the next generation of smartphones may usher in a revolutionary upgrade to memory architecture. industry sources indicate that mainstream manufacturers are actively evaluating the feasibility of replacing the current lpddr standard with llw dram (low-latency wide-bandwidth dynamic random access memory). this technology is optimized for high‑load ai computing scenarios and aims to overcome the memory bandwidth and latency bottlenecks currently faced by mobile devices during on-device inference of large models.
while traditional lpddr offers advantages in power efficiency and integration, it struggles to meet the stringent real-time data throughput demands of edge‑side ai. server‑grade hbm, though highly performant, is constrained by packaging complexity, thermal management requirements, and cost, making it impractical for direct deployment on smartphone platforms. llw dram, by contrast, strikes an efficient balance between the two—employing a single‑layer stacking design that eliminates the need for additional cooling modules while delivering memory bandwidth density and response speeds far exceeding those of lpddr. preliminary tests suggest its theoretical bandwidth could be 10–15 times that of existing lpddr5x, while reducing memory access latency to the microsecond range.
according to fixed focus digital, citing supply chain sources, llw dram also excels in energy efficiency: at equivalent computational performance, power consumption is reduced by approximately 50%, and overall ai task processing efficiency improves by up to 1.5 times. currently, this technology remains in the tape‑out verification stage, with mass production expected to begin as early as the third quarter of 2027. huawei and xiaomi are likely to be among the first brands to commercially adopt it.
