SoftBank, Tokyo Uni Partner for Energy-Saving AI Memory Chips

• SoftBank, Tokyo Uni develop energy-efficient AI memory chips with Intel tech.
• Sai Memory targets a low-power, cost-effective HBM alternative for data centers.
• Japan invests 10 trillion yen to revive the semiconductor industry by 2030.

Softbank, in collaboration with the University of Tokyo, the National Institute of Chemical Research (Riken), and Shinco Electric Industries, is spearheading a groundbreaking initiative to develop next-generation semiconductors through a newly established company, Sai Memory. This venture aims to address global challenges in high-bandwidth memory (HBM) by creating energy-efficient, cost-competitive alternatives for AI data centers.

In the next two years, Sai Memory will use patents obtained from the University of Tokyo and Intel’s technology to test and prototype advanced semiconductors, which are then intended to be mass-produced as planned. The project has Cy Memory, a subsidiary in charge of handling intellectual property matters. Cy Memory plans to reduce power consumption by changing the wiring linking DRAMs on a substrate in contrast to existing HBMs.

Softbank is planning to invest $30.7 million in Cy Memory and become the firm’s biggest investor, and is also working to obtain funds from the Japanese government to boost the project. This initiative is in line with Japan’s intentions to build up its semiconductor sector, with the government promising to allocate more than 10 trillion yen in 10 years for its growth.

The new memory developed by Cy Memory targets data centers, offering a competitive edge over HBM in both price and power efficiency. While HBM excels in data storage and transmission speed, its high costs and power demands, coupled with supply shortages, have created market gaps. SK Hynix, the leading HBM supplier, has reported that its entire stock for the next two years is already sold out.

Also Read: Vietnam Seeks Samsung AI Chip Design Partnership for Data, Telecom

It is evident from industry predictions that innovation is needed urgently. According to BCG, AI-related server shipments are projected to surge six-fold from 2023 to 2027, with DRAM shipments expected to grow 21% annually. Japan’s push to develop new memory solutions is partly driven by the decline of its domestic DRAM industry since the 1980s, with Kioxia, formerly part of Toshiba, now focusing solely on NAND flash memory.

This collaborative effort signals Japan’s determination to reclaim a leadership role in the global semiconductor market, addressing critical needs in AI infrastructure with innovative, sustainable technology.