MOS fully analog reinforcement neural network chip
M. Al-Nsour, H.S. Abdel-Aty-Zohdy · 2002
This paper addresses the design and implementation of an analog MOS reinforcement neural network by compact and novel subcircuits. System implementation was optimized for minimum silicon area and maximum input signal swing. The chip, consisting of two three-input neurons, is designed and implemented using 1.5 /spl mu/m CMOS n-well technology and occupied 0.114 mm/sup 2/. Due to the limited number of pads on a TinyChip, only two neurons were implemented. The ANN system is to be used for gas recognition applications, with present off-chip learning. Learning through digital genetic algorithms implementation is successfully achieved, and will be further implemented in silicon for integrated system-on-a-chip.