Artificial Intelligence
1 Nov 2021

Analogue neuromorphic computing for onboard artificial intelligence

Fig. 1: Photograph of the analogue neuromorphic chip HICANN-Xv2 developed at Heidelberg University. Image credit: Electronic Vision(s) Group, Heidelberg University.
Fig. 1: Photograph of the analogue neuromorphic chip HICANN-Xv2 developed at Heidelberg University. Image credit: Electronic Vision(s) Group, Heidelberg University.

With a power consumption of only around 20W, the human brain is still one of the most marvelous and fascinating "computing machines" we know of. To untangle how computing works in the brain and harness potential benefits from it, neuromorphic computing aims at building hardware systems which follow an architecture that more closely resembles the brain - with a focus on spike-based computing as well as massively parallel and distributed processing [1].

In the case of analogue neuromorphic hardware, physical equivalents that emulate the behaviour of real neurons and synapses are constructed in silicon [2]. Such a system comes with intruiging properties like strongly accelerated emulation speeds (compared to biology) and low energy consumption [3-5] - making these hardware systems an attractive candidate for future onboard artificial intelligence systems in spacecraft. Moreover, different from digital neuromorphic hardware, analogue neuromorphic computing explores the question of how robust computing can occur in a substrate with, e.g., non-identical components (i.e. due to the manufacturing process, neurons are not identical) and limited parameter resolution (such as synaptic weights) [4,6].


Project overview

The aim of this project is to explore and benchmark currently developed analogue neuromorphic systems for space applications (e.g. scene classification using Cubesats) and estimate the potential impact of this emerging technology for future space missions. This research is done together with colleagues of the University of Heidelberg as part of an ongoing Ariadna study.


References:

[1] Frenkel, C., Bol, D., & Indiveri, G. (2021). Bottom-up and top-down neural processing systems design: Neuromorphic intelligence as the convergence of natural and artificial intelligence. arXiv preprint arXiv:2106.01288.

[2] Schemmel, J., Billaudelle, S., Dauer, P., & Weis, J. (2022). Accelerated analog neuromorphic computing. In Analog Circuits for Machine Learning, Current/Voltage/Temperature Sensors, and High-speed Communication (pp. 83-102). Springer, Cham.

[3] Billaudelle, S., Stradmann, Y., Schreiber, K., Cramer, B., Baumbach, A., Dold, D., ... & Meier, K. (2020). Versatile emulation of spiking neural networks on an accelerated neuromorphic substrate. In 2020 IEEE International Symposium on Circuits and Systems (ISCAS) (pp. 1-5). IEEE.

[4] Kungl, A. F., Schmitt, S., Klähn, J., Müller, P., Baumbach, A., Dold, D., ... & Petrovici, M. A. (2019). Accelerated physical emulation of Bayesian inference in spiking neural networks. Frontiers in neuroscience, 1201.

[5] Göltz, J., Kriener, L., Baumbach, A., Billaudelle, S., Breitwieser, O., Cramer, B., ... & Petrovici, M. A. (2021). Fast and energy-efficient neuromorphic deep learning with first-spike times. Nature machine intelligence, 3(9), 823-835.

[6] Wunderlich, T., Kungl, A. F., Müller, E., Hartel, A., Stradmann, Y., Aamir, S. A., ... & Petrovici, M. A. (2019). Demonstrating advantages of neuromorphic computation: a pilot study. Frontiers in neuroscience, 13, 260.

Outcome

Artificial Intelligence Conference paper
Towards Large-scale Network Emulation on Analog Neuromorphic Hardware
Arnold, Elias and Spilger, Philipp and Straub, Jan V. and Müller, Eric and Dold, Dominik and Meoni, Gabriele and Schemmel, Johannes
arXiv preprint
(2024)
Download
BibTex
Hamburger icon
Menu
Advanced Concepts Team