Korea bets tomorrow belongs to AI chip leaders

Home > Business > Tech

Korea bets tomorrow belongs to AI chip leaders

Published: 24 May. 2022, 18:17 Updated: 24 May. 2022, 19:40

[JOONGANG PHOTO]

AI chips are the key to future semiconductor market dominance for Korean companies, academics and others argue.

Despite the country's strong position in memory chips and its growing presence in custom non-memory chips, they say that tomorrow belongs to companies that are best at designing and manufacturing high-performance processors, commonly called AI chips.

"Global chip companies are going full throttle on developing AI semiconductors," said Kim Hyung-jun, head of the Post-Silicon Semiconductor Institute under the Korea Institute of Science and Technology. "AI chips will play a crucial role in chip competitiveness in future."

AI chips are purposely built to perform very specific calculations as fast as possible. They are distinct from general-purpose processors, which are designed for a wider variety of functions, and are better suited for very specific, predetermined tasks that require impressive calculation capabilities and low energy consumption, such as voice recognition, autonomous driving, the metaverse and the Internet of Things.

The AI chip category is broad and the definition is somewhat flexible. It includes application-specific integrated circuits, graphical processing units (GPU), neural processing units (NPUs), which are similar but more powerful than GPUs, and neuromorphic chips, which are still largely experimental.

Introduced in December, the Electronics and Telecommunications Research Institute(ETRI)'s ArtBrain-K is capable of running 5,000 trillion computations per second. The system is powered by ETRI's AB9 high-performance AI chip, which is an NPU that emulates the neural networks of a human brain.

"ArtBrain-K has four times the computing capacity and seven times more energy efficiency compared to the previous GPU-based server," said Lyuh Chun-gi, head of ETRI's AI processor research team.

"A video processed through this system appears real-time, and the motion seems extra smooth," explained Lyuh.

The ArtBrain-K is going to be deployed in a facial recognition system at airports for automated passport control.

AI chips are becoming the linchpin of the global chip industry,

Over 50 companies worldwide are developing AI chips, according to Gartner. The global AI chip market will be $71.1 billion in size by 2025, more than double the market size of $34.3 billion in 2021

The Korea Information Society Development Institute forecasts that the global AI chip market size will reach $117.9 billion by 2030.

Santa Clara, California's Nvidia, which has 90 percent of the GPU market, is the clear frontrunner so far. In 2018, the company introduce computing hardware powered by its Xavier System-on-a-Chip, which is designed to support autonomous driving. The company also revealed its plan to build a $20-billion chip plant in Columbus, Ohio, to manufacture next-generation chip products.

Nvidia recently released the Gaudi 2, a second-generation deep learning processor. It has double the computation speed of the previous model.

Other big tech players, such as Google, Tesla and Huawei, have developed AI chips.

Korean chipmakers, which have been more focused on memory chips than processors, are eyeing the AI chip market.

Samsung Electronics is developing neuromorphic computing technology. Neuromorphic computing, or neuromorphic engineering, aims to develop integrated circuits mimicking the human nervous system. The company's goal is to develop a state-of-art memory chip that supports both data storage and processing functions. The company also took the processing speed of its mobile application processor Exynos 2100 up a notch by incorporating a high-performance NPU in the chip.

SK hynix has established Sapeon, a chip company, with SK Telecom and SK Square. Sapeon revealed its first AI chip for data centers, the Sapeon X220, in 2020, and plans to release the X330, which supports both inference and training, next year. Local semiconductor start-up FuriosaAI was recognized for an AI chip that performs better than Nvidia's product during a global AI chip event.

The Moon Jae-in administration's plan was to invest over 1 trillion won ($792 million) in the research and development of intelligent semiconductors through 2029, and pour an additional 402.7 billion won into processing-in-memory chips, where processing and memory are integrated.

President Yoon Suk-yeol also made semiconductors top priority, contemplating a wide range of support for chip education and enhanced support for chipmakers.

Experts point out that weak academia-industry relationships are hampering technological advances.

"The level of technological research in universities and research institutions is high," said Yoo Hoi-jun, an engineering professor at KAIST. "But Korea is lagging behind China and the United States in the commercialization of such technologies."

The semiconductor production process, which encompasses technological development by universities, product design by companies and manufacturing by foundries, is not flowing smoothly in Korea, according to Yoo. The disconnection between research institutions and corporations is most prominent.

"In China, the government is taking an active role in AI chip development," said Yoo, "And since the U.S. AI chip market was created early on, the investments there are immense."

Korea's investments in the AI chips is about one-tenth that of Silicon Valley's, said Yoo.

"Above all, we need a support system that allows anyone with a business idea to develop and market AI chips," argued Yoo.

Post-Silicon Semiconductor Institute Director-General Kim also emphasized the importance of industry-university collaboration.

"The cooperation system where research institutions, universities and corporations collectively carry out large-scale projects is needed in order to develop the source technology," said Kim.

BY CHOI EUN-KYUNG [shin.hanee@joongang.co.kr]