Logo image
Back
Compiling Spiking Neural Networks to Mitigate Neuromorphic Hardware Constraints
Conference proceeding   Open access

Compiling Spiking Neural Networks to Mitigate Neuromorphic Hardware Constraints

Adarsha Balaji and Anup Das
2020 11th International Green and Sustainable Computing Workshops (IGSC), pp 1-3
19 Oct 2020
DOI: https://doi.org/10.1109/IGSC51522.2020.9290830
url
http://arxiv.org/abs/2011.13965View
SubmittedOpen Access (License Unspecified),  Open

Abstract

Biological neural networks Computation Graph Computational modeling crossbar resources energy consumption Hardware homogeneous neural units learning (artificial intelligence) Machine Learning mapping SNN-based applications neuro-synaptic cores Neuromorphic Computing neuromorphic engineering neuromorphic hardware constraints Neuromorphics neuron function Neurons pattern recognition post-synaptic neuron power-constrained platforms pre-synaptic connections spatio-temporal pattern recognition spiking neural networks Spiking Neural Networks (SNNs) SpiNeMap Synapses tile-based neuromorphic hardware
Spiking Neural Networks (SNNs) are efficient computation models to perform spatio-temporal pattern recognition on resource- and power-constrained platforms. SNNs executed on neuromorphic hardware can further reduce energy consumption of these platforms. With increasing model size and complexity, mapping SNN-based applications to tile-based neuromorphic hardware is becoming increasingly challenging. This is attributed to the limitations of neuro-synaptic cores, viz. a crossbar, to accommodate only a fixed number of pre-synaptic connections per post-synaptic neuron. For complex SNN-based models that have many neurons and pre-synaptic connections per neuron, (1) connections may need to be pruned after training to fit onto the crossbar resources, leading to a loss in model quality, e.g., accuracy, and (2) the neurons and synapses need to be partitioned and placed on the neuro-sypatic cores of the hardware, which could lead to increased latency and energy consumption. In this work, we propose (1) a novel unrolling technique that decomposes a neuron function with many pre-synaptic connections into a sequence of homogeneous neural units to significantly improve the crossbar utilization and retain all pre-synaptic connections, and (2) SpiNeMap, a novel methodology to map SNNs on neuromorphic hardware with an aim to minimize energy consumption and spike latency.

Metrics

16 Record Views
13 citations in Scopus

Details

Logo image