According to South Korean media reports, Samsung Electronics plans to launch a dedicated solution for AI inference this August, integrating Processing-in-Memory (PIM) technology onto LPDDR5X memory.
In May this year, Samsung published research outcomes of its LP5X-PIM simulator in the form of a technical paper. The market widely expects Samsung to take another step forward in August by releasing measured performance metrics including bandwidth improvement and energy efficiency, showcasing physical hardware demo samples and disclosing its commercialization roadmap.
PIM technology addresses the performance bottleneck caused by data transfer between CPUs/GPUs and memory. Its core principle is embedding computing units (PIM logic circuits) inside memory chips to process data directly within memory, drastically cutting the volume of data movement.
PIM technology has long been adopted in products such as High Bandwidth Memory (HBM). However, Samsung will be the first company worldwide to integrate PIM technology into power-sensitive LPDDR memory for real-world AI inference deployments. This marks the technology’s transition out of the lab, with the goal of equipping end devices such as smartphones for on-device AI workloads at the earliest opportunity.
Simulation results of LPDDR5X-PIM show that when running AI matrix operations (GEMV), the solution delivers up to 6.2 times the performance of conventional architectures, significantly accelerating AI tasks including image recognition and natural language processing.
Samsung Electronics is pursuing a two-pronged strategy: leveraging HBM to target the server market, while tapping into consumer end markets with LPDDR5X-PIM. Unlike HBM, LPDDR5X-PIM targets power-sensitive devices such as smartphones and laptops. Against the backdrop of rising demand for privacy protection and lower network latency, the importance of on-device AI continues to grow.
When paired with GAIA, Samsung’s in-house AI accelerator built on a 4nm process node, the complete solution achieves optimal energy efficiency. GAIA, centered on an NPU, handles primary AI computations, while LPDDR5X-PIM delivers efficient collaborative support at the memory layer.
Besides Samsung, SK Hynix is also advancing development of PIM technology of a similar architecture. It has launched its proprietary PIM memory accelerator AiM, developed the GDDR6-AiM solution integrating this technology, and unveiled the AiMX AI accelerator equipped with multiple such chips.
The rollout of key technologies enabling real-time AI computing on AI PCs and flagship smartphones will provide strong momentum for the rapid growth of the on-device AI market.
All Rights Reserved © 2026 MemoryMarket.com
