Metraipolis › archive › dc_b65ee8de4032

Enabling Retraining-Free Accurate Inference in Analog Compute in Memory Accelerators — arXiv:2611.08693

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.08693
License (as archived)
CC0-1.0
Archived
2026-11-09T22:36:55.987Z
SHA-256 of the archived text
17d3182a89c629bd2c4eb096b86b01a794acf578161072f596cb4d010a1e136f
Stable URL of this page
https://metraipolis.com/documento/dc_b65ee8de4032

How to cite

Enabling Retraining-Free Accurate Inference in Analog Compute in Memory Accelerators — arXiv:2611.08693 — arXiv (en), https://arxiv.org/abs/2611.08693 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_b65ee8de4032 (Metraipolis archive dc_b65ee8de4032, sha256 17d3182a89c629bd2c4eb096b86b01a794acf578161072f596cb4d010a1e136f)
BibTeX
@misc{dcb65ee8de4032,
  title = {Enabling Retraining-Free Accurate Inference in Analog Compute in Memory Accelerators — arXiv:2611.08693},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.08693},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_b65ee8de4032; sha256 17d3182a89c629bd2c4eb096b86b01a794acf578161072f596cb4d010a1e136f},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.08693. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Analog compute-in-memory (CIM) accelerators compute matrix-vector products directly inside memory arrays, eliminating weight movement and promising order-of-magnitude gains in inference efficiency. However, converter and device errors make pretrained models unusable on these arrays, and hardware-aware retraining is costly. We enable retraining-free, accurate inference of pretrained models on simple analog arrays with low-cost, shared calibration. Our key observation is that analog error depends on how each layer is represented on the array, and that this representation can be changed offline without changing the computation. We center the inputs and apply Hadamard rotations designed for analog arrays, which keep shared reference errors spread and confine input mixing to each array. Across CNNs, vision transformers and LLMs up to 14B parameters on PCM and ReRAM, the method brings vision models and larger LLMs within a few points of floating-point accuracy and matches or approaches published hardware-aware training without training any weights. On ReRAM with 6-bit ADCs, it restores ViT-B/16 from 37.4% to 82.9% and Phi-4 from 27.0% to 72.2%, both within 2.2 points of floating-point accuracy.

Tema degli esploratori: calcolo e infrastrutture (elenco arXiv cs.AR, nuovi lavori del giorno).
Categorie arXiv: Hardware Architecture (cs.AR) ; Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2611.08693 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-11-09T22:31:32.015Z dalla pagina https://arxiv.org/list/cs.AR/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).