Semifive begins mass production of HyperExcel's 4nm, 500 mm²-plus AI accelerator

  • Semifive began volume production of HyperExcel's Bertha data-center AI inference accelerator.
  • The chip exceeds 500 mm² and uses Samsung Foundry's 4nm process.
  • Semifive secured 42.3 billion won in first-half 2026 volume-production orders.

Semifive, a South Korean AI application-specific integrated circuit (ASIC) design firm, has begun initial volume production of Bertha, an AI inference accelerator for data centers developed for HyperExcel. The launch marks Semifive's first large-scale production project using Samsung Foundry's 4-nanometer process. Bertha is a big-die chip exceeding 500 square millimeters and is optimized for repetitive inference workloads, in which trained large language models generate responses. Semifive provided a turnkey solution spanning front-end design and verification, packaging, testing and final volume-production supply. The company expects supply volumes to increase through additional purchase orders following HyperExcel's service launch and expansion. The program expands Semifive's production portfolio across security, high-performance computing and data-center AI, supporting recurring revenue beyond one-time non-recurring engineering work. Semifive's new volume-production orders totaled 42.3 billion won in the first half of 2026, compared with 21.2 billion won for all of 2025. Quarterly orders rose 71% from 15.6 billion won in the first quarter to 26.7 billion won in the second quarter, with overseas orders accounting for 45% of second-quarter bookings.

The information on this website is generated using AI and we cannot guarantee its accuracy. Please use it as reference information only.