Silicon Motion has a new SSD controller for AI PCs and other edge AI use cases.
The SM23524XT is a DRAM-less PCIe Gen5 SSD controller relying on its host memory buffer to provide the memory needed for its operations. It has a quad-core ARM Cortex-R8 design, 4 x PCIe 5 lanes and NAND interface speeds up to 4800 MT/s enabling it to reach up to 14 GBps sequential read speeds and up to 2.5 million random read IOPS. This is up to 25 percent faster random performance than the prior 3-core SM2504XT controller, which, it says, slashes latency and accelerates response times for the highly fragmented data access patterns that define KV Cache and AI inference workloads.
Nelson Duann, SVP of Client and Automotive Storage Business at Silicon Motion, said: “KV Cache has become a critical factor in AI inference performance, driving the need for sustained high random read/write throughput and low-latency data access. As AI PCs evolve to support increasingly complex Local Agent and on-device LLM workloads, the SM2524XT is designed to deliver the random I/O performance, latency stability, and power efficiency required for next-generation AI storage architectures.”
The SM2524XT integrates Silicon Motion’s Separated Command Address (SCA) technology, advanced FTL scheduling, and its 8th generation NANDXtend LDPC ECC technologies to improve parallel data processing efficiency, reduce latency interruptions, and maintain consistent performance during sustained AI workloads.
NANDXtend is an Error Correction Code (ECC) and data recovery technology that continuously identifies and corrects bit errors. SCA separates command and address transmission paths to improve controller efficiency by allowing faster command processing and parallel data access. This enhances overall NAND interface performance.
Silicon Motion says KV cache workloads feature streams of highly fragmented, latency-sensitive random read/write operations needing sustained IOPS throughput and reliable low-latency performance under continuous load. It claims the SM2524XT maintains stable random I/O performance during sustained AI inference sessions.
The SM2524XT has up to 25 percent more performance per watt than the previous SM2504XT model.
Silicon Motion supplies its controllers to OEMs and system builders and products taking advantage of the SM2524XT could appear in a few months time, perhaps by the end of the year or in early 2027.