Jackson Mowry and James S. Plank

July, 2025

ICONS: International Conference on Neuromorphic Systems

https://iconsneuromorphic.cc/Schedule-2025/

View Article

Abstract

VRISP is a variant of the RISP neuroprocessor whose goal is very fast CPU simulation, in both embedded and non-embedded computing scenarios. Its design focuses on memory efficiency and the leveraging of readily available vector instructions to improve the effective speed of the neuroprocessor. In this paper, we describe the methods we have employed to speed up execution, and then we perform three experiments to demonstrate the improvements. The first employs a dense, two-layer network as a benchmark to explore the various improvements, and to compare with other neuroprocessor implementations. The second evaluates a very challenging embedded computing scenario where the DBSCAN clustering algorithm is applied neuromporphically to event-based camera data. The last demonstrates performance improvements during training and inference, where the simulator’s performance is the major determining factor of overall performance. VRISP is available as open-source software.

Citation Information

Text

author     J. Mowry and J. S. Plank
title      {VRISP}: A Vectorized Open-Source Simulator for Efficient Neuromorphic Computing
booktitle  International Conference on Neuromorphic Computing Systems (ICONS)
publisher  ACM
month      July
year       2025

Bibtex

@INPROCEEDINGS{mp:24:vrisp,
    author = "J. Mowry and J. S. Plank",
    title = "{VRISP}: A Vectorized Open-Source Simulator for Efficient Neuromorphic Computing",
    booktitle = "International Conference on Neuromorphic Computing Systems (ICONS)",
    publisher = "ACM",
    month = "July",
    year = "2025"
}