Everspin Technologies has successfully demonstrated what it describes as the world's first Compute Express Link (CXL)-connected MRAM platform at the SNIA Developer Conference 2026, and the system is now ready for customers to test with their own workloads.
The demonstration introduces a fast, persistent memory tier positioned between DRAM and NAND flash storage, giving CXL-based systems another place to store data that needs to remain available through a power interruption or system restart. Everspin sees potential applications in areas including checkpointing, meaning saving a system's state so it can resume after an interruption, persistent key-value (KV) cache, write buffering, and persistent caching.
The working system connects Everspin's PERSYST MRAM to CXL-enabled platforms as persistent memory using memory semantics, combining CXL's standard high-speed connectivity with MRAM's ability to retain data without needing continuous power. This creates a fast, persistent buffer and cache layer offering cache-line-level, nanosecond-class access, with write speeds up to 100 times faster than NAND-based solid-state storage. Data stays available through a power interruption or system restart without relying on battery-backed DRAM or needing to first transfer data to an SSD.
The demo setup uses a Supermicro AS-1116CS-TN server with an AMD EPYC 9355 32-core processor, an AMD Alveo U250 FPGA platform paired with a Wolley CXL controller, and Everspin PERSYST 1GB DDR4 MRAM UDIMMs. This configuration provides an open, standards-based platform demonstrating coherent access to persistent MRAM through CXL. It also allows memory resources to be pooled and shared across multiple host systems, letting persistent memory capacity be allocated wherever specific workloads need it. For AI systems, this opens up opportunities for applications like checkpointing and persistent KV cache.
Sanjeev Aggarwal, president and CEO of Everspin Technologies, said the company has worked with data center customers for 15 years and continues innovating as AI changes the memory requirements of computing systems. He said CXL gives Everspin a standard way to bring persistent MRAM to a much broader range of computing platforms, and that the company now has a working system so customers can run their own workloads and measure the benefit for their specific applications.
At the conference, Everspin and Wolley demonstrated read and write access to MRAM through the CXL interface under demanding test conditions. Dr. Bernard Shung, CEO of Wolley, said Wolley enables memory and storage connectivity through CXL, and that this demonstration with Everspin shows how persistent MRAM can be brought into that environment using memory semantics. He said together, the two companies are demonstrating a practical way to connect persistent memory through CXL and make it available to today's computing platforms.
The demonstration builds on Everspin's ongoing work extending MRAM into new memory architectures, including persistent caching, write buffering, and metadata applications.
