Association Hardware for Intelligent Electronic Systems

摘要

Aiming at real-time event recognition, basic associa- tion hardware was developed using neuron MOS (VMOS), a multi-input threshold processing element device. TO store a large amount of past events as analog vectors, new rewri- table analog EEPROM cells were first designed, featuring high-accuracy analog voltage control without “read/ver- ify” Besides, associator hardware was developed to detect and associate nearest stored events to input events. In addi- tion, a circuit was designed for detection of motion vectors of 2D images as the most simple application of the associa- tor, and it was shown that processing only takes several hundred nanoseconds, while even dedicated DSP requires about 10 ms. As an application of 2D image compression and processing, vMOS circuit for real-time center-of-mass detection was also developed. The salient feature of vMOS circuits is the unique combination of flexibility of analog processing with accuracy of digital processing. Appropriate use of analog processing is efficient in eliminating digital bottleneck. Besides, along with analog circuitry develop- ment, vector quantization processor was designed as a digital circuit, and applied to real-time compression of moving images.