SK hynix Develops LPDDR6 DRAM on 1c Node, 33% Faster Than LPDDR5X
Key Facts
- What: SK hynix has successfully developed its first LPDDR6 DRAM chips using its leading-edge 1c (10nm-class) process node.
- Performance: 16Gb chips deliver base operating speeds over 10.7 Gbps, providing up to 33% faster data processing than LPDDR5X.
- Efficiency: More than 20% lower power consumption compared to previous-generation LPDDR memory, achieved through sub-channel architecture and Dynamic Voltage and Frequency Scaling (DVFS).
- Target Markets: Initially aimed at smartphones and tablets for on-device AI workloads, with future adoption expected in AI data centers and Nvidia’s post-Vera Rubin superchips.
- Context: Announcement comes eight months after JEDEC finalized the LPDDR6 standard; Samsung previously revealed its own LPDDR6 design.
SK hynix announced the successful development of its first LPDDR6 DRAM on its advanced 1c process node, delivering base speeds exceeding 10.7 Gbps on 16Gb chips while offering up to 33% faster performance and more than 20% better power efficiency than LPDDR5X. The new low-power memory is designed primarily for smartphones and tablets handling on-device AI tasks but is also expected to play a major role in next-generation AI servers that rely on LPDDR-based SOCAMM modules.
The South Korean memory giant developed the LPDDR6 on its 1c 10nm-class process, which it first announced in 2024. According to the company, the new memory achieves its speed gains through wider bandwidth and higher data transfer rates. Power savings come from two key innovations: a new sub-channel structure that activates only the data paths currently in use, and Dynamic Voltage and Frequency Scaling (DVFS) that lowers clock speed and voltage during periods of lighter memory demand.
This marks SK hynix’s entry into the LPDDR6 market eight months after JEDEC published the official standard last July. Samsung had already unveiled its own LPDDR6 product at CES 2025, also reaching 10.7 Gbps speeds, highlighting the rapid industry transition toward the new low-power DRAM specification.
Technical Details and Design Improvements
The 16Gb LPDDR6 chips represent a significant leap over current LPDDR5X solutions. While the base operating speed is listed as “over 10.7 Gbps,” SK hynix emphasizes that the effective performance improvement reaches 33% compared with the best available LPDDR5X modules. This boost stems from both the increased per-pin data rate and architectural enhancements that improve overall bandwidth.
Power efficiency was a core focus during development. By implementing the sub-channel design, SK hynix can selectively power only active data paths, avoiding unnecessary energy consumption across the entire memory array. Combined with DVFS technology, the new memory reduces overall power draw by more than 20% relative to previous-generation LPDDR products, an important advantage for battery-powered mobile devices and power-constrained AI accelerators.
The 1c process node itself contributes to these gains by enabling higher transistor density and improved electrical characteristics compared with earlier 10nm-class nodes. This manufacturing technology allows SK hynix to deliver both higher speeds and better efficiency within the same physical footprint.
Growing Role in AI Infrastructure
While LPDDR6 will first appear in smartphones and tablets optimized for on-device AI inference, its long-term impact may be even greater in the data center. LPDDR memory has gained significant traction in AI servers through SOCAMM and SOCAMM2 module formats, which are more compact and power-efficient than traditional server DIMMs.
Nvidia’s GB300 Grace Blackwell Ultra Superchip already uses SOCAMM modules based on LPDDR memory. The company’s latest Vera Rubin Superchip adopts the newer SOCAMM2 specification. SK hynix previously indicated that it expects post-Vera Rubin Nvidia AI designs to incorporate LPDDR6, suggesting the new memory could become a key component in future AI supercomputers.
The combination of higher bandwidth and improved power efficiency makes LPDDR6 particularly attractive for AI workloads that require both fast memory access and careful thermal and power management. As AI models continue growing in size and complexity, the ability to deliver high throughput with lower power consumption becomes increasingly valuable.
Competitive Landscape and Future Roadmap
SK hynix is not the first to announce LPDDR6 — Samsung demonstrated its version earlier — but the company’s use of its latest 1c process node positions it strongly in the race to ramp production. Both manufacturers are targeting the same 10.7 Gbps baseline speed, indicating that this figure represents the initial standard for LPDDR6 products.
Looking further ahead, JEDEC anticipates that LPDDR6 will eventually support data rates as high as 14,400 MT/s. This would significantly exceed even the fastest DDR5 overclocking records, pointing to substantial headroom for future performance improvements as memory makers refine their designs and manufacturing processes.
The development of LPDDR6 reflects the broader industry shift toward specialized memory solutions for AI. Traditional DRAM architectures are being supplemented — and in some cases replaced — by formats optimized for the unique bandwidth, power, and form-factor requirements of AI training and inference systems.
Impact on Mobile and Data Center Markets
For smartphone and tablet makers, LPDDR6 offers a compelling upgrade path. The combination of 33% higher performance and 20% better power efficiency should translate into faster app loading, smoother multitasking, and improved battery life, particularly for AI-enhanced features such as real-time image processing, natural language understanding, and on-device generative AI.
In the data center, the adoption of LPDDR6 in SOCAMM2 modules could help Nvidia and other AI chip designers build more power-efficient systems. As energy costs and thermal constraints become increasingly important in large-scale AI deployments, the efficiency gains from LPDDR6 could prove strategically significant.
The memory industry’s rapid development of LPDDR6 also signals confidence that demand for high-bandwidth, low-power memory will continue growing as AI capabilities expand from cloud data centers into edge devices and consumer electronics.
What's Next
SK hynix has not yet disclosed specific timelines for mass production or commercial availability of its 1c LPDDR6 chips. However, given the company’s statement that the memory is developed and ready, sampling to key customers is likely already underway or imminent.
Future iterations of LPDDR6 are expected to push speeds well beyond the initial 10.7 Gbps baseline as manufacturers optimize their designs. The JEDEC roadmap to 14,400 MT/s suggests multiple generations of performance improvements remain ahead.
The memory will likely first appear in flagship smartphones and AI-focused tablets before migrating into enterprise AI servers. Its success in both markets will depend on how effectively system designers can leverage the increased bandwidth and improved power characteristics.
As AI workloads continue proliferating across consumer and enterprise environments, LPDDR6 represents an important step in the ongoing evolution of memory technology to meet the unique demands of artificial intelligence.