XCENA, a fabless semiconductor company, showcased its CXL-based memory solutions for AI inference at the OCP Global Summit 2026 in San Jose. The demonstrations address the growing memory requirements and data-movement costs of AI inference. Using its MX1 computational memory solution, XCENA presented two key use cases.
The first demonstrated multiple AI inference servers sharing a KV cache through a CXL memory pool, reducing redundant data copies. The second utilized MX1's near-data processing (NDP) to accelerate vector search for retrieval-augmented generation (RAG) directly in memory. Additionally, XCENA and SK hynix confirmed that their CXL hybrid memory architecture, using Infinite Memory, offers greater capacity and performance than conventional solutions.
Executives from XCENA, including Harry Kim and Grant Mackey, participated in panels and technical sessions alongside partners like Samsung, Marvell, and Microsoft to discuss disaggregated memory architectures and addressing the memory wall for AI.
Source: TechPowerUp News · Summarized by HeadlinesBriefing