JEDEC officially announced that the formulation of the JESD328 standard specification is close to completion, a newSOCAMM2(Small Outline Compression Attached Memory Module) memory form, serving the next generation of AI and data center platforms. The SOCAMM2 standard definesA low-profile, replaceable and upgradeable memory based on LPDDR5 and LPDDR5X, which can provide higher bandwidth, lower power consumption and better scalability.
Its maximum data transmission rate for a single pin can reach 9.6Gbps, which is 9600MT/s, a significant improvement compared to the existing 8533MT/s, meeting the bandwidth requirements of modern AI training and inference, while having lower energy consumption and heat dissipation requirements.
It also introduces the SPD function, which can be used for identity recognition and verification to ensure that reliability requirements are met.
Vendors such as NVIDIA are moving their server platforms to SOCAMM2-based designs, although they may be different at the physical level, they are basically similar.
JEDEC confirmed that the JESD328 SOCAMM2 specification will be released in the near future.
CAMM memory was originally Dell's proprietary design and was created by Dell senior engineer Tom Schnell. The Dell Precision 7000 series notebook workstations released in April 2022 were equipped with DDR5 CAMM memory for the first time.
At the end of 2022, Dell will contribute CAMM technology to JEDEC, hoping to become a unified industry standard.
In December 2023, JEDEC completed the formulation of the smaller CAMM2 standard, and motherboard manufacturers such as MSI and ASUS and memory manufacturers such as G.Q. immediately released related products.
Earlier in September 2023, Samsung announced that it would use LPDDR5X memory for smaller LPCAMM products.
In January 2024, Micron announced that it would use LPDDR5X memory for LPCAMM2, and it would be installed on the Lenovo ThinkPad P1 Gen7 notebook for the first time in May 2024.
In March 2025, Micron and NVIDIA jointly announced the development of SOCAMM memory for AI servers, paired with LPDDR5X memory, with a maximum single memory capacity of 128GB. Samsung and SK Hynix also joined later.
