{"id":26010,"date":"2026-05-18T09:39:23","date_gmt":"2026-05-18T13:39:23","guid":{"rendered":"https:\/\/blog.tmcnet.com\/blog\/rich-tehrani\/?p=26010"},"modified":"2026-05-18T09:40:46","modified_gmt":"2026-05-18T13:40:46","slug":"university-of-tokyo-research-points-to-faster-lower-power-switching-for-future-ai-infrastructure","status":"publish","type":"post","link":"https:\/\/blog.tmcnet.com\/blog\/rich-tehrani\/networking\/university-of-tokyo-research-points-to-faster-lower-power-switching-for-future-ai-infrastructure.html","title":{"rendered":"University of Tokyo Research Points to Faster, Lower Power Switching for Future AI Infrastructure"},"content":{"rendered":"\n<p><strong>Key Takeaways:<\/strong><\/p>\n\n\n\n<ul>\n<li>University of Tokyo researchers demonstrated nonvolatile switching in an antiferromagnetic device using 40 picosecond electrical pulses.<\/li>\n\n\n\n<li>The work focuses on a major computing bottleneck: the energy and speed cost of moving data between processors and memory.<\/li>\n\n\n\n<li>The device used Mn3Sn, a topological antiferromagnet, and tantalum in a thin film structure.<\/li>\n\n\n\n<li>The research also demonstrated switching using 60 picosecond photocurrent pulses generated from a communications wavelength laser and a high speed photodetector.<\/li>\n\n\n\n<li>The findings are promising, but still in the research stage. Further work is needed before this could influence commercial data center systems.<\/li>\n<\/ul>\n\n\n\n<p>As AI workloads grow, the conversation around computing infrastructure usually starts with GPUs, networking gear, power availability, cooling, and the cost of building new data centers. Fair enough. Those are the visible pressure points. But deeper inside the system, another problem keeps getting more important: moving data is expensive.<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"alignleft is-resized\"><img decoding=\"async\" src=\"https:\/\/www.t.u-tokyo.ac.jp\/hs-fs\/hubfs\/press-release\/2026\/0515\/001\/fig.jpg?width=700&amp;height=220&amp;name=fig.jpg\" alt=\"fig\" style=\"width:468px;height:auto\"\/><figcaption class=\"wp-element-caption\"><strong>Schematic diagram of non-volatile quantum switching combining a photoelectric conversion element and an antiferromagnetic switching element.<\/strong><\/figcaption><\/figure><\/div>\n\n\n<p>Not expensive in the everyday sense of a line item on an invoice, though it eventually becomes that. Expensive in energy, latency, heat, and system design complexity. Modern computing architectures still separate processing and memory. That means data has to travel back and forth, again and again, between where it is stored and where it is processed. As AI models and large scale machine learning systems demand more data movement, this so called memory wall has become a more serious constraint on performance and power efficiency. The University of Tokyo release frames this issue as one of the reasons researchers are exploring in memory and near memory computing, along with optical interconnects and other approaches that bring computation and data movement closer together.<\/p>\n\n\n\n<p>A University of Tokyo led research group has now <a href=\"https:\/\/www-t-u--tokyo-ac-jp.translate.goog\/press\/pr2026-05-15-001?_x_tr_sl=es&amp;_x_tr_tl=en&amp;_x_tr_hl=en&amp;_x_tr_pto=wapp&amp;_x_tr_hist=true\">reported<\/a> a notable step in that direction: a nonvolatile quantum switching device that operates at extremely high speed while using very low power. The team demonstrated switching in a device based on the antiferromagnetic material Mn3Sn using electrical pulses as short as 40 picoseconds. A picosecond is one trillionth of a second, so this is a time scale far below the nanosecond range typically associated with high speed electronic switching.<\/p>\n\n\n\n<p>The key detail is not just speed. It is speed combined with nonvolatility and low power operation. Nonvolatile means the device can retain its state even when power is turned off. In practical computing terms, that matters because memory or switching elements that do not need constant power to hold information could help reduce wasted energy. The University of Tokyo team says the device rewrites a magnetic binary state using very short pulses, while avoiding the large heat buildup that has made picosecond scale switching difficult to move toward practical use.<\/p>\n\n\n\n<p>The researchers used a thin film device made from Mn3Sn and tantalum. Mn3Sn is described in the release as a topological antiferromagnet, a class of material that can show useful magnetic and electronic behavior even though its overall magnetization is close to zero. That sounds counterintuitive if you are used to thinking about magnets as things with a strong external magnetic field. But that is part of why antiferromagnets are interesting for future spintronics. They can offer very fast spin dynamics and may avoid some limitations associated with conventional ferromagnetic devices.<\/p>\n\n\n\n<p>Here is the simpler way to think about it. Conventional electronic systems move charge. Spintronic systems try to use another property of electrons, their spin. In this work, the switching mechanism is tied to spin orbit torque, where current is converted into spin flow that can influence the magnetic order of a neighboring material. The release says the 40 picosecond switching behavior was consistent with an intrinsic spin torque mechanism rather than heat driven switching. That distinction matters because heat is often the enemy in high speed electronics. If switching only works by dumping a lot of energy into a tiny region, the device may be fast in a lab but difficult to scale into durable hardware.<\/p>\n\n\n\n<p>The reported endurance is also important. The team demonstrated more than 10 to the 11th switching operations using picosecond pulses. It also estimated an energy density of about 1.7 picojoules per square micrometer and a power density of about 0.04 watts per square micrometer for a device with a 10 nanometer Mn3Sn layer and 5 nanometer tantalum layer. When scaled to a 30 by 30 by 10 nanometer one bit element, the release says the energy consumption would be on the order of 1 femtojoule.<\/p>\n\n\n\n<p>That does not mean a commercial chip is around the corner. It does mean the research addresses several problems at once: speed, energy, durability, and nonvolatile behavior. In hardware research, solving one of those problems is useful. Showing progress on several of them together is more meaningful.<\/p>\n\n\n\n<p>Another interesting part of the work is the optical connection. The team used a communications wavelength laser and a high speed photodetector to generate 60 picosecond photocurrent pulses, then used those pulses to switch the antiferromagnetic state. In one experiment, the release says the team observed error free switching over 250 repetitions. The significance is that optical signals could potentially be converted into electrical signals and connected directly to nonvolatile memory writing. That makes the work relevant not only to memory devices, but also to future data center I\/O, where optical links are already central to high speed communication between systems.<\/p>\n\n\n\n<p>This is where the research becomes easier to connect to the broader AI infrastructure discussion. Data centers are not just running out of compute. They are running into limits around power delivery, thermal management, and data movement. Faster accelerators help, but if memory access and interconnect efficiency do not improve, some of the system level gains can be lost. A switching device that is ultrafast, low power, and compatible with optoelectronic interfaces could eventually support new <a href=\"https:\/\/gizmodo.com\/new-quantum-processing-technology-points-to-life-after-the-transistor-maybe-2000759222\">architectures<\/a> that reduce some of this friction.<\/p>\n\n\n\n<p>Still, the word \u201ceventually\u201d is doing real work here. The University of Tokyo release says future work includes achieving picosecond class operation without an external magnetic field and optimizing the device for circuit and implementation conditions. Those are not small details. Materials research has a long road from proof of concept to manufacturable technology, especially when it touches memory, logic, packaging, and system architecture at the same time.<\/p>\n\n\n\n<p>The research was published in Science under the title \u201cPicosecond ultralow-power switching device based on an antiferromagnet.\u201d The author group includes researchers from the University of Tokyo and RIKEN, with the release listing Hanshen Tsai and Takuya Matsuda as equal contributors and Satoru Nakatsuji as corresponding author.<\/p>\n\n\n\n<p>What makes the work worth watching is not that it immediately changes the economics of AI data centers. It does not. The more grounded takeaway is that researchers are attacking the data movement problem at the material and device level, not just at the software, chip, or system design layer. If AI keeps pushing infrastructure toward higher performance and higher power consumption, those lower level breakthroughs may become increasingly important.<\/p>\n\n\n\n<p>There is a quiet lesson here. The future of computing may not depend only on larger clusters, better cooling, or faster GPUs. It may also depend on tiny switching events happening in trillionths of a second, using less energy, creating less heat, and holding state without constant power. That is still a research story today. But it is the kind of research story that points toward where the next generation of computing infrastructure may need to go.<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"alignright\"><a href=\"https:\/\/blog.tmcnet.com\/blog\/rich-tehrani\/wp-content\/uploads\/2025\/05\/image-10.png\"><img loading=\"lazy\" decoding=\"async\" width=\"299\" height=\"136\" src=\"https:\/\/blog.tmcnet.com\/blog\/rich-tehrani\/wp-content\/uploads\/2025\/05\/image-10.png\" alt=\"\" class=\"wp-image-20657\"\/><\/a><\/figure><\/div>\n\n\n<p>If you liked this post, you\u2019ll love one of the the leading global business communications and technology events since 1999, the&nbsp;<a href=\"http:\/\/www.itexpo.com\/\">ITEXPO #TECHSUPERSHOW<\/a>, Feb 9-11, 2027 Fort Lauderdale, Florida.<\/p>\n\n\n\n<p>Don\u2019t forget the collocated&nbsp;<a href=\"http:\/\/www.mspexpo.com\/\">MSP Expo<\/a>&nbsp;\u2013 just for managed service providers!<\/p>\n\n\n\n<p><em>Aside from his role as CEO of&nbsp;<a href=\"http:\/\/www.tmcnet.com\/\">TMC<\/a>&nbsp;and chairman of&nbsp;<a href=\"http:\/\/www.itexpo.com\/\">ITEXPO<\/a>&nbsp;#TECHSUPERSHOW Feb 9-11, 2027,&nbsp;Rich Tehrani is CEO of&nbsp;<a href=\"https:\/\/www.rt-advisors.com\/\">RT Advisors<\/a>&nbsp;and a Registered Representative (investment banker) with and offering securities through&nbsp;<a href=\"https:\/\/www.4pointscapital.com\/\">Four Points Capital Partners LLC&nbsp;<\/a>(Four Points) (Member FINRA\/SIPC). He handles capital\/debt raises as well as M&amp;A. RT Advisors is not owned by Four Points.<\/em><\/p>\n\n\n\n<p>The above is not an endorsement or recommendation to buy\/sell any security or sector mentioned. No companies mentioned above are current or past clients of RT Advisors.<\/p>\n\n\n\n<p>The views and opinions expressed above are those of the participants. While believed to be reliable, the information has not been independently verified for accuracy. Any broad, general statements made herein are provided for context only and should not be construed as exhaustive or universally applicable.<\/p>\n\n\n\n<p><em>Portions of this article may have been developed with the assistance of artificial intelligence, which may have contributed to ideation, content generation, factual review, or editing<\/em><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Key Takeaways: As AI workloads grow, the conversation around computing infrastructure usually starts with GPUs, networking gear, power availability, cooling, and the cost of building new data centers. Fair enough. Those are the visible pressure points. But deeper inside the system, another problem keeps getting more important: moving data is expensive. Not expensive in the<\/p>\n","protected":false},"author":44,"featured_media":26006,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[164],"tags":[],"post_mailing_queue_ids":[],"_links":{"self":[{"href":"https:\/\/blog.tmcnet.com\/blog\/rich-tehrani\/wp-json\/wp\/v2\/posts\/26010"}],"collection":[{"href":"https:\/\/blog.tmcnet.com\/blog\/rich-tehrani\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/blog.tmcnet.com\/blog\/rich-tehrani\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/blog.tmcnet.com\/blog\/rich-tehrani\/wp-json\/wp\/v2\/users\/44"}],"replies":[{"embeddable":true,"href":"https:\/\/blog.tmcnet.com\/blog\/rich-tehrani\/wp-json\/wp\/v2\/comments?post=26010"}],"version-history":[{"count":2,"href":"https:\/\/blog.tmcnet.com\/blog\/rich-tehrani\/wp-json\/wp\/v2\/posts\/26010\/revisions"}],"predecessor-version":[{"id":26012,"href":"https:\/\/blog.tmcnet.com\/blog\/rich-tehrani\/wp-json\/wp\/v2\/posts\/26010\/revisions\/26012"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/blog.tmcnet.com\/blog\/rich-tehrani\/wp-json\/wp\/v2\/media\/26006"}],"wp:attachment":[{"href":"https:\/\/blog.tmcnet.com\/blog\/rich-tehrani\/wp-json\/wp\/v2\/media?parent=26010"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blog.tmcnet.com\/blog\/rich-tehrani\/wp-json\/wp\/v2\/categories?post=26010"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blog.tmcnet.com\/blog\/rich-tehrani\/wp-json\/wp\/v2\/tags?post=26010"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}