Stanford EE Computer Systems Colloquium

4:15PM, Wednesday, February 23, 2011
HP Auditorium, Gates Computer Science Building B01

How we wanted to revolutionize X-ray radiography -- and how we then accidentally discovered single-photon CMOS imaging

Peter Seitz
About the talk:

Since the discovery of X-rays in 1895, radiography imagery has only been black-and-white. The reason for this can be found in any textbook on X-rays: The linear attenuation coefficient of a pure chemical element factorizes into a universal product of a power of the atomic number Z times a power of the X-ray photon energy E.

Our careful analysis has shown that the simple textbook equation with constant exponents is incorrect, and that there is a small dependency of the exponents of Z and E: In particular, the exponent of the energy term depends monotonously on Z, allowing us to distinguish between different elements in a sample by simple transmission measurement, albeit with good spectral resolution. This possibility to create color radiographs and CT imagery requires large arrays of miniaturized, low-noise X-ray detectors and spectrometers. Thorough noise analysis of conventional spectrometer circuitry shows the way to a novel ultra-low-noise X-ray spectrometer pixel that can be realized with commercially available CMOS technologies. Our test pixel covers an area of 20x30 microns, has a fill factor of 56% and a record noise performance of only 12 electrons r.m.s. at room temperature. Since this noise is dominated by Johnson noise originating in the continuous reset resistor, a switched reset scheme could bring down the noise to about 5 electrons r.m.s.

These insights into the various origins of noise in X-ray pixels can also be used for the design of a novel type of low-noise CMOS image sensor for electronic imaging in the visible and NIR spectral range. By using an in-pixel single-transistor voltage amplifier with a gain of about 10, optimum noise-shaping in the columns of the image sensor and correlated double sampling, sub-electron noise becomes possible.

Employing UMC's commercially available 0.18 micron CMOS technology, we have realized a 256x256 array of 11x11 micron pixels for which we measured a record readout noise of 0.86 electrons r.m.s. at room temperature and at a speed of 60 full frames per second.


Download the slides for this presentation in PDF format.

About the speaker:

[speaker-photo] Peter Seitz received his M.Sc. and Ph.D. degree in physics from the Swiss Federal Institute of Technology ETH in Zurich, Switzerland, in 1980 and 1984 respectively. He then worked for the RCA research laboratory in Zurich and the David Sarnoff Research Center in Princeton, NJ, in the domains of optical metrology and digital image processing. In 1987 he joined the Paul Scherrer Institute in Zurich, where he created and headed the research group for solid-state image sensing. In 1997 he transferred to CSEM, the Swiss Center for Electronics and Microtechnology, in Neuchatel and Zurich, Switzerland, where he became Vice President and head of the Photonics Division in 2000. In 2007 he created CSEM's new Nanomedicine Division in Landquart, Switzerland, which he is heading still today. Since 1998 he has also been professor in optoelectronics at the University of Neuchâtel and the Swiss Federal Institute of Technology EPFL, and he continues research and teaching in this capacity. He is a life member of the Swiss Academy of Technical Sciences SATW, and he is a Fellow of the European Optical Society EOS. He has authored and co-authored 180 scientific publications, and he is holding 35 patents in semiconductor imaging and optical measurement techniques.

Contact information:

Peter Seitz
CSEM SA, Nanomedicine Division
Bahnhofstrasse 1
CH-7302 Landquart
+31 81 307 8111
+31 81 307 8100