Research in VLSI System Implementation of Neuromorphic Learning Networks

摘要

The methodology of the researchers was to build experimental prototype learningsystems they wanted: to develop a prototype of an enhanced neuron/synapse chip using some ideas that they have gained from existing chips, develop a prototype VME based experimental platform for the above devices, write experimental prototype system software to run the above prototype boards and chips as co-processors for typical computer system such as a SUN4 and develop new algorithms to perform other types of learning suitable for prototype VLSI implementation. The following results were achieved: System Level Hardware-redesigned prototype learning chips were fabricated, System Level Software-software modules to interface with their prototype system has has been written, Algorithms-theoretical and simulation experiments were carried out to gauge the efficiency of one-weight-at-a-time vs. parallel perturbations.