A solid-state electronic linear adaptive neuron with electrically alterable synapses

The authors address the hardware implementation of a semiconductor device which emulates the biological synaptic interconnection for the hardware realization of neural network systems. Specifically, they describe work on the electrically reprogrammable (alterable) SONOS (silicon blocking oxide nitride tunneling oxide silicon) nonvolatile synapse and a simple electronic neuron which incorporates these alterable synapses. The electronic synaptic interconnection strength, or the weight value, can be electrically altered at CMOS voltage levels. The authors have incorporated these modifiable synaptic weights into a solid-state electronic linear adaptive neuron with a Widrow-Hoff's delta learning rule as the updating algorithm to examine the electrical performance of these programmable synapses. The experimental results and the desirable features of these electronic synapses are discussed.<>