In this paper emerging parallel/distributed architectures are explored for the digital VLSI implementation of adaptive bidirectional associative memory (BAM) neural network. A single instruction stream many data stream (SIMD)-based parallel processing architecture, is developed for the adaptive BAM neural network, taking advantage of the inherent parallelism in BAM. This novel neural processor architecture is named the sliding feeder BAM array processor (SLiFBAM). The SLiFBAM processor can be viewed as a two-stroke neural processing engine, It has four operating modes: learn pattern, evaluate pattern, read weight, and write weight. Design of a SLiFBAM VLSI processor chip is also described. By using 2-mum scalable CMOS technology, a SLiFBAM processor chip with 4+4 neurons and eight modules of 256x5 bit local weight-storage SRAM, was integrated on a 6.9x7.4 mm(2) prototype die. The system architecture is highly flexible and modular, enabling the construction of larger BAM networks of up to 252 neurons using multiple SLiFBAM chips.

ABSTAACTNeural networks are presently being extensively explored for use in Pattern Recognition applications. Bi-directional Associative Memory (BAM) is a two-level non-linear neural network suitable for pattern recognition applications. One important performance attribute of the discrete BAM is its ability to recall stored pattern pairs, particularly in the presence of noise. In this paper the VLSI implementation of BAM is presented. A modular VLSI processor chip implementing BAM was designed. By using 2 micron CMOS technology, 4 neurons with 8 modules of 256x5 bit local weightstorage memory were integrated on a 6.9 x 7.4 mm2 die. With 4 operating modes (Learn, Evaluate, Read and Write), it is suitable to serve as a eo-processor. The system architecture is highlyflexible and modular, enabling the construction of larger BAM networks of upto 252 neurons using multiple BAM chips. Results show that real-time speeds can be achieved. The total training time for a full network of upto 252 neurons is 1.5997 ms at a clock-rate of lOh4HZ, which is fast eneough for numerous pattern recognition applications.

No abstract