A finite physical quantity neural network VLSI with a learning capability

A finite physical quantity neural network (FPQNN) VLSI with a learning capability is proposed. The FPQNN model has a feature that the total electric charge stored in all neurons in a network is monotone decreasing while recalling. Despite the restriction of never increasing, the FPQNN model can produce a good performance for pattern classification. This paper presents an analog circuit implementation with a learning capability that can calculate the movement of electric charge of the FPQNN model precisely and rapidly.