CMOS Implementation of Phase-Encoded Complex- Valued Artificial Neural Networks

摘要

The model of a simple perceptron using phase-encoded inputs and complex-valued weights is presented. Multilayer two-input and three-input complex-valued neurons (CVNs) are implemented as mixed-signal CMOS integrated circuits. High frequency AC signals are used to carry information. Analog differential amplifier and comparator circuits implement the aggregation function and activation function. Using offline learning, the CVN is shown to be superior to traditional perceptrons, with a single CVN capable of implementing all 16 functions of two Boolean variables and 245 of the 256 function of three Boolean variables without additional logic, neuron stages, or higher order terms such as those required in polynomial logic.