A 6.8 TOPS/W Energy Efficiency, 1.5µW Power Consumption, Pulse Width Modulation Neuromorphic Circuits for Near-Data Computing with SSD

This paper proposes Pulse Width Modulation Neuromorphic (PWMNM) circuits, best suitable for near-data computing with SSD. The concept of neuromorphic computing is mimicking human brain with a number of mimicked-neurons of analog circuits. A mimicked-neuron in proposed PWMNM consists of 10 pairs of oscillators & clock counters, a micro controller and a charge pump. Proposed PWMNM achieves high energy efficiency of 6.8TOPS/W (Tera operations per second per Watt) and low power consumption of 1.5μW, thanks to matrix product operations are processed with number of output pulses as a function of the time instead of conventional current or voltage operation. Both proposed PWMNM and peripheral circuits of NAND flash memory are fabricated with a 180nm standard CMOS process and PWMNM can be embedded in SSD. Character recognition system is demonstrated by using proposed PWMNM, and the recognition accuracy of 93.3% is achieved.