On Chip Customized Learning on Resistive Memory Technology for Secure Edge AI
This paper presents the first experimental demonstration of few-shot on-chip training on an in-memory computing resistive memory (ReRAM) platform. We use the Model-Agnostic Meta-Learning (MAML) algorithm to reduce training iterations and associated ReRAM conductance updates by orders of magnitude. Through co-optimization of device programming conditions and the algorithm, we achieve <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$>97 {\%}$</tex> accuracy on the Omniglot dataset after just five training iterations (i.e., ReRAM programming operations) while improving device retention at 150°C.
Authors
- Djohan Bonnet (ORCID: https://orcid.org/0009-0002-1018-3712)
- G. Molas (ORCID: https://orcid.org/0000-0002-7345-4164)
- Marco Piccoli (ORCID: https://orcid.org/0000-0002-9461-2715)
- S. D’Agostino (ORCID: https://orcid.org/0009-0007-2006-4764)
- N. Castellani (ORCID: https://orcid.org/0009-0007-0763-2130)
- Damien Querlioz (ORCID: https://orcid.org/0000-0002-0295-1008)
- Louis Hutin (ORCID: https://orcid.org/0000-0001-6429-3867)
- F. Andrieu (ORCID: https://orcid.org/0000-0002-1951-1033)
- E. Vianello
- G. Piccolboni
- J. F. Nodin
- M. A. Iftakher
- M. Pallo
Institutions
- Centre National de la Recherche Scientifique (FR)
- Commissariat à l'Énergie Atomique et aux Énergies Alternatives (FR)
- Université Paris-Saclay (FR)
- Centre Inria de Saclay (FR)
- Laboratoire d'Électronique des Technologies de l'Information (FR)
- Université Grenoble Alpes (FR)
Publication Details
- Published
- 2025-06-08
- DOI
- https://doi.org/10.23919/vlsitechnologyandcir65189.2025.11074789
- Citations
- 1
- Primary Topic
- Advanced Memory and Neural Computing
- Type
- article
- Field-Weighted Citation Impact
- 0.73