"What the domestic fabless (chip design) industry needs more than government subsidies is the market."
Park Jae-hong, 61, head of Boss Semiconductor, met with ChosunBiz on the 16th and cited "lack of demand" as the challenge for Korea's system chip industry. He noted that while research and development (R&D) support from the government matters, it will be hard to secure industrial competitiveness unless a market forms where domestically made chips are actually used and validated.
Boss Semiconductor is a car-focused artificial intelligence (AI) fabless company founded in 2022 by Park, a former executive vice president of Samsung Electronics' System LSI and foundry divisions. The core team includes talent who led the mobile application processor (AP) and application-specific integrated circuit (ASIC) businesses at Samsung Electronics, and the company recently secured a $65.2 million series A round. It is a homegrown fabless player challenging Qualcomm's dominance in the in-vehicle AI chip market.
Boss Semiconductor is pushing to commercialize the Eagle-N, an in-vehicle AI accelerator. The chip targets advanced driver-assistance systems (ADAS), Autonomous Driving, and in-car Generative AI services. As the auto industry moves beyond basic driving functions to adopt in-vehicle large language model (LLM)-based voice assistants and agentic AI services, related AI compute demand is rising quickly.
Last month, Boss Semiconductor completed tape-out for Eagle-N prototypes using Samsung Electronics' 5-nanometer process. It will then begin full-fledged performance evaluations with automaker clients in Germany and China.
Park said, "Clients in Germany and China are reviewing whether Eagle-N can be applied," adding, "China makes decisions quickly, so we expect it may be adopted in mass-produced models next year, but Europe and Japan have longer validation steps, and it could take more than three years to reach mass production."
Boss Semiconductor's bet is the "chiplet" architecture. Instead of implementing all functions on a single large chip, it combines multiple chips, which can raise Production yield and lower manufacturing expense. The larger the die, the higher the defect risk. By contrast, connecting multiple small chips in a chiplet architecture can boost production yields, drawing attention across the industry as a next-generation design approach.
Park said, "It's a method first used in data center chips by AMD and Intel," adding, "Our analysis shows adopting a chiplet architecture can cut manufacturing expense by up to half."
Automakers are also actively moving to build a chiplet ecosystem. Boss Semiconductor is taking part in a chiplet consortium led by Japan's auto industry and a related project centered on European research institutes. The company said this reflects efforts by carmakers to lower dependence on specific chip corporations and diversify supply chains.
On the Qualcomm-led market for in-vehicle chips, Park said, "From the standpoint of carmakers, they can't help but worry about rising dependence on a particular supplier," adding, "There will be continued demand for new supply chains."
Beyond cars, Boss Semiconductor sees Autonomous Driving farm machinery and the Edge AI market as new growth opportunities. As in-car use of Generative AI and agentic AI increases, demand is growing for AI computing platforms rather than simple Autonomous Driving chips, the company said.
Park, however, pointed to a shortage of demand-side corporations as a structural limit of Korea's system chip industry. He said, "Fabless businesses can only grow if someone uses their products," adding, "It's meaningful for the government to give subsidies directly to fabless corporations, but it's also worth considering incentives for corporations that adopt domestic chips."
He added, "Clients in Germany and China often evaluate based on technology and business viability rather than corporate size," and stressed, "It's important in Korea to build a virtuous cycle where supplier corporations and demand-side corporations grow together."
He went on, "For the fabless industry to gain competitiveness, not only technology development but also a market that can accumulate real-world use cases must grow together," adding, "Building an ecosystem where demand-side corporations and supplier corporations grow together is the biggest task for Korea's system chip industry." The following is a Q&A with Park.
― How far along is Eagle-N development?
"In July this year, we carried out tape-out using Samsung Electronics' 5-nanometer process. We will then begin full-fledged evaluations with automaker clients in Germany and China. For in-vehicle chips, evaluation and validation are critical. Clients will test in real vehicle environments and decide on mass production based on the results."
― When do you expect commercialization?
"It varies by region. China's decision-making is relatively fast, so we expect possible adoption in mass-produced models next year. Europe and Japan, by contrast, have longer validation. It usually takes more than three years. Because once adopted, in-vehicle chips are used for a long time, customers take a conservative approach."
― What are Boss Semiconductor's differentiators?
"The biggest differentiator is the chiplet architecture. It splits functions that used to be implemented on a single large chip into multiple chips. As the die grows, Production yield drops, but the chiplet approach can raise Production yield. We are actively applying chiplets in automotive chips."
― How economical is the chiplet architecture?
"Analyses of cases like AMD and Intel show manufacturing expense can be reduced by up to half. In Japan and Europe, efforts to build chiplet ecosystems are also progressing actively, centered on automotive chips. The auto industry likewise sees chiplets as a key technology to reduce reliance on specific corporations."
― Are there markets beyond automobiles where it can be applied?
"Autonomous Driving farm machinery is a prime example. Recently, even inside cars, use of Generative AI and agentic AI is increasing. Functions that converse with drivers or perform booking services are becoming more important. Eagle-N can be used not only for Autonomous Driving but also as an Edge AI AI Semiconductor."
― What is the hardest part as a startup?
"The in-vehicle chip market is conservative. Clients typically want companies that can supply stably for more than 10 years. Since we are still a startup, earning trust is crucial. We are persuading customers based on past project experience and the reputation we've built in the industry."
― What does Korea's system chip industry need most?
"The market matters more than subsidies. You need customers who actually use the products. Competitiveness comes from repeating the process of getting feedback in the market and improving the product. Building that virtuous cycle is most important."
― Any takeaways from working with overseas clients?
"Clients in Germany and China strongly tend to evaluate based on technology and business viability rather than corporate size. If there is good technology and a good product, there is a willingness to work with startups. I hope a similar environment forms at home where demand-side corporations and supplier corporations can grow together."