JEDEC has officially released the LPDDR6 memory standard (JESD209-6), which is designed to significantly boost performance, power efficiency, and security for mobile devices and AI applications.
Key Features of LPDDR6 #
‘Performance:’
LPDDR6 introduces a dual-subchannel architecture, maintaining a minimum access interval of 32 bytes while allowing for more flexible operations.
- ‘Dual Subchannels:’ Each die supports two subchannels, with each subchannel featuring 12 data signal lines (DQ) for optimized channel performance.
- ‘Instruction Optimization:’ Each subchannel includes four command/address (CA) instructions, reducing ball count and improving data access speeds.
- ‘Static Efficiency Mode:’ Supports larger capacities and maximizes the use of bank resources.
- ‘Flexible Data Access:’ Offers real-time burst length control, supporting 32/64-byte access.
- ‘Dynamic NT-ODT (Non-Target On-Die Termination):’ Adjusts ODT based on load requirements to improve signal integrity.
While JEDEC did not specify the exact data transfer rates, LPDDR6 is expected to start at over 10 Gbps (10,667 Mbps) and reach a maximum of 14,400 Mbps (14.4 GHz). In comparison, LPDDR5 started at 6400 Mbps, LPDDR5X reached 8533 Mbps, and SK Hynix’s LPDDR5T hit 9600 Mbps.
Efficiency and Power Savings: #
LPDDR6 reduces voltage and power consumption by adopting VDD2 power supply and a dual power supply system.
- Alternating Clock Command Input: Improves performance and efficiency.
- Low-Power Dynamic Voltage Frequency Scaling (DVFSL): Reduces VDD2 power supply during low-frequency operation to save energy.
- Dynamic Efficiency Mode: Uses a single subchannel interface suitable for low-power, low-bandwidth scenarios.
- Partial Self-Refresh and Active Refresh: Reduces refresh power consumption.
Security and Reliability: #
LPDDR6 also includes significant enhancements for security and reliability.
- Per-Row Activation Count (PRAC): Supports memory data integrity.
- Carve-Out Meta Mode: Allocates specific memory regions for critical tasks, improving overall system reliability.
- Programmable Link Protection, ECC Error Correction, Command/Address (CA) Parity, Error Scrubbing, and Memory Built-In Self-Test (MBIST): Enhances error detection capabilities and system reliability.
Major semiconductor test vendors, memory chip manufacturers, and terminal manufacturers—including Advantest, Cadence, Synopsys, Samsung, SK Hynix, Micron, Qualcomm, and MediaTek—have expressed strong support for LPDDR6.
Samsung and Chinese Manufacturers Race for LPDDR6 #
Last month, Samsung Electronics announced it will begin mass production of LPDDR6 memory in the second half of this year using its sixth-generation “1c DRAM” process, supplying tech giants like Qualcomm. The 1c DRAM process achieves higher transistor density and better energy efficiency compared to previous generations. Samsung has notably improved the yield of 1c DRAM and plans to expand production lines at its Hwaseong factory in South Korea.
Industry sources indicate that Qualcomm’s next-generation flagship chip, the “Snapdragon 8 Elite Gen2,” will be the first to support LPDDR6 memory, expected to be unveiled at the Snapdragon Summit this year. Samsung is also utilizing 1c DRAM technology in HBM4 and other high-end storage products, aiming to cover all AI-related memory solutions.
China’s memory technology is also advancing rapidly. CXMT (ChangXin Memory Technologies) has made significant strides in low-power memory semiconductors (LPDDR), successfully developing and mass-producing LPDDR5X, and is now pushing toward LPDDR6. Analysis suggests that CXMT could achieve LPDDR6 mass production as early as 2026, potentially within a year of Samsung’s timeline.
Desktop Memory Advances: G.Skill Launches High-Capacity DDR5 Kits #
In the desktop memory sector, performance continues to accelerate.
G.Skill has announced the availability of the world’s first 256GB memory kit, consisting of four 64GB DDR5-6000 CL32 modules. These kits are available in the Trident Z5 Neo RGB and Flare X5 series.
This memory kit uses the latest energy-efficient SK Hynix chips and is designed to run stably at 6000MHz with tight timings of CL32-44-44-126. G.Skill also offers a 128GB kit (two 64GB modules) with timing options of CL34-44-44 or CL36-44-44.
These memory kits are optimized for AMD Ryzen 9000 series processors and support AMD EXPO overclocking technology. Users should verify motherboard support and update the BIOS to the latest version for compatibility with single 64GB modules.
Additionally, G.Skill has launched a new high-frequency overclocking memory kit featuring two 64GB modules for a total of 128GB, reaching speeds of DDR5-6400 with timings of CL36-44-44. This kit is suited for Intel platforms requiring high-frequency memory and supports Intel XMP 3.0 technology. It is available in the Trident Z5 RGB series.
Both memory kits are currently shipping and will soon be widely available.
