Micron Technology is developing an edge inference memory product, or EIM, for physical AI, according to Li Zhe, a vice president at the company, speaking at the SEDEX 2026 semiconductor exhibition held at COEX in Seoul.
The report, citing industry expectations carried by Kechuangban Daily, said the product is expected to enter mass production in the second half of 2028. Micron is studying the use of EIM with LPDDR6 to serve demand for low-power DRAM in applications such as autonomous driving and humanoid robots, and is in discussions with major customers about related products.
EIM is aimed at edge inference workloads
According to the report, EIM sits between AI chips and memory. The approach involves stacking LPDDR on a logic chip to process inference data at high speed. The full computing architecture has not been fully disclosed.
Li said current robot AI models are around 1 billion to 2 billion parameters, and their bandwidth requirements are not high. As more advanced vision-language models emerge, LPDDR alone will not be enough, and HBM will eventually be used in humanoid robots.
Micron shared its view on robot memory requirements
Micron estimates that future high-performance humanoid robots may need about 200 GB of DRAM and 4 TB of storage. Current service robots, by comparison, use about 12 GB of DRAM and 256 GB of storage.
Micron CEO Sanjay Mehrotra said the storage capacity of a single humanoid robot is about 10 times that of a typical L2+ passenger vehicle. He said the memory industry could enter a demand upcycle lasting decades around 2030, driven by the need to process data from hundreds of sensors at the same time and run large models locally for millisecond-level control.
Counterpoint projects humanoid robot DRAM demand at 172,000 Tbit by 2030
Counterpoint Research expects global humanoid robot DRAM demand to increase from about 8,300 Tbit this year to about 172,000 Tbit by 2030.
