AN ANALOG NEUROCHIP AND ITS APPLICATIONS TO MULTILAYERED ARTIFICIAL NEURAL NETWORKS

Fujitsu developed a multi-purpose neurochip (analog neuroprocessor) wh ich was designed for practical applications that would incorporate neu ral networks into equipment. This paper describes this analog neurochi p and the multilayered artificial neural networks that run on it. This new chip was built using 2 mum Bi-CMOS technology. Input-output signa ls are analog; while weights are represented in digital. An analog tim e-sharing bus is used for input and output; analog data can be sampled at intervals of 13 mu s. The current chip has some problems, such as dispersion of device characteristics, loss of significant digits of we ight, and digit overflow. Some countermeasures for these problems were also examined.