PFN Continues Vertical Stacking Logic and DRAM Approach, MN-Core L2000 Chip Aims to Double Memory Bandwidth
IT Home news, November 17: Japanese AI startup Preferred Networks (PFN) announced today local time that it will continue its cooperation with chip design services company SEMIFIVE and chip contract manufacturing service provider Samsung Electronics on the AI inference accelerator MN-Core L2000, continuing to build products that vertically stack logic chips and DRAM memory.

PFN's first-generation MN-Core series product, the L1000, has a single-chip peak FP8 compute of 200 TFLOPS and integrates 32GB of on-chip memory with over 10TB/s of bandwidth. The second-generation product announced this time, the L2000, aims to achieve more than 2x the memory bandwidth of the L1000 series, further improving AI inference speed and improving the energy efficiency of AI workloads.
PFN plans to officially release the MN-Core L1000 in 2027, offering single-chip, four-chip box, and 16-chip workstation solutions. The MN-Core L2000 series will also complete chip design in 2027Q1.
Related Articles
- SK hynix: Plans to invest 1.4 trillion won over the next 5 years to support domestic chip suppliers
- Axelera AI unveils architecture of data-center-class AIPU chip Titania, production to begin in 2027Q3
- South Korea unveils on-device AI semiconductor roadmap, plans to complete development and verification of 10 chips by 2030
- UK AI chip startup Fractile receives government funding, valued at $7 billion
- AI data centers squeeze mature-process capacity, automotive chip supply chain faces bottlenecks
- DEEPX 2nm low-power AI chip DX-M2 physical chip expected to arrive in early 2027