SEMIFIVE said it has begun mass production of HyperAccel’s Bertha data-center AI inference accelerator on Samsung Foundry’s 4-nanometer process, calling it the firm’s first large-scale mass-production project on that node. Company materials describe a big die of more than 500 square millimeters aimed at data-center inference workloads, including large language models.
SEMIFIVE also reported KRW 42.3 billion in new mass-production orders in the first half of 2026, versus KRW 21.2 billion for full-year 2025, with intake rising from KRW 15.6 billion in the first quarter to KRW 26.7 billion in the second. That KRW 42.3 billion figure is company-wide H1 mass-production order intake. It is not a disclosed Bertha-only contract value; no Bertha-only deal price was disclosed.
DigiTimes on Sep. 9 framed Samsung’s SF4 capacity as tight; utilization percentages were not disclosed.
Sources: SEMIFIVE PR Newswire, Sep. 8, 2026; Herald Business; DigiTimes, Sep. 9.
