Korea eyes DPU development, aims to lead AI chip industry

[Graphics by Song Ji-yoon]

South Korean chip manufacturers must make efforts to take the lead in the development of next-generation semiconductors as competition in the global artificial intelligence (AI) chip sector intensifies, according to experts.

The Institute for Future Strategy at Seoul National University on Sunday listed the future semiconductor technologies that Korea must stake its survival on. These include data processing units (DPUs) specialized in data processing acceleration and ultra-small, ultra-low-power semiconductors that significantly reduce size and power consumption compared to current semiconductors.

“Korea could surpass Nvidia Corp. in the DPU chips field, which could become the mainstream after graphic processing units (GPUs),” according to Kim Jang-woo, chief executive officer and founder of MangoBoost Inc. and a professor of electrical and computer engineering at Seoul National University.

Seoul National University Materials Science & Engineering Professor Kim Sang-bum also noted that advancement in the semiconductor industry in the future is impossible without DPUs or ultra-small, ultra-low-power semiconductors, adding that “although research in these areas is demanding, it is crucial for enhancing national competitiveness.”

Experts view DPUs and ultra-small, ultra-low-power semiconductors as key weapons in the chips war because they can address the major challenge of power consumption, which is currently the most significant obstacle in both the semiconductor industry and the AI ecosystem.

The computational requirements and associated energy consumption in the semiconductor industry are increasing exponentially with the emergence of AI and big data.

For example, ChatGPT requires a large number of GPUs every time it runs, with an annual electricity cost exceeding 600 billion won ($450.5 million).

“A computer and power that is enough to run a nuclear power plant will be necessary to receive AI services just once moving forward,” Professor Kim Jang-woo said. He also warned that if this problem is not resolved, humanity’s technological advancement will come to a halt.

DPUs are anticipated to lead the AI semiconductor market after GPUs by accelerating data movement and processing. The market is projected to grow from an annual rate of 26.9 percent from 2021 to $5.5 billion by 2031.

Professor Kim Jang-woo, who is developing future semiconductor DPUs, is known as a DPU pioneer. He expressed confidence, saying that “Nvidia is also investing in DPUs, but we are ahead in this field.”

Kim also revealed a recent collaboration with AMD to develop MangoBoost software embedded in AMD DPU chips in preparation for the upcoming semiconductor competition.