel A v q C sense column line Rl v n, thermal kta v C column Stanford EE380 Peter Seitz Page 25
Single electron detection Gain Pixel : Reset and Amplifying State Stanford EE380 Peter Seitz Page 26
Single electron detection Gain Pixel : Column-Level Bandwidth Engineering Stanford EE380 Peter Seitz Page 27
Single electron detection Single Electron/Photon Detection With CMOS Imagers! Parameter Value pixel pitch 11 µm fill factor 50% transistor count 4 sense node capacitance 5.3 ff voltage gain (linear) 9.9 pixel conversion factor (lin.) 300 µv/e - linear range 4 ke - full well capacity 29 ke - readout noise (60 fps, 300K) 0.86 e - rms dynamic range (t exp = 17ms) 90.4 db 11x11 μm CMOS pixel with 50% fill factor Sample picture of 256x256 imager. Average: 6 photoelectrons per pixel Stanford EE380 Peter Seitz Page 28
Summary 105 years after Röntgen s discovery, it has been shown that color X-ray imagery (element-sensitive radiography) is possible, in principle. Determination of elemental composition from spectral measurements is an ill-posed problem. Only decomposition into a few components seems practical. Relevance of color radiography? Too early to tell! Spectral measurements require integrated (monolithic) energy-resolved single-photon X-ray detectors. State of the art (ΔE = 50-100 ev) can be improved by an order of magnitude (ΔE = 5-15 ev), using bandwidth optimization, in-pixel amplif., input capacitance reduction, synchronous reset The same techniques can be applied to CMOS/APS image sensors, resulting in sub-electron (photo-) charge detection at room temperature and at video rates. Night vision for everybody is around the corner! Stanford EE380 Peter Seitz Page 29
Stanford EE380 Peter Seitz Page 30