A reconfigurable neuromorphic VLSI multi-chip system applied to visual motion computation

We present a multi-chip neuromorphic system in which an address event representation is used for inter-chip communication. The system comprises an analog VLSI transient imager with adaptive photoreceptors, an analog VLSI motion receiver chip and a prototyping communication infrastructure which allows for programmability of connections between the elements on the two chips. We describe the properties of the two VLSI chips and of the communication infrastructure. To characterize the whole system, we present examples of connectivity tables which allow it to compute translational motion and expanding motion and show data from the transient detector array and motion receiver chips.