High-Density Digital Neuromorphic Processor with High-Precision Neural and Synaptic Dynamics and Temporal Acceleration

Abstract

This paper presents a large-scale digital neuromorphic processor for spiking neural network emulation. The processor is fabricated in a 28 nm CMOS and occupies a 9.00 mm(2) die area for 262,144 neurons. A time-embedded floating-point leaky integrate-and-fire neuron model is implemented to reduce the size of SRAM and the number of SRAM accesses. It achieves a high-density neuron integration (34.9 k neurons/mm(2)), which is 13 times higher than other state-of-the-art designs. Additionally, it achieves a high-dynamic range synapse (8-bit floating-point) and low energy consumption (28.26 pJ/synaptic operation).