惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

阮一峰的网络日志
阮一峰的网络日志
The GitHub Blog
The GitHub Blog
酷 壳 – CoolShell
酷 壳 – CoolShell
雷峰网
雷峰网
U
Unit 42
Y
Y Combinator Blog
I
InfoQ
P
Proofpoint News Feed
Engineering at Meta
Engineering at Meta
量子位
Microsoft Security Blog
Microsoft Security Blog
B
Blog
The Cloudflare Blog
F
Fortinet All Blogs
Google DeepMind News
Google DeepMind News
MyScale Blog
MyScale Blog
C
Check Point Blog
S
SegmentFault 最新的问题
爱范儿
爱范儿
博客园 - 叶小钗
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
Hugging Face - Blog
Hugging Face - Blog
罗磊的独立博客
T
Tailwind CSS Blog

... eeNews Europe

Molex Teramount deal targets co-packaged optics NVIDIA and ServiceNow extend AI governance from desktops to data centres Faraday Future launches Physical AI robotics institute with BIBS System Check: Should engineers learn analog? Decoupled by Design: How Gateworks and NXP are rethinking edge AI architecture NVIDIA and Corning partner on AI photonics expansion GCT taps satellite partner to speed 5G rollout Sodankylä supersite to support ESA Earth observation SiTime posts 88% revenue growth on AI infrastructure demand Infrared LEDs support in-cabin sensing for vehicle safety Anthropic compute deal taps SpaceXAI Colossus 1 Quantum Brilliance CEO Mark Luo on deployable quantum systems and the future of diamond-based computing SEMI Summit spotlights Europe’s chiplet and packaging push AI data center infrastructure drives Pennsylvania energy expansion NXP CoreRide gains Vector software support for SDV platforms Elektor Lab Talk covers Red Pitaya and reconfigurable test gear SEMI names Julie Rogers to lead ESD Alliance ASML CEO backs joint call for Europe tech competitiveness push FlexIC RFID inlays bring NFC to paper packaging ROHM targets smart rings with ultra-compact NFC wireless power chipset China silicon wafers push boosts Eswin capacity ESD Alliance outlook spotlights agentic AI in chip design AI robotics sales growth rises as Faraday Future expands into education Microchip expands dsPIC33A controllers for AI data center power sensiBel MEMS microphone heads to Silex production SEMI: Global silicon wafer shipments jump 13% on AI demand AI drives photonics innovation Advantech adds Intel Core Series 3 to edge AI systems NI CHESS enables software-driven RF channel emulation into aerospace testing Forsee Power battery system powers new electric fire pump
Spintronic memory switches in 40 ps
Brian Tristam Williams · 2026-05-21 · via ... eeNews Europe

Spintronic memory switches in 40 ps

News |

By Brian Tristam Williams



A University of Tokyo-led research team has demonstrated a non-volatile spintronic memory switching device that can rewrite a magnetic state in 40 ps, using an antiferromagnetic manganese-tin structure rather than stored electrical charge.

The work, announced by the University of Tokyo School of Science and published in Science, is aimed at one of the central limits in AI servers and high-speed computing: switching and moving data without converting most of the input energy into heat.

Spintronic memory uses magnetic state switching

The device uses the antiferromagnetic material Mn3Sn with tantalum. Instead of storing data as capacitor charge, as in DRAM, it uses magnetic states to represent binary values. Short electrical pulses generate spin-orbit torque, transferring angular momentum into the magnetic structure and changing its state.

The researchers report switching with 40 ps electrical pulses. They also demonstrated switching using 60 ps photocurrent pulses produced by combining a telecom-wavelength laser with a photoelectric converter. That second result is important because it links optical signalling to direct writing of a non-volatile state.

For context, a 5 GHz processor cycle lasts 200 ps. A memory-switching process that takes 1 ns spans several such cycles, while a 40 ps event fits well inside one. That does not make the device a commercial memory chip, but it shows why picosecond switching is attractive for future processor and data-centre architectures.

Spintronic memory still has practical hurdles

The claim is not that DRAM has been replaced. This is still a laboratory switching device, not an addressable memory array ready for manufacturing. Density, retention, endurance in real systems, CMOS integration, bias-field requirements and cost all remain open engineering questions.

Even so, the result adds weight to the case for magnetic and antiferromagnetic approaches to future memory. As we reported previously when researchers in Japan developed low power spintronic ME-MRAM memory, magnetic memories are being pursued as a way to reduce standby power and avoid the continual refresh overhead of charge-based memory.

The latest spintronic memory demonstration pushes that discussion into the picosecond regime. Its near-term value is as a materials and device result, particularly for optical-to-electrical conversion and low-heat switching, rather than as a near-term drop-in replacement for DRAM, SRAM or NAND.

If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :

   eeNews on Google News