EUVIMEDEUVIMED entdecken
Open Access

Single-shot optical neural network

Liane Bernstein iD, Alexander Sludds iD, Christopher Panuski, Sivan Trajtenberg-Mills, Ryan Hamerly iD, Dirk Englund iD

Science Advances · 23.06.2023 · 9 (25)

DOI: 10.1126/sciadv.adg7904 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Science Advances, erschienen am 23.06.2023. Verfasst von Liane Bernstein und 5 weiteren Autor:innen. Beteiligte Einrichtung: Research Laboratory of Electronics, Massachusetts Institute of Technology, 50 Vassar St, Cambridge, MA 02139, USA. Die Arbeit wurde bisher 79-mal zitiert (Crossref) und nennt 75 Literaturangaben. Der Volltext ist frei in PubMed Central verfügbar.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Analog optical and electronic hardware has emerged as a promising alternative to digital electronics to improve the efficiency of deep neural networks (DNNs). However, previous work has been limited in scalability (input vector length K ≈ 100 elements) or has required nonstandard DNN models and retraining, hindering widespread adoption. Here, we present an analog, CMOS–compatible DNN processor that uses free-space optics to reconfigurably distribute an input vector and optoelectronics for static, updatable weighting and the nonlinearity—with K ≈ 1000 and beyond. We demonstrate single-shot-per-layer classification of the MNIST, Fashion-MNIST, and QuickDraw datasets with standard fully connected DNNs, achieving respective accuracies of 95.6, 83.3, and 79.0% without preprocessing or retraining. We also experimentally determine the fundamental upper bound on throughput (∼0.9 exaMAC/s), set by the maximum optical bandwidth before substantial increase in error. Our combination of wide spectral and spatial bandwidths enables highly efficient computing for next-generation DNNs.

Autor:innen und Institutionen

  • Research Laboratory of Electronics, Massachusetts Institute of Technology, 50 Vassar St, Cambridge, MA 02139, USA.
  • Lightmatter Inc., 100 Summer St, Boston, MA 02110, USA.
  • NTT Research Inc., Physics and Informatics Laboratories, Sunnyvale, CA 94085, USA.

ORCID-Profile: Liane Bernstein, Alexander Sludds, Ryan Hamerly, Dirk Englund

Zitieren

APA

Bernstein, L., Sludds, A., Panuski, C., Trajtenberg-Mills, S., Hamerly, R., & Englund, D. (2023). Single-shot optical neural network. Science Advances, 9(25). https://doi.org/10.1126/sciadv.adg7904

Vancouver

Bernstein L, Sludds A, Panuski C, Trajtenberg-Mills S, Hamerly R, Englund D. Single-shot optical neural network. Science Advances. 2023;9(25). doi:10.1126/sciadv.adg7904

BibTeX
@article{Bernstein2023,
  author = {Bernstein, Liane and Sludds, Alexander and Panuski, Christopher and Trajtenberg-Mills, Sivan and Hamerly, Ryan and Englund, Dirk},
  title = {Single-shot optical neural network},
  journal = {Science Advances},
  year = {2023},
  volume = {9},
  number = {25},
  doi = {10.1126/sciadv.adg7904}
}
Kostenlos starten →