SK hynix unveils new AI memory and 3D FeNAND technologies

SK hynix has introduced new memory technologies designed to expand AI infrastructure capabilities beyond traditional High Bandwidth Memory (HBM). At the OCP Global Summit in California, the company showcased several advancements aimed at meeting the increasing capacity and efficiency demands of AI ecosystems.

Key developments include the MARS (Memory-Centric AI Rack-Scale System), an architecture that allows for the flexible coordination of computing and memory resources. The company also presented progress in High Bandwidth Flash (HBF) technology and a second-generation 256 GB memory product based on Compute Express Link (CXL) 3.2, which offers double the bandwidth and 30% better energy efficiency than its predecessor.

Additionally, research published by SK hynix demonstrates the use of 3D Ferroelectric NAND (3D FeNAND) to achieve Analog Computing-in-Memory (A-CiM). By utilizing the polarization states of ferroelectric materials, this technology allows NAND flash to perform basic AI operations, such as multiply-accumulate (MAC), directly within the memory. Simulations indicate that this design can improve throughput by up to 20 times and energy efficiency by up to 5,000 times compared to traditional 2D non-volatile memory arrays.