Science paper turns memristor drift into a 2.12 ms neural dynamical system
jiqizhixin · x · 2026-07-23
Science paper shows a memristor chip running neural dynamical systems in 2.12 ms
A new Science paper turns a flaw in phase-change memristors into a feature: the devices’ natural conductance drift is used to compute neural dynamical systems directly in memory.
Reported results
- Custom 40 nm chip
- 2.12 ms latency
- Up to 36× faster and 24× more power efficient than prior NDS hardware
- Up to 478× faster than an A100 GPU on the reported setup
Why it matters
The paper suggests a path toward sub-10-millisecond neural dynamical systems by moving computation into the device physics itself rather than treating drift as a bug.
More from Infra
- AMD lines up 2GW of MI450 GPUs for Anthropic and may invest up to $5B — Beth_Kindig · 2026-07-23
- Signal65 says AMD MI355X delivered up to 2.15x better tokens per dollar than NVIDIA B200 — ryanshrout · 2026-07-23
- AI job listings mentioning evals rise to 10.7% by July 2026 — HamelHusain · 2026-07-23
- Writer Study: Optimizing AI Harness Reduces Costs by 41% Without Losing Accuracy — bendee983 · 2026-07-23
- Voice assistant tool calls sped up instantly after moving the backend to Europe — ur_piyo_a_hoe · 2026-07-23
- Meta overhauls blob storage to cut GPU stalls across exabyte-scale clusters — Meta_Engineers · 2026-07-23