We explore non-traditional forms of sensing and computation based on asynchronous and event based architectures inspired by the transmission of information on the brain (spike-based). This research is carried out by designing VLSI chips that use transistor in the sub-threshold regime1. In particular, we design sensors with embedded motion estimation and analysis circuitry. In general we would like to develop algorithms that exploit the unique structure of non-uniform sensors together we the efficiency of event based visual processing.
People: Chiara Bartolozzi, Charles Clercq
Related project: eMorph
|Details of a space-variant arrangement of pixels in a silicon retina. Left: foveal region; right: periphery|
|Layout of the Tracker Motion Sensor (TMS) chip including sensors, spatial and temporal derivatives, and attention based output coding||Block diagram of the Tracker Motion Sensor (TMS) chip including sensors, spatial and temporal derivatives, velocity detectors and attention based output coding|
1 S.-C. Liu, J. Kramer, G. Indiveri, T. Delbr¨uck, and R. Douglas. Analog VLSI:Circuitsand Principles MIT Press, 2002.
Bartolozzi, C and Indiveri, G. A silicon synapse implements multiple neural computational primitives The Neuromorphic Engineer, vol. 4, pp. 1–3. Available from: DOI 19 May 2009.