Modern age intelligent systems will require extensive computational power, complex parallel processing units, and low-power design. Biologically inspired neuromorphic VLSI systems present a viable solution to the demands of both highly parallel, and low-power consuming processors. Among biological sensory systems, vision is the most important one with the largest portion of the brain devoted to visual computations. Biological models for tasks like visual motion computation, target pursuit, and crash-avoidance have been extensively studied. In this thesis, a biologically inspired target-fixation model has been analyzed and implemented in VLSI. A modular approach for designing a sender-receiver based tracking system has also been discussed. A spiking-neuron sender chip has been implemented using a frequency-encoded event driven communication protocol. This sender chip is used to relay information about changes in image intensity to a computational unit in a modular visual system. The use of these neuromorphic chips has been suggested for developing monolithic and modular target-tracking systems for yaw-torque control in a robot. The mixed-signal chips work in the subthreshold region of the MOSFET and consume very little electrical power. Subthreshold implementations are very well suited for the low-frequency behavior of real-world tracking systems. System level architecture and simulations of such a tracking system have also been presented.
Vivek Pant, "Modular Neuromorphic VLSI Architectures for Visual Motion and Target Tracking," MS thesis (advisor: Charles M. Higgins), Department of Electrical and Computer Engineering, The University of Arizona, June 2003.