“There are more than 4,000 artificial intelligence (AI) semiconductor corporations in China. Just that number makes it difficult to gauge the depth of manpower and the breadth of the market.”
Kim Hyung-joon, Director General of the Next-Generation Intelligent Semiconductor Business Division and a professor emeritus at Seoul National University, said in an interview on April 15, “China has surpassed South Korea in the AI field, developing its own large language models (LLMs) like DeepSeek, even impressing OpenAI.”
Director General Kim is a leading expert in semiconductor devices and processes in South Korea, having served as the head of the Semiconductor Research Institute at Seoul National University and as president of the Korean Society for Semiconductor Display Technology. The next-generation intelligent semiconductor project he leads aims to develop semiconductor designs, core technologies, and fine processing specialized in AI computations, with a total investment of 1.096 trillion won from 2020 to 2029. This project is also a resumption of a trillion-won level semiconductor research and development (R&D) project after 10 years.
Director General Kim warned, “In the legacy semiconductor sector, even despite a high-cost structure, China is rapidly closing the technology gap by pushing domestic semiconductor self-sufficiency based on government subsidies,” adding that, “China’s policies, which do not adhere to market logic, are posing a structural threat to the South Korean semiconductor industry.” Current analysis suggests that the technology gap between South Korea and China has narrowed to 2-3 years for DRAM and 1-2 years for NAND flash. He expressed concern that, “If we do not proactively develop our technology, we could be overtaken at any time.”
As a solution for recovering technological competitiveness, Director General Kim emphasized the need to reform the government R&D system. He stated, “The current structure, where the Ministry of Science and ICT, Ministry of Trade, Industry and Energy, Ministry of Education, and Small and Medium Business Administration each support individual projects, has clear limitations,” and argued that, “A presidential task force that can integrate and adjust R&D budgets and strategies is necessary.” He further emphasized that “the urgent establishment of a comprehensive government-level ‘technology strategy headquarters’ that encompasses strategic technologies such as semiconductors, AI, quantum computing, and batteries is needed.” The following is a question-and-answer session.
The ratio of R&D investment to gross domestic product (GDP) in South Korea is one of the highest in the world. Nevertheless, what is the background that allowed China to catch up technologically?
“The ratio of R&D investment to GDP in South Korea (4.96%) is the second highest in the world. However, in terms of absolute investment amounts, the United States invests about 10 times more than South Korea, and China invests 7-8 times more. In R&D, the absolute amount is more important than the ratio to GDP. Looking at semiconductors alone, the R&D investment ratio compared to sales is 18% in the U.S., 15% in Europe, and 11% in Taiwan, while South Korea’s is only 9%. Even in semiconductors, which are our key industry, R&D investment is not greater than that of other countries compared to sales.”
It is shocking that South Korea, which boasts the ultra-competitive semiconductor sector, has been stingy with R&D investment in semiconductors.
“The semiconductor industry has long been a blank space for large-scale national R&D projects. The perception that ‘large semiconductor corporations can operate without government support’ has been widespread, and corporations have concentrated their investments only in profitable sectors. As a result, future technologies such as fabless (semiconductor design) or packaging have been neglected. After the disappearance of large projects, even professors in semiconductor fields at universities have moved to other areas such as display, solar energy, and NANO. This has intensified the shortage of talent in semiconductor development. The government R&D has been operated in a fashion that chases fashion trends rather than long-term strategies, making it difficult to develop critical technologies like semiconductors that require a long-term outlook.”
There are serious concerns about decoupling between universities and the industrial field in R&D.
“Currently, the standards for hiring university professors are heavily focused on the number of papers and impact factors. Under such circumstances, getting published in prestigious journals like ‘Nature’ becomes advantageous. University R&D becomes focused on researching future technologies that could be commercialized in 10 years rather than research that corporations actually need. This leads to a lack of trust in university research among corporations. They perceive that data from university labs, extracted through small experiments in poor conditions, is difficult to apply for mass production compared to using the latest equipment. This creates a technological and trust gap between corporations and universities, which results in a structure where joint research is not activated.”
How should this issue be resolved?
“The government should build high-performance semiconductor research infrastructure directly at universities. For instance, Peking University in China has established megabyte-level experimental facilities, but domestic universities are stuck at kilobyte levels. It is essential to construct advanced infrastructure that can be utilized collaboratively, rather than leaving it to individual universities. Additionally, a government R&D task force that collaborates with government ministries should connect universities and research institutions, creating structures that provide shared facilities to turn specific material or component research results into actual chips, enabling corporate linkages and technology validation.”
Is there no aspect of R&D being distorted into support for small and medium enterprises?
“Currently, over 150 small fabless firms depend on government R&D funding of around 200 million won, but this is far too insufficient compared to the hundreds of millions of won required for chip production, leading most to exhaust their funds on operational costs. To reduce inefficiency, it would be effective to reallocate support to focus on around 20 promising corporations for concentrated support.”
There is a serious shortage of engineering talent in South Korea, and the current government is responding by expanding contract departments and AI graduate schools.
“This is merely a short-term remedy. Creating a single graduate school does not immediately produce the workforce, and in reality, it is difficult to fill the enrollment quotas. A structural approach is necessary. Fundamentally, societal recognition must be fostered that engineering and technology development are noble endeavors. It should be recognized that technology development is not merely about making a lot of money, but rather a calling that represents a challenge suited to one’s abilities. Especially, success stories of startups like Furious AI or Rebellion, which are challenging the global market based on talent and passion in South Korea, can serve as important role models for engineering talent. We need to properly nurture and promote them to instill confidence in the next generation that “they can succeed through technology.”
What is the direction of the Korean-style AI industry strategy?
“The AI industry has three core elements. The first is computing power, the second is data, and the third is models. China is training LLMs based on massive graphics processing unit (GPU) resources, and it has rich data and its own model development capabilities. In contrast, South Korea is lacking in GPU infrastructure and has little access to large-scale datasets or advanced models. All three elements are weak. Considering this reality, the strategic point where we can realistically aim for an advantage is in computing power, specifically the development of computational chips and the establishment of related infrastructure. It is urgent to mandate public computing centers to purchase a certain percentage of AI semiconductors developed by domestic fabless companies to create early success. Through this, startups with technological prowess can secure a foothold in the market and attract follow-up investment. We need to start from hardware, laying the foundation, and create conditions for the private sector to grow organically in data and model areas as well.”
Is the political discussion about creating or operating a state-owned AI corporation, like ‘K NVIDIA,’ a valid strategy?
“It is unlikely that the government can successfully create AI corporations. Government organizations often struggle to keep pace with speed and flexibility due to bureaucratic processes. Since AI is a field that evolves at a rapid pace, the government should focus on cultivating talent and building infrastructure, acting as a facilitator that enables private corporations to operate freely. The government’s role should be to relax regulations and lay down the foundation.”
Plus Point
Did weak government semiconductor R&D investments lead to Samsung Foundry’s downturn?
Concerns have been raised that the structural weakness in South Korea’s semiconductor R&D is the reason why Samsung Electronics' foundry business has not surpassed TSMC.
Director General Kim stated, “Samsung Electronics puts forth a one-stop service equipped with foundry, logic, and system LSI functions, but it lacks advanced packaging technology that combines high bandwidth memory (HBM) and GPUs, failing to meet the demands of clients such as NVIDIA.” He added, “Samsung Electronics has not actively invested in future technologies such as packaging and HBM due to a revenue-focused management strategy, leading to a decline in foundry competitiveness.”
However, Director General Kim believes that the root cause must be found in the weak government R&D system. The United States invests in semiconductor design, packaging, and materials, led by government departments such as the Defense Advanced Research Projects Agency (DARPA) and the Department of Energy (DOE) over the long term. Taiwan has fostered an ecosystem through public research institutions like the Industrial Technology Research Institute (ITRI) that collaborate with universities and industry. Director General Kim emphasized that, “The reason NVIDIA in the United States dominates the AI semiconductor market is due to continued government R&D investment in packaging technology development and supporting fabless firms,” stating that “For future technologies that are difficult for corporations to tackle, the government must pave the way first.”