Keynote Information

Title: Energy Efficiency and Scalability of the TrueNorth Neurosynaptic Processor

Abstract

Inspired by biological brains, we built a semiconductor chip called the TrueNorth processor, which is equipped with 4096 computing cores, one million digital neurons and 256 million binary synapses. This chip can run various types of neural networks while consuming 65mW of power. This extreme efficiency is achieved by the architectural choice of moving data around as little as possible. We further reduced the power consumption by using asynchronous event-driven circuits in the chip implementation. The TrueNorth processor scales naturally to multi-chip systems with its built-in chip-to-chip communication interface. This allows us to envision a neurosynaptic supercomputer that matches mammalian brains in scale within a realistic power budget.

Bio

Jun Sawada received the B.S. and M.S. degrees in mathematics from Kyoto University, Kyoto, Japan, and the Ph.D. degree in computer sciences from the University of Texas at Austin, Austin, TX, USA, for the study in formal verification of hardware, VLSI microarchitecture, automated theorem proving and automated reasoning. He has been a Research Staff Member at IBM Research since 2000. At IBM, he participated in many chip development projects, including the ones for Power microprocessors and BlueGene supercomputers, and pioneered the application of formal verification technologies to large scale microprocessor designs. Recently he has been working on neurosynaptic chip and system designs, and took a leading role in the development team of the TrueNorth chip.

Proceedings available at the ACM Digital Library

Please do not forget to provide an evaluation here