Demonstration of a subthreshold analog CMOS reservoir chip for temporal signal processing
Abstract We present a low-power analog CMOS reservoir computing chip that employs a simple ring topology for unidirectional nonlinear analog processing. Each node integrates a sample-and-hold circuit operating at 1 kHz, enabling compact and energy-efficient implementation in standard CMOS. The fabricated chip was evaluated on standard benchmarks, achieving a linear memory capacity of 13.4 and information processing capacities of 7.2 (second-order), 3.3 (third-order), and 1.2 (fourth-order). It also demonstrated competitive accuracy in both short- and long-term time-series forecasting tasks. Most notably, this performance was achieved with a power dissipation of only 20 μ W per core, highlighting the potential of analog CMOS reservoir computing for deployment in energy-constrained edge devices.
Authors
- Yukio Terasaki (ORCID: https://orcid.org/0000-0002-1304-936X)
- Kota Ando (ORCID: https://orcid.org/0000-0001-8648-3768)
- Kazuki Nakada (ORCID: https://orcid.org/0000-0003-2375-2603)
- Shin-ichiro Mochizuki
- Keita Suda
- Eiji Suzuki
- Tomoyuki Sasaki
- Abe Yuki
- Tetsuya Asai
- Shimon Matsuno
- Yukihiro Urakawa
Institutions
- TDK (Japan) (JP)
- Hokkaido University (JP)
Publication Details
- Journal
- npj Unconventional Computing
- Published
- 2026-03-02
- DOI
- https://doi.org/10.1038/s44335-026-00059-3
- Primary Topic
- Neural Networks and Reservoir Computing
- Type
- article
- Field-Weighted Citation Impact
- 0.00