4×4×2 neural network design using modular neural chips with on-chip learning

Describes a feedforward artificial neural network with learning capability in a modular design. The design uses a modified error backpropagation continuous-time learning rule. The modular design is implemented onto (identical) tiny chips where each chip implements a 4-input, 2-output modular network. The modular chips can then be cascaded in order to realize a large network by adding modular chips vertically and horizontally. The authors report on test results that demonstrate the successful operation of the chip and a constructed 4×4×2 neural network using three module chips