Ifi Research: Microelectronic Systems: Nanooelectronic Systems: Neuromorphic Systems: Student Projects
Our eye is in many respects different from a camera. For instance, it does not make sense at all to ask at what `frame rate' our eyes opperate. There are no frames! Every `pixel' opperates on its own timing, asynchronously. It conveys its information independently from all other pixels as a series of voltage pulses to the brain. In general, light intensity is encoded as the frequency of these pulses.
It is, however, a fact shown in psychophysical experiments that only a few pulses from every pixels are necessary for humans to process a picture, so few that no average pulse rate can be computed with any accuracy. Therefore, a certain encoding of information with only one spike from each pixel has been suggested, the so called time-to-first-spike encoding. In this scheme, pixel brightness is encoded in the time that a pixel needs to fire a pulse/spike after it has been reset: the brighter the pixel, the shorter the delay.
It is not so trivial to produce a colour camera chip the traditional way, that is with colour filters that are placed very precisely over the pixels. Complicated post processing of a CMOS chip is necessary to achieve this placement.
There is a more recent technique, though, that does not require any post processing at all. The company Foveon refined the idea to use photo sensors that are burried in different depths below the surface of the silicon chip. Light of different colour/wavelength actually penetrates the silicon up to different depths. A photo sensor close to the surface will see blue light while one deeper down will only see red light. Conveniently, a light sensor in CMOS is nothing else but a PN-junction (a PN-diode) which is easily obtained in different depths in any normal CMOS chip production process. One can even have several of those diodes in the same 2D location, stacked on top of each other at different depths. Thus, one can get information on different colours from the very same pixel location instead of only one colour per pixel as with colour filters.
The aim of this project is to develop a color vision sensor that uses the time-to-first-spike encoding to convey vision information to a computer. It entiles ASIC and PCB design and testing using optical end electronic equipment. Come and talk to me to find out more.