Samsung LPDDR6: Advancing Low-Power Memory for AI with Industry-First Validation on Qualcomm Technologies' New Snapdragon® Platform
Across cloud and edge environments, AI workloads are becoming more complex, persistent and data-intensive. As a result, energy efficiency is becoming as critical as computing performance, with AI systems expected to deliver more capability within increasingly constrained power and thermal budgets.
Against this backdrop, LPDDR is expanding beyond its traditional role in mobile devices. Samsung is extending LPDDR technology across a broader range of AI computing environments, from LPDDR-PIM, which integrates processing capabilities directly into memory, to SOCAMM2, an LPDDR-based modular solution for AI infrastructure.
Meanwhile, mobile computing is also evolving rapidly. On-device AI is moving beyond standalone features toward multimodal interactions and AI agents that can maintain context across longer interactions. These experiences place new demands on how efficiently mobile systems move and manage data within tight power constraints.
Samsung LPDDR6 was developed with these shifts in mind, combining advances in performance, power efficiency and reliability to support increasingly sophisticated mobile AI experiences.
Keeping data moving for AI workloads
During LLM inference, the system repeatedly accesses large model weights and KV Cache data as it generates responses. As models grow and interactions become longer, efficient data movement between memory and the processor becomes increasingly important.
Samsung LPDDR6 addresses this challenge with enhanced per-pin I/O speed and an expanded I/O architecture, delivering up to 114GB/s of memory bandwidth. This helps alleviate data-movement bottlenecks and allows the system to access the data needed for AI processing more efficiently, supporting more responsive on-device AI experiences.
But higher bandwidth alone is not enough. On-device AI workloads can shift rapidly between intensive and lighter processing, making efficient power management and reliable operation equally important.
Balancing performance, power and reliability
Samsung LPDDR6 incorporates Dynamic Voltage and Frequency Scaling (DVFS) and Dynamic Efficiency Mode (DEM) to optimize power consumption in real time as workload demands shift. DVFS dynamically adjusts operating voltage and frequency based on performance needs, while DEM reduces unnecessary power use during lower-utilization periods. Together, these technologies improve power efficiency by up to 21% compared with the previous generation, LPDDR5X.
This adaptive approach allows LPDDR6 to align power use more closely with workload intensity, helping maintain performance without drawing peak power unnecessarily.
As memory activity intensifies, reliability becomes increasingly important. LPDDR6 incorporates Per-Row Activation Counting (PRAC), which monitors repeated activations of specific memory rows and helps manage potential data interference associated with these activation patterns. This added protection supports reliable memory operation during complex multimodal processing and prolonged on-device AI agent activity.
Recommended by LinkedIn
Industry-first validation on Qualcomm Technologies' Snapdragon® platform
Bringing a new memory technology to a mobile platform requires more than component-level advances. Memory must be optimized and validated together with the SoC and system architecture to ensure that it can meet platform-level requirements for performance, power efficiency and reliability.
Samsung has successfully validated its 16GB LPDDR6, operating at I/O speeds of up to 10.7Gbps, on Qualcomm Technologies' Snapdragon 8 Elite Extreme Gen 6. The achievement marks the industry's first LPDDR6 validation on Qualcomm Technologies' next-generation mobile platform.
The milestone reflects close engineering collaboration between Samsung and Qualcomm Technologies from the early stages of development. It demonstrates that Samsung LPDDR6 can meet the demands of advanced mobile platforms and underscores the two companies' continued efforts to advance memory technology for more demanding on-device AI workloads.
“We are proud to have successfully completed sampling and validation of LPDDR6 on Qualcomm Technologies' next-generation mobile platform, following Samsung's industry-first LPDDR6 development," said Jangseok Choi, Vice President, Product Planning Team, Memory Business at Samsung Electronics. "This achievement demonstrates both the technology leadership and product readiness of our LPDDR6. Working closely with Qualcomm Technologies, we will continue advancing high-performance, low-power memory solutions that enable the next generation of AI devices.”
“The successful validation of Samsung's LPDDR6 memory on Snapdragon 8 Elite Extreme Gen 6 demonstrates the close collaboration between Qualcomm Technologies and Samsung to advance next-generation mobile technologies,” said Salman Saeed, Vice President, Product Management, Qualcomm Technologies, Inc. “LPDDR6 delivers the bandwidth and power efficiency needed to support increasingly sophisticated on-device AI experiences, and we are pleased to achieve this industry-first milestone as we continue driving innovation across the mobile ecosystem.”
More details on the collaboration will be shared at Snapdragon Summit 2026.
* The contents of this page are provided for informational purposes only. No representation or warranty (whether express or implied) is made by Samsung or any of its officers, advisers, agents, or employees as to the accuracy, reasonableness or completeness of the information, statements, opinions, or matters contained in this page, and they are provided on an "AS IS" basis.
* Samsung will not be responsible for any damages arising out of the use of, or otherwise relating to, the contents of this page. Nothing in this page grants you any license or rights in or to information, materials, or contents provided in this document, or any other intellectual property.
* The contents of this page may also include forward looking statements. Forward looking statements are not guarantees of future performance and that the actual developments of Samsung, the market, or the industry in which Samsung operates may differ materially from those made or suggested by the forward-looking statements contained in this page.
* All data on products herein, including their performances, are based on internal testing using standard Samsung benchmarks under laboratory conditions.
* All design, features and specifications represented herein may change without notice.
* Test results do not guarantee future performance under such test conditions, and the actual throughput or performance that any user will experience may vary depending upon many factors.
Semiconductor Materials Expert | CMP / Post-CMP Cleaning | Advanced Semiconductor Manufacturing | Yield Improvement & Process Integration | Materials Strategy & Supply Chain
3wThis is an important milestone for on-device AI because the memory subsystem is becoming just as critical as compute performance itself. As mobile AI workloads become more agentic, multimodal, and persistent, processors need to move increasingly large amounts of model weights, KV cache data, and sensor information with very tight power constraints. In that environment, higher bandwidth alone is not enough. Power efficiency, reliability, and dynamic workload management all have to improve together. What stands out about LPDDR6 is that these challenges are being addressed at the architecture level through higher I/O performance, DVFS, DEM, and enhanced reliability mechanisms such as PRAC. The validation with Snapdragon 8 Elite Extreme Gen 6 is equally significant. Memory technology only creates real system value when it is optimized and validated together with the SoC and platform. The next generation of mobile AI will depend not only on faster NPUs, but on how efficiently compute and memory can be co-designed to move data with higher bandwidth, lower power, and greater reliability.
Impressive validation — 10.7Gbps I/O speeds are a massive leap for on-device AI. In our daily work supporting procurement and engineering teams, we often see that software and silicon move faster than the supporting hardware supply chain. When next-gen memory is validated and ramps up, older memory and companion chips quietly slide toward obsolescence. Finding a reliable source for these legacy parts — without introducing counterfeit risks or triggering an unplanned requalification — becomes the real bottleneck. At NOVA COMPONENTS, we help teams navigate exactly that transition: sourcing original, factory-sealed long-lead-time and hard-to-find components so that a breakthrough in AI doesn't get stalled by a missing legacy part in the BOM.
Looking forward to it!!!
On-device AI depends on efficient data movement as much as compute. LPDDR6’s bandwidth, power efficiency, and reliability are key to scaling mobile intelligence.