A fast, novel, cascadable design for multi-layer networks

It is now well established that multi-layer neural networks can achieve results in pattern recognition problems which are comparable to, if not slightly better than, the results obtained with the best conventional algorithms available. The authors argue that if analogue VLSI technology is to represent a real alternative to digital implementations of these multilayer networks, it must not only deliver the high computational rates made possible by massive parallelism, but also be capable of high throughput; the designs must be cascadable and their operation made as independent of process variation as possible