Output Estimation of Photonic Reservoir Systems Considering Characteristics of Analog MAC Circuits
Photonic reservoir computing (PRC) has attracted considerable attention as an emerging AI accelerator for ultrafast intelligent processing with low power consumption. However, performing all processing optically remains challenging, neces-sitating the support of electrical circuits for post-processing. To address this, we have prototyped an analog multiply-and-accumulate (MAC) circuit designed for the readout process of the reservoir computing. Our results demonstrate that nonlinearity and noise due to the circuit do not significantly affect the PRC system. It suggests analog circuits could be used for readout and actively use this nonlinearity for reservoir computing.
Authors
- Satoru Tanoi (ORCID: https://orcid.org/0000-0001-9173-8569)
- Akifumi Kasamatsu (ORCID: https://orcid.org/0000-0002-2217-6272)
- Satoshi Sunada (ORCID: https://orcid.org/0000-0003-0466-8529)
- Kawori Sekine (ORCID: https://orcid.org/0000-0003-4166-0796)
- Shinsuke Hara (ORCID: https://orcid.org/0000-0002-5183-9725)
- Kazuyuki Wada
- Yuichiro Ohtsuka
Institutions
- Kanazawa University (JP)
- Meiji University (JP)
- National Institute of Information and Communications Technology (JP)
Publication Details
- Published
- 2024-11-18
- DOI
- https://doi.org/10.1109/icecs61496.2024.10849040
- Primary Topic
- Neural Networks and Reservoir Computing
- Type
- article
- Field-Weighted Citation Impact
- 0.00