Semifive begins mass production of AI inference accelerator using Samsung 4nm process

Semifive, an AI application-specific integrated circuit (ASIC) specialist, has begun mass production of ‘Bertha’, a data center AI inference accelerator developed for HyperAccel. This marks the first time Semifive has utilized Samsung Electronics’ 4nm foundry process for large-scale production.

The Bertha chip is a ‘Big Die’ product with a surface area exceeding 500mm², a scale that presents significant challenges in managing power consumption, heat dissipation, and yield. Semifive provided a complete turnkey solution, managing the entire process from front-end design and verification to packaging, testing, and mass supply.

Semifive reported significant financial growth, with new orders for mass production in the first half of this year totaling 42.3 billion won, approximately double the total orders from the previous year. The company is expanding its product lineup, following previous successes with AI chips for security cameras and high-performance computing (HPC) in Japan.