Implementing Intelligence in Silicon Integrated Circuits Using Neuron-Like High-Functionality Transistors

Abstract

The primary objective of this article is not to present integrated circuit implementation of neural networks in the sense that neurophysiological models are constructed in electronic circuits, but to describe new-architecture intelligent electronic circuits built using a neuron-like high-functionality transistor as a basic circuit element. This has greatly reduced the VLSI hardware/software burden in carrying out intelligent data processing and would find promising applications in robotics. The transistor is a multiple-input-gate thresholding device called a neuron MOSFET (neuMOS or νMOS) due to its functional similarity to a simple neuron model. vMOS circuits are characterized by a high degree of parallelism in hardware computation, large flexibility in the hardware configuration, and a dramatic reduction in circuit complexity compared to conventional integrated circuits. As a result, a number of new-concept circuits has been developed. Examples include a real-time reconfigurable logic circuit called flexware and associative memory conducting a fully parallel search for the most similar targets. A simple hardware model for self-learning systems is also presented. The enhancement in functionality at a very elemental transistor level is critical to building human-like intelligent systems on silicon.