Samsung used Hot Chips 2026 to detail what Tom's Hardware describes as the industry's first LPDDR5X-PIM, a memory chip with a logic unit built directly into it.
The idea behind processing-in-memory, or PIM, is simple even if the engineering isn't. In a normal device, data sits in memory and has to be shuttled back and forth to a processor every time something needs to be computed. For AI workloads, that shuttling is often the bottleneck — the processor spends much of its time waiting on data rather than doing math. PIM flips that around by doing some of the work where the data already lives.
According to Tom's Hardware, Samsung's LPDDR5X-PIM delivers 3.01x faster AI inference than standard LPDDR5X, with 8x the bandwidth. Samsung positioned the technology as a way to speed up data-intensive workloads, with AI inference — the stage where a trained model actually answers your question or generates an image — as the headline use case.
The "LP" in LPDDR5X stands for low power, and that's the part worth paying attention to. This is the memory family used in phones, laptops and other battery-powered devices, not the exotic high-bandwidth memory stacked next to data center accelerators. Samsung is aiming its in-memory logic at the mobile and edge tier.
Samsung presented the design at Hot Chips, the annual conference where chipmakers walk through architectural details of new silicon. The company has not, in the material reported here, announced shipping products or timelines.
Why it matters: if AI inference can run meaningfully faster inside low-power memory, more AI could run on the device in your pocket instead of in a distant data center.