SEMIFIVE, a South Korea-based custom AI semiconductor solutions provider, has commenced mass production of an AI inference accelerator called Bertha for HyperAccel, the company announced on September 8, 2026. The chip is manufactured on Samsung Foundry's 4nm process node and represents SEMIFIVE's first large-scale mass production project at that technology level.

A Large-Die Challenge at Advanced Nodes

The Bertha accelerator is designed to handle AI inference workloads, including those driven by large language models. What sets the project apart from a manufacturing standpoint is the chip's physical size: it exceeds 500 square millimeters, placing it in the category of what the industry refers to as a "Big Die."

At that scale, managing power consumption, heat dissipation, and manufacturing yield becomes significantly more difficult, particularly when combined with the demands of an advanced process node such as Samsung's 4nm.

SEMIFIVE delivered what it describes as a complete turnkey solution for the project, covering front-end design and verification through packaging, testing, and volume manufacturing supply.

The company said that bringing a demanding project of this complexity into successful mass production validates its execution capabilities at the cutting edge of semiconductor process technology.

Follow-on purchase orders are anticipated as HyperAccel expands its services, with mass production volumes projected to grow in step with that rollout.

Broadening Portfolio Across Application Domains

The HyperAccel program is the third distinct mass production milestone SEMIFIVE has reached within roughly a year. In the third quarter of last year, the company completed mass production of the Wisenet 9 AI ASIC chip, developed for Hanwha Vision's security camera product line.

Earlier this year, in the second quarter, SEMIFIVE brought a high-performance computing AI chip for an unnamed Japanese customer into mass production. The data center AI inference accelerator for HyperAccel follows in the third quarter of this year.

The progression across security, high-performance computing, and data center AI reflects a deliberate effort by SEMIFIVE to diversify its portfolio across application domains and levels of technical complexity.

The company said this approach strengthens business continuity and earnings momentum by moving beyond one-time non-recurring engineering projects toward stable, recurring mass production revenue aligned with customers' long-term product roadmaps.

Order Intake Nearly Doubles Year Over Year

The expanding production lineup has translated into rapid financial growth for the company. SEMIFIVE secured KRW 42.3 billion in new mass production orders during the first half of this year alone. That figure represents nearly double the KRW 21.2 billion in full-year order intake the company recorded across the entirety of last year.

The growth accelerated further within the first half itself. Quarterly order intake rose 71 percent, climbing from KRW 15.6 billion in the first quarter to KRW 26.7 billion in the second quarter.

International business also gained ground, with overseas orders accounting for 45 percent of second-quarter bookings. SEMIFIVE characterized that proportion as marking the full-scale activation of its global business pipeline.

The company noted that the quality of its order intake has also improved alongside the volume increase, supported by stable original equipment manufacturer volumes backed by secured end customers and high-value data center ASIC projects.

Shift from Training to Inference Drives Demand

SEMIFIVE's chief executive officer addressed the broader market context driving demand for the type of chip the company is now producing at scale. As the AI industry transitions its focus from model training to inference deployment, demand for custom ASIC accelerators in data center environments is increasing.

Inference workloads, which involve running already-trained models to generate outputs in real time, place different demands on hardware than training does, and purpose-built chips designed specifically for inference can offer advantages in efficiency and performance over general-purpose processors.

The CEO said the successful mass production of the HyperAccel chip validates the company's capabilities in advanced process execution. The statement described the combination of HyperAccel's chip architecture with SEMIFIVE's turnkey production capabilities as enabling a seamless path from design through to volume manufacturing for a die of more than 500 square millimeters. The CEO also stated the company's intention to carry the momentum from the first half through the rest of the year.

Turnkey Model as Competitive Differentiator

The turnkey delivery model that SEMIFIVE emphasizes covers the full span of chip development and production, from early design and verification stages through to final packaging, testing, and high-volume supply.

For customers developing custom AI silicon, this approach removes the need to coordinate across multiple independent vendors for different stages of the production process.

The Bertha project illustrates the scope of what that model entails in practice. Producing a chip exceeding 500 square millimeters on a 4nm node requires managing yield challenges that grow in difficulty as die area increases.

Defects that might be acceptable on a smaller die become more likely to appear somewhere on a larger one, making yield management a central concern in achieving commercially viable mass production at scale.

SEMIFIVE's completion of the initial production contract and anticipation of follow-on purchase orders suggest the yield and supply challenges associated with the large-area chip have been addressed to a degree sufficient for commercial volume production to proceed.

Tap Into Every Semiconductor Opportunity Across South Korea Before Your Competitors Do

The South Korean semiconductor landscape is evolving at a pace that leaves little room for guesswork. From advanced fabrication plants breaking ground to emerging chip manufacturing corridors in South Korea, the volume and velocity of project activity demand more than periodic market reports it demands a dedicated intelligence system built for the industry.

That is precisely what the Global Project Tracking (GPT) platform by Blackridge Research delivers. Designed for manufacturers, equipment suppliers, EPC contractors, and investors operating across the semiconductor value chain, it consolidates the project intelligence you need into a single, continuously updated source of truth.

Whether you are mapping capital expenditure cycles, identifying procurement windows, or benchmarking facility development timelines across South Korea, the platform gives your team the structured data to act with confidence at every stage of the project lifecycle.

• Upcoming Projects

• Tender Notices

• Contract Awards

• Projects Under Construction

• Completed Projects

See how the Global Project Tracking (GPT) platform by Blackridge Research can sharpen your market strategy across South Korea. Book a Free Demo with our team today.