Micron EIM memory is the company's new bet for the physical AI market. Micron is developing Edge Inference Memory (EIM), a memory technology designed for physical artificial intelligence applications such as humanoid robots, autonomous vehicles, augmented and virtual reality.
What happened?
The announcement came from Lee Cheol, Micron vice president, during the “Semiconductors Powering Humanoids” forum at SEDEX 2026 in Seoul on October 16. EIM stacks low-power DRAM (LPDDR) on a logic die, positioning itself between the AI processor and conventional memory to reduce data processing latency and increase bandwidth.
Unlike Processing-In-Memory (PIM), the exact computational architecture of EIM has not yet been disclosed. Micron is exploring its integration with LPDDR6, and industry observers expect mass production in the second half of 2028.
Why does this matter?
Humanoid robots need energy efficiency because they are battery-powered, unlike data centers that use HBM. Current robot models use about 1 to 2 billion parameters, but more advanced vision-language models will require more bandwidth. Micron predicts that high-performance robots will need about 200 GB of DRAM and 4 TB of storage, versus 12 GB and 256 GB in current service robots.
The company is already collaborating with robot manufacturers, chipset companies and hyperscalers from the early design stages.
What changes in practice?
If EIM delivers on its promise, robots could operate longer with faster decisions and lower energy consumption. This accelerates the adoption of physical AI in factories, logistics and assistance. However, specific performance targets have not yet been confirmed by Micron.
Sources: The Elec, IT Home and Wccftech.
Image credit: Micron / The Elec
By GeekikiBot
