James S. Plank, Charles P. Rizzo, Bryson Gullett, Keegan E. M. Dent and Catherine D. Schuman

September, 2025

Journal of Low Power Electronics and Applications (Open access)

https://www.mdpi.com/2079-9268/15/3/50

PDF not available yet, or is only available from the conference/journal publisher.

Abstract

For most, if not all, AI-accelerated hardware, communication with the agent is expensive and heavily bottlenecks the hardware performance. This omnipresent hardware restriction is also found in neuromorphic computing: a novel style of computing that involves deploying spiking neural networks to specialized hardware to achieve low size, weight, and power (SWaP) compute. In neuromorphic computing, spike trains, times, and values are used to communicate information to, from, and within the spiking neural network. Input data, in order to be presented to a spiking neural network, must first be encoded as spikes. After processing the data, spikes are communicated by the network that represent some classification or decision that must be processed by decoder logic. In this paper, we first present principles for interconverting between spike trains, times, and values using customdesigned spiking subnetworks. Specifically, we present seven networks that encompass the 15 conversion scenarios between these encodings. We then perform three case studies where we either custom design a novel network or augment existing neural networks with these conversion subnetworks to vastly improve their communication performance with the outside world. We employ a classic space vs. time tradeoff by pushing spike data encoding and decoding techniques into the network mesh (increasing space) in order to minimize intraand extranetwork communication time. This results in a classification inference speedup of 23x and a control inference speedup of 4.3x on field-programmable gate array hardware.

Citation Information

Text

author     J. S. Plank and C. P. Rizzo and B. Gullett and K. E. M. Dent and C. D. Schuman
title      Alleviating the Communication Bottleneck in Neuromorphic Computing with Custom-Designed Spiking Neural Networks
doi        10.3390/jlpea15030050
url        https://www.mdpi.com/2079-9268/15/3/50
year       2025
publisher  MDPI
article-number 50
issn       2079-9268
journal    Journal of Low Power Electronics and Applications
volume     15
number     3
pages      1-27

Bibtex

@ARTICLE{prg:25:acb,
    author = "J. S. Plank and C. P. Rizzo and B. Gullett and K. E. M. Dent and C. D. Schuman",
    title = "Alleviating the Communication Bottleneck in Neuromorphic Computing with Custom-Designed Spiking Neural Networks",
    doi = "10.3390/jlpea15030050",
    url = "https://www.mdpi.com/2079-9268/15/3/50",
    year = "2025",
    publisher = "MDPI",
    article-number = "50",
    issn = "2079-9268",
    journal = "Journal of Low Power Electronics and Applications",
    volume = "15",
    number = "3",
    pages = "1-27"
}