Neuromorphic intermediate representation: A unified instruction set for interoperable brain-inspired computing
Spiking neural networks and neuromorphic hardware platforms that simulate neuronal dynamics are getting wide attention and are being applied to many relevant problems using Machine Learning. Despite a well-established mathematical foundation for neural dynamics, there exists numerous software and hardware solutions and stacks whose variability makes it difficult to reproduce findings. Here, we establish a common reference frame for computations in digital neuromorphic systems, titled Neuromorphic Intermediate Representation (NIR). NIR defines a set of computational and composable model primitives as hybrid systems combining continuous-time dynamics and discrete events. By abstracting away assumptions around discretization and hardware constraints, NIR faithfully captures the computational model, while bridging differences between the evaluated implementation and the underlying mathematical formalism. NIR supports an unprecedented number of neuromorphic systems, which we demonstrate by reproducing three spiking neural network models of different complexity across 7 neuromorphic simulators and 4 digital hardware platforms. NIR decouples the development of neuromorphic hardware and software, enabling interoperability between platforms and improving accessibility to multiple neuromorphic technologies. We believe that NIR is a key next step in brain-inspired hardware-software co-evolution, enabling research towards the implementation of energy efficient computational principles of nervous systems. NIR is available at neuroir.org.
Authors
- Steven Abreu (ORCID: https://orcid.org/0000-0002-2272-315X)
- Vittorio Fra (ORCID: https://orcid.org/0000-0001-9175-2838)
- Sadique Sheik (ORCID: https://orcid.org/0000-0003-0302-8511)
- Gianvito Urgese (ORCID: https://orcid.org/0000-0003-2672-7593)
- Dylan R. Muir (ORCID: https://orcid.org/0000-0003-3856-826X)
- Peng Zhou (ORCID: https://orcid.org/0000-0003-1686-6354)
- Matthias Jobst (ORCID: https://orcid.org/0000-0003-3147-1625)
- Sadasivan Shankar (ORCID: https://orcid.org/0000-0002-1555-639X)
- Gregor Lenz (ORCID: https://orcid.org/0000-0003-0619-3555)
- Felix Christian Bauer (ORCID: https://orcid.org/0000-0002-1631-7403)
- Jens Egholm Pedersen (ORCID: https://orcid.org/0000-0001-6012-7415)
- Terrence C. Stewart (ORCID: https://orcid.org/0000-0002-1445-7613)
- Bernhard Vogginger (ORCID: https://orcid.org/0000-0001-9042-5405)
- Jason K. Eshraghian (ORCID: https://orcid.org/0000-0002-5832-4054)
- Kade Heckel (ORCID: https://orcid.org/0009-0009-7021-8834)
Institutions
- University Medical Center Groningen (NL)
- University of Groningen (NL)
- Politecnico di Torino (IT)
- University of California, Santa Cruz (US)
- University of Cambridge (GB)
- SLAC National Accelerator Laboratory (US)
- Lux Research (United States) (US)
- Laboratoire de Neurobiologie Cellulaire et Moléculaire (FR)
- National Research Council Canada (CA)
- Deutsche Telekom (Slovakia) (SK)
- Technische Universität Dresden (DE)
- KTH Royal Institute of Technology (SE)
- Stanford University (US)
Publication Details
- Journal
- Nature Communications
- Published
- 2024-09-16
- DOI
- https://doi.org/10.1038/s41467-024-52259-9
- Citations
- 43
- Primary Topic
- Advanced Memory and Neural Computing
- Type
- article
- Field-Weighted Citation Impact
- 8.97
Funders
- National Science Foundation
- U.S. Department of Energy
- European Commission
- National Research Foundation
- Deutsche Forschungsgemeinschaft
- Danmarks Grundforskningsfond
- Technische Universität Dresden
- Kungliga Tekniska Högskolan
- Electronic Components and Systems for European Leadership
- Innosuisse - Schweizerische Agentur für Innovationsförderung
- Office of Science
- Office of Energy Efficiency and Renewable Energy
- Staatssekretariat für Bildung, Forschung und Innovation
- Division of Electrical, Communications and Cyber Systems
- Office of Energy Efficiency
- SLAC National Accelerator Laboratory