SEMIFIVE Starts Mass Production of HyperAccel Bertha AI Chip

SEMIFIVE has begun mass production of HyperAccel's Bertha inference accelerator, the company said in a press release. The data centre chip is built for large language model inference.

Bertha is made on Samsung Foundry's 4nm process and is what the industry calls a big die, exceeding 500 square millimetres. SEMIFIVE supplied a turnkey service covering front end design and verification, packaging, testing and volume manufacturing supply, and says follow on purchase orders are expected as HyperAccel expands its services.

The programme is SEMIFIVE's first large scale mass production project on that process node, which is the part that matters beyond this one chip. Designing a chip and shipping one in volume on a leading edge process are different problems, and the second is where custom silicon efforts usually stall.

The company's order book has moved with it. SEMIFIVE says it took KRW 42.3 billion in new mass production orders in the first half of 2026, against KRW 21.2 billion for the whole of the previous year, with quarterly intake rising from KRW 15.6 billion in the first quarter to KRW 26.7 billion in the second, and overseas customers accounting for 45% of second quarter bookings. Its pipeline also includes the Wisenet 9 AI chip for Hanwha Vision security cameras and a high performance computing AI chip for a Japanese customer.