Accelerating ML models and NP-hard optimisation problems using light
With Moore’s law plateauing and digital specialization approaching its ultimate limit, digital computing faces scalability, performance, and sustainability challenges, calling out for non-traditional compute paradigms. Our novel computer leverages optical and analog technologies and hardware-application codesign to accelerate ML inference and combinatorial optimization, achieving over 100x efficiency improvement in TOPS/Watt.
Authors
- Daniel Cletheroe (ORCID: https://orcid.org/0009-0003-6444-9149)
- L. Pickup (ORCID: https://orcid.org/0000-0002-1728-3145)
- Babak Rahmani (ORCID: https://orcid.org/0000-0002-0856-9635)
- Jannes Gladrow (ORCID: https://orcid.org/0000-0002-2973-4492)
- Greg O’Shea
- Francesca Parmigiani (ORCID: https://orcid.org/0000-0001-7784-2829)
- Christos Gkantsidis (ORCID: https://orcid.org/0000-0002-6898-2368)
- Burcu Canakci (ORCID: https://orcid.org/0000-0002-8406-7768)
- Antony Rowstron (ORCID: https://orcid.org/0009-0009-5936-6895)
- Grace Brennan (ORCID: https://orcid.org/0000-0001-7081-0509)
- James Clegg (ORCID: https://orcid.org/0009-0002-3428-7161)
- Kirill P. Kalinin (ORCID: https://orcid.org/0000-0003-0622-3056)
- Hitesh Ballani (ORCID: https://orcid.org/0000-0003-1573-3314)
- Jiaqi Chu
- Douglas J. Kelly
- H. Kremer
Institutions
- Microsoft Research (United Kingdom) (GB)
Publication Details
- Published
- 2024-11-10
- DOI
- https://doi.org/10.1109/ipc60965.2024.10799574
- Primary Topic
- Robotic Path Planning Algorithms
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- Microsoft Research