USE OF A-SI-H MEMORY DEVICES FOR NONVOLATILE WEIGHT STORAGE IN ARTIFICIAL NEURAL NETWORKS

An Artificial Neural Network (ANN) is an ensemble of simple processing units interconnected by variable strength weights. VLSI ANNs use eith er dynamic techniques or non-volatile EEPROM technology for weight sto rage. a-Si:H memory devices offer an alternative method for the non-vo latile storage of analogue weight values. Results are presented from a test chip on which a-Si:H analogue memory devices were fabricated on the surface of a conventional CMOS chip. The design of a second chip i s discussed, in which a-Si:H devices are used to store the synaptic we ights.