Intel is developing an alternative to HBM memory with the University of Tokyo and Softbank

The fashion Keep pulling towards Integrated artificial intelligence devicesdevices of all kinds that usually resort to NPU solutions or even to Data processing in the cloud to be able to implement ia. This is raising (even more so if it fits) Accelerators demand To be able to offer faster response, for what is needed greater amount of HBM memory which are usually included in these. To offer Better and faster solutions Intel (along with some partners) is preparing An alternative To this memory of great bandwidth.

A new one Startup call Saimemory composed of Intel, the University of Tokyo and Softbank. Intel will contribute his experience as a manufacturer of semiconductors, which he will use in the patents of the University of Tokyo, while SoftBank has joined as a capitalist partner contributing 21 million dollars. In addition, other partners such as the Riken and Shinko Electric Industries Research Institute want to join either to contribute capital or experience. From this new startup they are also Looking for government support that will allow you to cut this long project in time.

The idea for this alternative to HBM is to use a reorganized signal routing system that will allow Improve energy efficiency and latency which will lead to a higher performance. A DRAM -based solution that is committed to A risky alternative outside the standards current, but has the support of Intel that can Implement this memory in your solutions. A project that the first are expected prototypes for 2027 With one mass production for within at least 5 years.