Researchers in South Korea have developed the world’s first artificial intelligence (AI) semiconductor that operates at ultra-high speeds with minimal power consumption for processing large language models (LLMs), based on principles that mimic the structure and function of the human brain. The semiconductor successfully ran GPT 2 using only 1/625 of the power consumption and at 1/41 the size of Nvidia’s A100 graphics processing unit (GPU). This breakthrough is considered a key development in the escalating global AI semiconductor war.
The research team from the Korea Advanced Institute of Science and Technology (KAIST) PIM Semiconductor Research Center and the Graduate School of AI Semiconductor led by Professor Yu Hoi-jun developed this ultra-low power “complementary transformer” semiconductor using Samsung Electronics’ 28 nm process as announced by the Ministry of Science and ICT on Feb. 6.
Traditionally, operating GPT 2 required multiple GPUs and consumed 250 watts of power. However, the research team managed to process GPT 2 in just 0.4 seconds using a single AI semiconductor, measuring 4.5x4.5 mm, with only 400 milliwatts of power. The results of this study were presented and demonstrated at the International Solid-State Circuits Conference (ISSCC) held in San Francisco from Feb. 19 to 23.
The “complementary transformer” is a technology that allows selective use of “spiking neural networks” (SNNs), which mimic the way neurons in the brain process information, and “deep neural networks” (DNNs), AI models capable of recognizing complex patterns and learning through multiple layers of neural networks. It emulates the principle of human brain operation where energy consumption varies depending on the cognitive load. The technology minimizes power consumption by using DNNs for large input values and SNNs for smaller ones.
Professor Yu Hoi-jun of KAIST stated, “Neuromorphic computing, which mimics brain functions, is a technology that significant companies like IBM and Intel have not meaningfully implemented. We are proud to be the world's first to operate LLMs with a super-efficient neuromorphic accelerator (complementary transformer).”