A low power CMOS imager based on time-to-first-spike encoding and fair AER

The paper presents a CMOS vision sensor based on a biologically inspired data representation referred to as time-to-first-spike (TFS) encoding combined with a fair address event representation (AER) scheme. Our approach is different from conventional methods because the read-out of information is initiated by the pixel itself, while access to the read-out bus is granted only once to each pixel, after which it enters into a stand-by mode. This approach allows a great saving in dynamic power consumption and extensive reduction of the inefficiencies due to periodic requests of the bus in the case of spiking pixels. Transmission bandwidth is significantly improved using the proposed circuitry. Each pixel includes only 14 transistors and occupies an area of 15×15 μm², with a fill factor of 33% using a 0.35 μm process. The average current consumption is estimated to be 10 nA per pixel, which is 3 orders of magnitude lower compared with that of the spiking pixel.