System of Interconnected Reconfigurable Linear Threshold Circuits

We have developed and fabricated a reconfigurable system based on Linear Threshold Elements (LTEs) that perform a variety of common logic operations using one basic structure. A single LTE compares a sum of weighted-inputs to a threshold and produces a Boolean output. This operation is equivalent to the basic functions defined for artificial neural networks (ANNs). The reconfigurable structure was created by incorporating memory in the LTE and combining multiple LTEs together to produce a wider range of possible functions. The LTE structure presented in this paper is explicitly programmed by a host for logic tasks spanning multiple LTEs. However, the possibility exists to implement supervised learning or other neural network learning algorithms to train the LTE network for the desired operation.