Development of World-Class AI Semiconductor with Domestic Technology... Demonstration Planned for Second Half of the Year

[Asia Economy Reporter Seolgina Jo] A world-class artificial intelligence (AI) semiconductor has been developed using domestic technology. It is evaluated as a stepping stone for technological independence in the AI semiconductor field, a new battleground in the semiconductor industry.

On the 7th, the Ministry of Science and ICT announced that domestic companies including the Electronics and Telecommunications Research Institute (ETRI) and SK Telecom have jointly developed an AI semiconductor based on a neural processing unit (NPU) applicable to high-performance servers and Internet of Things (IoT) devices, and plan to promote demonstration through AI infrastructure and product application.

AI semiconductors refer to semiconductors that execute computations required by AI-based application services with high performance and high power efficiency. The newly developed NPU-based AI semiconductors include ▲ultra-low power AI semiconductors for servers and ▲visual intelligence AI semiconductors for mobile and IoT devices, which mimic the neural networks of the human brain to process large-scale computations simultaneously.

The Ministry of Science and ICT stated, “As high-performance and low-power AI semiconductors optimized for AI execution have emerged as next-generation core technologies determining future AI industry competitiveness, since 2016, proactive technology development has been pursued through national R&D involving domestic large corporations, SMEs, and ETRI. After years of research and development, independent design technology has been secured, and world-class AI semiconductor implementation has been successfully achieved.”

First, the ultra-low power AI semiconductor for servers, developed for the first time domestically, can be used in high-performance servers such as cloud data centers providing AI application services. ETRI and SK Telecom invested a total of 7.33 billion KRW from March 2017 to the end of last year for development. Currently, semiconductors used for AI computation consume a large amount of power and have large chip sizes, limiting efficient production and utilization; however, the research team minimized chip size while achieving high computational capability and power efficiency.

In particular, it is characterized by maximizing performance by highly integrating 16,384 computing cores on a small area the size of a coin (17mm x 23mm), and minimizing power consumption by applying software technology that can turn on and off the power of each computing core. Through this, it can process 40 trillion operations per second (40 TFLOPS) while consuming low power of 15 to 40 watts.

The Ministry of Science and ICT expects that applying this to cloud data centers will improve power efficiency for AI services by more than tenfold. The research team plans to demonstrate the developed chip in real environments and accelerate commercialization starting in the second half of this year by applying it to SK Telecom data centers that provide services such as intelligent CCTV and voice recognition.

Alongside this, the visual intelligence AI semiconductor for mobile and IoT devices recognizes objects like human vision and can be applied to intelligent CCTV and drones. It was developed in cooperation with ETRI, Korea Electronics Technology Institute (KETI), FA Linux, Nextchip, and AD Technology since April 2016, with a budget of 12 billion KRW.

This semiconductor features a minimized circuit area about half the size of an adult fingernail (5mm x 5mm) and achieves object recognition performance of 30 times per second with power consumption of 0.5 watts, which is less than one-tenth compared to existing semiconductors. The research team plans to promote demonstration and commercialization linked to AI-based intelligent device productization in fields such as video surveillance and reconnaissance starting in the second half of this year.

Minister of Science and ICT Choi Ki-young emphasized, “AI semiconductors are the core foundation for South Korea to leap from an ‘ICT powerhouse to an AI powerhouse’ in the AI era,” and added, “Independent AI semiconductor development will be an important challenge for innovation in the domestic AI and data ecosystem.” He continued, “Through public-private cooperation, we will establish an ‘AI Semiconductor Development Strategy’ and focus on nurturing AI semiconductors as a future innovation growth engine. We will begin full-scale investment this year in core AI semiconductor technologies such as innovative design and low-power new devices, and actively support challenging research leading the global market, including new concept semiconductor technology integrating memory and computation (PIM).”

This content was produced with the assistance of AI translation services.