"For small and medium-sized developers developing AI chips, it will be crucial to focus on the edge."
Edge-to-edge movement is actively driving personalized AI and resource efficiency.
The market for solutions that combine AI and semiconductors will continue to grow.
Many industry experts have expressed positive views on the vision of AI semiconductors at the edge. The edge AI market is poised for a full-blown boom, driven by the explosion of data, electrification of mobility, and increasing demand for low power.

On the 9th, the Seoul National University System Semiconductor Industry Promotion Center held the System Semiconductor Symbiosis Forum at the Seoul National University Faculty Hall. Under the theme, "K-Semiconductor Strategy: Looking Back on 2022 and Looking Ahead to 2023," the forum explored various industry trends, including next-generation semiconductor technologies and CES 2023.

On this day, the forum was attended by △Lim Ha-ni, CEO of Roa Intelligence, △Professor Kim Dong-sun of Sejong University, and △Vice President Lee Gyu-bok of the Korea Electronics Technology Institute and gave presentations.

Next-Generation Semiconductors: AI on the Rise

CEO Lim Ha-ni looked at big tech companies developing their own chips. She mentioned Google, which has developed the TPU, an AI processor specialized for running TensorFlow; Apple, which is developing an Apple chip; and the Trainium chip, which specializes in machine learning.

They claimed that the reasons for their decision to develop their own chips were service differentiation, chip supply shortages, and the complexity of the semiconductor market's supply chain. “We believe there is room for big tech companies to exploit gaps in the semiconductor supply chain,” he said, adding, “As generative AI increases, computing demand will increase, which will increase demand for AI inference chips.”

Professor Kim Dong-soon explained, “The domestic industry is also at a turning point to become more competitive through AI semiconductors,” and “With the growing demand for highly personalized artificial intelligence and increased efficiency in network and computing resources, cloud and AI service models at the edge are being actively created to improve the efficiency of big data operations.”

In addition, with electrification of mobility becoming a major trend, he emphasized that low-power, lightweight, AI-supported system semiconductors are emerging as next-generation core technologies to capture the edge and mobility.

Edge AI, a growth engine for fabless companies

CEO Lim Ha-ni emphasized in her Edge AI focus strategy, “We need to find a service-level edge by focusing on specific usability in specific applications, not just the edge at the device level.” He said that it is important to keep in mind that Edge AI is integrated in terms of functionality, citing the fact that ByteDance, the maker of TikTok, also makes its own chips.

Professor Kim Dong-jun also said, “Intel, Qualcomm, and others are paying attention to edge NPU technology,” and predicted, “Edge AI is a field where domestic fabless companies can achieve success, and full-scale market maturity is predicted to take about 3 to 4 years, and the peak is expected to take about 10 years.”

Vice President Lee Gyu-bok also said, “It is more effective to target the edge, and AI is needed at the edge, such as IoT and smart factories,” and urged people to pay attention to John Deere’s autonomous tractor, which attracted attention at CES 2023.

While the central computing that controls autonomous driving will likely use chipsets from global companies like Nvidia, the edge of the car will house numerous modules, making edge AI inevitable to reduce central computing resources and achieve low power consumption.

John Deere's seed drills use sensors to control seeding and fertilizer application, but the vice president added, "These tractors have dozens of individual modules, and it needs to be addressed with cameras and edge devices."

A representative of a fabless company who attended the forum said, “It is difficult for small and medium-sized companies to approach AI semiconductors,” and added, “Big tech companies such as Coupang, Baedal Minjok, and Naver are the ones using AI the most, so linking up with demand companies like these is very important.”

AI chips lack a general-purpose solution market, and are often developed by big tech companies for their own consumption needs. To revitalize the domestic system semiconductor ecosystem, a win-win strategy, such as a project to develop AI chipsets for demand-side companies through collaboration between big tech and fabless companies, appears necessary.