As per a report from the Korean media outlet The Elec, Micron is teaming up with major customers to create a novel memory product specifically designed for the realm of physical AI—Edge Inference Memory (EIM). This innovative solution is set to be primarily utilized in areas like humanoid robots, autonomous vehicles, and AR/VR devices. At an event held during SEDEX 2026, a prominent Korean semiconductor exhibition, Micron executive Li Zhe highlighted the booming nature of the physical AI market. He emphasized that the company is actively collaborating with customers to drive forward product development that caters to this burgeoning sector. By stacking low-power LPDDR on top of a logic chip, EIM memory facilitates high-speed processing of AI inference data directly within the memory unit. This approach minimizes data processing latency and boosts system bandwidth. While the specific computational architecture of EIM remains under wraps, unlike Processing-in-Memory (PIM), its design objective is clear: to offer ample memory bandwidth to satisfy the requirements of physical AI devices, all while keeping power consumption in check. Presently, humanoid robots predominantly rely on LPDDR memory. However, with the expansion of AI model parameter scales, the need for increased memory bandwidth and capacity is on the rise. Micron forecasts that future high-performance humanoid robots may necessitate around 200GB of DRAM and a hefty 4TB of storage space. In light of this, Micron is delving into a technical solution that integrates EIM with LPDDR6. Industry insiders anticipate that mass production of this solution could commence in the latter half of 2028.