Back to today’s briefing

Story perspectives

Breakthrough Neural Network Enhances Real-Time Computing Efficiency

9/25/2025

41 4

Summary

Story summary

Researchers at the University of Tokyo developed a compact, field-programmable spiking neural network with CMOS-based reservoir computing, enabling dense on-chip networks.
It uses Leaky Integrate-and-Fire neurons and voltage-controlled oscillators. It delivers 21.7 pJ per pulse and eliminates the need for analog-to-digital converters.
FORCE learning improves real-time performance for complex tasks, with benchmark results on the NARMA10 problem.

1 of 1

Story summary

Researchers at the University of Tokyo developed a compact, field-programmable spiking neural network with CMOS-based reservoir computing, enabling dense on-chip networks.
It uses Leaky Integrate-and-Fire neurons and voltage-controlled oscillators. It delivers 21.7 pJ per pulse and eliminates the need for analog-to-digital converters.
FORCE learning improves real-time performance for complex tasks, with benchmark results on the NARMA10 problem.