Implementation of the Neural Engineering Framework on the TrueNorth Neurosynaptic System

Abstract

The Neural Engineering Framework (NEF) provides a methodology for implementing algorithms and models using spiking neurons. Although it is possible to run simulations based on the NEF on Von Neumann hardware, neuromorphic hardware holds the promise of increased computational efficiency and lower power implementation. This work describes an implementation of the NEF on IBM's TrueNorth Neurosynaptic system. Using one TrueNorth chip, a NEF neural population of 629 neurons representing five dimensions is demonstrated on hardware. However, the crossbar array architecture itself, utilized in the TrueNorth hardware, can be used to compute the basic NEF calculations for any sized neural population, representing any dimensionality. The computation time is a function of the maximum values used in the computations.