Slow electronics with reservoir computing

Isao H. Inoue
AIST, Tsukuba, JAPAN

Processing low-frequency signals is challenging for digital computers as they require significant energy despite their high-speed data processing abilities. This poses an issue for biomedical wearables and environmental monitors that require real-time processing of slow signals in energy-limited ‘edge’ environments with small batteries. One proposed solution is event-driven processing with non-volatile memory, which is highly energy-consuming and unsuitable for edge environments. To tackle this challenge, we suggest the development of ‘slow electronics’ that can process low-frequency signals more efficiently. The biological brain is an excellent example of slow electronics, which processes low-frequency signals in real time with exceptional energy efficiency.
We have integrated slow electronics with a spiking neural network (SNN) to enable real-time data processing in edge environments without internet connectivity. This integration can reveal the determinism or periodicity behind unconscious behaviours and habits that have been difficult to explore due to privacy barriers. It may also provide a deeper understanding of a craftsman’s skills, which they may not even be aware of.
In this talk, I will discuss the recent concepts and technological developments in slow electronics, including reservoir calculation, analogue CMOS circuits, artificial neuromorphic devices, and numerical simulation with extended time constants.