A feedforward neural network for CMOS VLSI implementation

A feedforward neural network circuit model suitable for CMOS VLSI implementation is introduced. The model captures the principles of operation of artificial neural nets, and is suitable for analog all-MOS VLSI circuit implementations. Each unit consists of a control device and one operational amplifier which is implemented with two CMOS inverters in series. A control device is implemented with one n-MOS and one p-MOS transistors to model biasing. A single n-MOS transistor is used to connect the output of a unit of one layer to the input of the next layer. All the connection weights are set by applying analog signals to the gates of these transistors.<>