Micron Technology is advancing the development of Micron EIM, or Edge Inference Memory, a specialized memory for physical AI applications including humanoid robots, autonomous vehicles, and AR/VR devices.
What happened?
At the Semiconductors Powering Humanoids forum at SEDEX 2026 in Seoul on October 16, 2026, Micron vice president Lee Cheol highlighted the rapid growth of the physical AI market. The company collaborates with key customers on Micron EIM, which stacks low-power LPDDR DRAM on a logic die. This setup sits between the AI processor and standard memory to cut latency and boost bandwidth.
Unlike PIM (Processing-In-Memory), which places compute units inside memory, the compute architecture of Micron EIM remains confidential. Micron is investigating integration with LPDDR6, and experts forecast mass production in the second half of 2028, according to reports from The Elec and IT Home.
Why is this important?
Physical AI systems, especially robots, need low-latency inference and high efficiency to run for hours on battery. Micron estimates high-performance robots will require about 200 GB of DRAM and 4 TB of storage, far above the 12 GB and 256 GB of current service robots. Micron EIM aims to combine LPDDR efficiency with HBM-like high bandwidth without the latter's high power draw.
What changes in practice?
Success of Micron EIM could boost robotics and edge AI, enabling larger models and snappier responses in mobile devices. The 2028 target shows development is ongoing, with early partnerships already with robot makers, chip suppliers, and hyperscalers. Sources: The Elec, IT Home, and SEDEX 2026 reports.
By GeekikiBot
