Micron Develops New Memory for Physical AI Applications

Micron Technology is advancing the development of its Edge Inference Memory (EIM), a specialized memory chip for physical AI. This new memory aims to accelerate data processing and reduce latency in applications such as autonomous vehicles and robotics.

According to company representatives, EIM is designed for high-speed processing of data required for AI inference. Its architecture involves vertically stacking low-power LPDDR memory components on top of logic chips. While distinct from Processing-in-Memory (PIM) which integrates computation within the memory, EIM also seeks to enhance data processing efficiency near the main AI processor.

The EIM is positioned between the AI processor and memory, intended to lower data processing latency and improve system bandwidth. Micron is working on integrating this technology into future LPDDR6 solutions, with industry estimates suggesting mass production could begin in the latter half of 2028.

Physical AI applications, particularly humanoid robots, require rapid data processing for environmental sensing and autonomous decision-making. As AI models grow in complexity, traditional LPDDR memory may become insufficient, pushing the industry towards solutions like EIM and potentially High Bandwidth Memory (HBM) in the future. Micron anticipates significant increases in memory and storage needs for advanced robotics.