LPDDR6 will support up to 512GB memory modules to target the needs of the AI ​​era

[ Gearbest Technology News]With the rapid development of generative AI and agent AI applications, the demand for memory capacity and energy efficiency in data centers continues to rise. Recently, JEDEC announced the technical plan for the next generation LPDDR6 memory standard. The new standard will focus on improving memory density and energy efficiency. The maximum capacity of a single module is expected to reach 512GB, providing stronger support for future AI training and inference scenarios.

memory chip

According to the disclosed information, LPDDR6 will adopt SOCAMM2 packaging form and achieve higher capacity integration through new architectural design. Compared with the current LPDDR5X platform’s maximum single module capacity of 256GB, LPDDR6 is expected to double the capacity to 512GB. JEDEC stated that the core goal of the new standard is not to simply pursue bandwidth growth, but to prioritize meeting the demand for large-capacity memory in the artificial intelligence era.

From the perspective of performance planning, LPDDR6 is expected to bring about a 10% to 20% increase in bandwidth, and the increase is relatively limited. Compared with performance indicators, R&D focuses more on improving data storage capabilities per unit space. To this end, the new standard will use narrower internal interfaces and new memory architecture designs to integrate more memory cells in the same area.

In addition to capacity upgrades, LPDDR6 will also introduce LPDDR6-PIM (Processing-In-Memory) technology. This technology allows some computing tasks to be completed directly inside the memory without having to hand them over to the central processor for execution. By completing some operations in the storage link, the performance loss caused by data transfer can be effectively reduced, while the load on the server processor is reduced, and the overall energy efficiency performance is improved.

The industry believes that as the scale of AI models continues to expand, the traditional development path that relies on improving processor performance is facing higher power consumption pressure, and the importance of memory systems is also increasing. The LPDDR platform already has the advantage of low power consumption, and the addition of PIM technology further enhances its application value in artificial intelligence infrastructure.