Neuromorphic Visual Processing Systems Implemented in Analog VLSI

National Science Foundation, Career Award (ECS-9984386), A.H. Titus

The project presented in this proposal combines educational and research activities through exploration and development.

The research portion of the work will be formed around the field of analog VLSI systems with an emphasis on neuromorphic processing.
A unique approach will be taken to implement an integrated artificial visual system that is formed from distinct components that are
designed to work together seamlessly. The entire system will be designed to operate as an entire, functional unit, but will be designed as
modular pieces with each piece created individually; the complexity of the visual system requires that it be modeled in manageable
pieces. The visual system will consist of the low-level processing functions performed by the retina, mid-level stereopsis processing,
and high-level object recognition. The retina function has been implemented successfully, but modifications will be made to previous
designs to accommodate the connections to other levels of processing. A unique object recognition system will be designed based on current psycho-physical and biological models of object recognition that is compatible with analog VLSI. The final system will be a silicon-based system that performs multiple visual functions and implemented on an analog VLSI platform. The proposed research will provide new insights into the development of artificial visual systems through the concept of modularization. This type of processing-specific analog circuit is compact and has very low power consumption, making it ideally suited for low cost, high volume, and autonomous exploration devices. Also, this work will generate a set of design standards for analog VLSI, which will aid the mainstreaming of this technology.