An Adaptive Threshold Logic Gate Using Capacitive Analog Weights

A physical realization of an adaptive threshold logic gate which can be used to realize linearly separable switching functions is presented. The system is completely electronic and is easily implemented in the sense that standard components may be used throughout. The memory circuit used for each weight of the device stabilizes the voltage across a capacitor by means of a sampling technique. Over the range that the capacitor voltage is stabilized, the net average capacitor current is zero at a discrete number of stable voltages which can be made numerous enough to approximate analog memory. The device exhibits both long-term stability and ease in weight adjustment. Training results are given for several linearly separable switching functions. This circuit shows that a practical adaptive threshold logic gate can be realized. The incorporation of integrated circuitry in the device presented would greatly enhance the feasibility of building more complex trainable machines.