Micron EIM (Edge Inference Memory) is the new edge inference memory Micron is developing for the emerging physical AI market. This technology stacks low-power LPDDR memory on a logic die to reduce latency and increase bandwidth in applications such as humanoid robots, autonomous vehicles and augmented reality devices.
The information came from statements by Micron vice president Lee Cheol during the “Semiconductors Powering Humanoids” forum at the Semiconductor Exhibition (SEDEX 2026), held on October 16, 2026 at COEX in Seoul. The original report was published by The Elec and covered by Chinese outlets such as IT Home.
What happened?
Lee Cheol stated that physical AI markets — including autonomous driving, humanoid robots, AR and VR — are growing rapidly, and Micron is discussing suitable products with key customers from the early design stages. Micron EIM differs from Processing-In-Memory (PIM): while PIM integrates compute units inside the memory, the specific computing architecture of EIM has not yet been publicly disclosed.
The focus of Micron EIM is to reduce data-processing latency and raise bandwidth between the AI processor and memory. The company has not revealed numerical performance targets, but is working on integrating the technology with LPDDR6. Industry observers estimate that mass production of Micron EIM will occur in the second half of 2028.
Why does this matter?
Humanoid robots and physical AI systems require high bandwidth and low latency to process vision data, tactile sensors and make real-time decisions. Traditional HBM memories offer high performance but are expensive, power-hungry and hard to integrate into compact, battery-powered edge devices. Micron EIM aims to combine LPDDR’s energy efficiency with bandwidth characteristics closer to those of HBM.
According to Micron’s own projections, high-performance robots in the future will need about 200 GB of DRAM and 4 TB of storage. By comparison, current service robots typically use only 12 GB of DRAM and 256 GB of storage. This gap illustrates the growing demand for memory solutions optimized for the edge.
What changes in practice?
If Micron EIM reaches the market in 2028, it could accelerate the adoption of larger and more responsive AI models in robots, autonomous cars and AR/VR glasses, with lower power consumption. Robot manufacturers, chipset suppliers and hyperscalers are already collaborating with Micron in the early development stages. Although concrete performance numbers are not yet available, the initiative demonstrates how the memory industry is positioning itself for the next wave of AI outside data centers.
For more details, see the original report from The Elec and the coverage by IT Home.
Image credit: Wikimedia Commons — Source: Thingreenline4546 (CC BY-SA 4.0)
By GeekikiBot
