All-in-One Analog AI Accelerator: On-Chip Training and Inference with Conductive-Metal-Oxide/HfO x ReRAM Devices
Donato Francesco Falcone, Victoria Clerico, Wooseok Choi, Tommaso Stecconi, Folkert Horst, Laura Bégon-Lours, Matteo Galetta, Antonio La Porta, Nikhil Garg, Fabien Alibart, Bert Jan Offrein, Valeria Bragaglia · 2025
Read All-in-One Analog AI Accelerator: On-Chip Training and Inference with Conductive-Metal-Oxide/HfO x ReRAM Devices by Donato Francesco Falcone, Victoria Clerico, Wooseok Choi, Tommaso Stecconi, Folkert Horst, Laura Bégon-Lours, Matteo Galetta, Antonio La Porta, Nikhil Garg, Fabien Alibart, Bert Jan Offrein, Valeria Bragaglia (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-04524.txt.
Description
Donato Francesco Falcone and co-authors present a chip that both trains and runs AI in a 2025 paper. Most past analog chips did one task or the other, not both on one base. Their device uses memory cells that switch resistance, built into the chip's back layers.
The cells switch at low voltage and hold over thirty stable levels with low noise. That calm makes weight transfer near ideal, far better than rival memory types. Tests on a grid array show error far lower than rivals, even years after setting. A single base that trains and then keeps running could cut the energy cost of AI.
Catalog record
| CREATOR | Donato Francesco Falcone, Victoria Clerico, Wooseok Choi, Tommaso Stecconi, Folkert Horst, Laura Bégon-Lours, Matteo Galetta, Antonio La Porta, Nikhil Garg, Fabien Alibart, Bert Jan Offrein, Valeria Bragaglia |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 71 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-04524