YOU ARE VIEWING ONE ITEM FROM THE AICRIER FEED

Bi2Se3 memristor powers analog neural network

AICrier tracks AI developer news across Product Hunt, GitHub, Hacker News, YouTube, X, arXiv, and more. This page keeps the article you opened front and center while giving you a path into the live feed.

// WHAT AICRIER DOES

7+

TRACKED FEEDS

24/7

SCRAPED FEED

Short summaries, external links, screenshots, relevance scoring, tags, and featured picks for AI builders.

Bi2Se3 memristor powers analog neural network
OPEN LINK ↗
// 76d agoRESEARCH PAPER

Bi2Se3 memristor powers analog neural network

University of Michigan researchers published an ACS Nano paper on a Bi2Se3 memristor built from 2D bismuth selenide layers that combines long-term retention, analog tuning, and regulator-free operation. In a fully analog reservoir-computing demo, the device controlled a balance lever using about 7 microwatts, pointing to lower-power neuromorphic hardware.

// ANALYSIS

This is the rare memristor paper that feels like a systems advance, not just a materials curiosity. The useful part is that it closes a real feedback-loop problem without forcing the circuit back through digital conversion.

  • Gold filaments grow and retract without bridging the gap, giving smooth analog modulation instead of abrupt switching.
  • The paper reports 10-40% conductance tuning, less than 1% retention loss over 10,000 seconds, about 7 microwatts controller power, and NRMSE 0.094.
  • The gold-assisted, plasma-etching-free PVD crossbar is the commercialization clue: it looks much closer to a manufacturable process than a one-off lab artifact.
  • For developers, the promise is edge control loops where sensing, memory, and actuation stay analog, though array-scale uniformity and endurance still need proof.
// TAGS
bi2se3-memristoredge-aiinferencerobotics

DISCOVERED

76d ago

2026-03-26

PUBLISHED

76d ago

2026-03-25

RELEVANCE

8/ 10

AUTHOR

jferments