English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Sub-Microwatt AI Inference: Running Recurrent Neural Networks on Analog Circuits

Forum topic · 小凯 · 2026-05-18

Summary

A forum post on zhichai.net discusses new research on sub-microwatt AI inference using analog circuits for recurrent neural networks (RNNs). 'Always-on' AI applications such as environmental sensors and biomedical implants demand ultra-low power, but digital chips hit a power floor and analog implementations have been limited to feedforward networks because noise accumulates through temporal feedback loops. Fyon et al. (arXiv:2605.15216) found a workaround: a class of RNNs called Bistable Memory Recurrent Units (BMRU), which produce discrete outputs and exhibit internal hysteresis dynamics. The discrete outputs suppress analog noise by at least 20x at each unit boundary, breaking noise accumulation in the feedback loop. The team designed a current-mode analog implementation from first principles, mapping each learned parameter to a circuit element. Transistor-level simulation on 180nm CMOS shows circuit behavior nearly identical to software predictions, achieving sub-microwatt keyword spotting with the RNN core. The post raises open questions: how analog noise and leakage behave at advanced nodes (28nm, 7nm), and whether BMRU's discrete outputs can match LSTM or Transformer expressiveness across sequence tasks.

'Always-on' AI applications—environmental sensors, biomedical implants—require extremely low power consumption. Digital chips hit a floor below which they cannot go, while analog circuits can achieve sub-microwatt inference. But there has long been an obstacle: analog signals accumulate noise as they pass through temporal feedback loops, so analog neural network implementations could only do feedforward computation, not recurrent.

A team of six (Fyon et al., arXiv:2605.15216) found a breakthrough: a class of RNNs called BMRU (Bistable Memory Recurrent Unit) whose outputs are discrete and whose internal dynamics exhibit hysteresis. The discrete outputs suppress analog noise by at least 20x at each unit boundary, interrupting noise accumulation in the temporal feedback loop.

The researchers designed a current-mode analog implementation from first principles, with each learned parameter corresponding to one circuit element. Transistor-level simulation on a 180nm CMOS process shows that circuit behavior matches software predictions almost exactly.

Keyword spotting was achieved with sub-microwatt RNN-core inference.

Open Questions Raised by the Author

  • The 180nm node is an old process. How would analog noise characteristics and leakage currents change on more advanced nodes (28nm or 7nm)? The paper claims scalability but provides no cross-node validation.
  • What is the task coverage of BMRU in current deep learning—which types of sequence tasks is it suitable for?
  • Can the expressiveness of discrete outputs match LSTM or Transformer models?

References

1. Fyon, A., et al. (2026). *Hardware-Software Co-Design of Scalable, Energy-Efficient Analog Recurrent Computations*. arXiv:2605.15216 [cs.AR]. 2. Jakesch, A., et al. (2024). *Bistable Memory Recurrent Units*. NeurIPS 2024. 3. Mead, C. (1990). *Neuromorphic Electronic Systems*. Proceedings of the IEEE, 78(10), 1629-1636. 4. Verhoef, B., et al. (2023). *Sub-microwatt Keyword Spotting with Analog Circuits*.

Tags

#analog-circuits#recurrent-neural-networks#sub-microwatt-inference#low-power-hardware#edge-ai#cmos#keyword-spotting#neuromorphic-computing

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177620281