Processor's Parallelism Helps Neural Networks Learn Quickly

For fast learning and recognition with artificial neural networks, NC3001 VLSI parallel processor is based on an architecture optimized for the implementation of the learning algorithm, a competitive alternative to back-propagation that is said to lead to very low cost, compact VSLI implementations. Its high processing power, low power dissipation and limited chip size make it well-suited for embedded neural, fuzzy and general filtering applications. The device provides 32 fixed-point, fully parallel digital multiply-and-accumulate processors (MACs) operating in parallel with three-stage pipeline and on-chip weight memory. It also offers a simple chip interface for coprocessor operation in µP systems. Performance is 1000 MOPS with a 30-MHz clock, processing efficiency is 30 MOPS/W, and power consumption is 1W at 30 MHz in a 132-pin PGA package. Evaluation boards are available for various systems.