Toward an analog VLSI implementation of adaptive resonance theory (ART2)

We describe the architecture for an analog VLSI implementation of the ART2 network. This design uses primary analog circuitry, with some digital clocking for controlling switches for storage and timing. The system operates in current-mode, which simplifies much of the circuitry. The network is broken into four major blocks and we describe the circuitry that can implement these blocks. This represents an important first step toward an analog hardware implementation of ART2.

1. Introduction

Artificial Neural Networks (ANN) have proven to be very useful in many applications related to pattern classification and recognition, signal and image processing and data analysis [4] [5]. Software simulations can implement these ANN models, but the massive parallelism, which is one of the significant properties of ANN, can be realized from the hardware implementation of these models. Even though it is not an easy task to implement all ANN models on a chip or give a circuit representation for each model, but there are significant steps taken to address these hardware implementation issues [6]-[8].