DGIST Professor Hyeokjun Kwon's Team Develops AI Semiconductor 'Resembling the Human Brain' ... Simultaneous Computing and Memory Processing

'10,000 Times Faster Than Humans'

Published in a World-Renowned Academic Journal

The research team led by Professor Kwon Hyuk-jun from the Department of Electrical Engineering and Computer Science at DGIST (President Lee Geon-woo), with Song Chong-myeong, a master's and integrated Ph.D. student, as the first author, has developed next-generation AI semiconductor technology that mimics the human brain by enhancing the efficiency of artificial intelligence and neuromorphic systems.

With the advancement of artificial intelligence technology, there is a rapidly growing demand for semiconductor technology that is both energy-efficient and capable of high-speed operation. However, existing computing devices have a 'von Neumann architecture,' where the processing unit and memory are separate, causing bottlenecks during data processing that reduce speed and energy efficiency.

Biological neurons possess the characteristic of performing computation and memory functions simultaneously, drawing attention to research on neuromorphic devices that imitate this feature.

Professor Hyeokjun Kwon (left) and Song Chongmyeong, an integrated master's and doctoral program student at DGIST.

In response, Professor Kwon Hyuk-jun's team developed a synaptic field-effect transistor using hafnium oxide with strong electrical properties and thin layers of tin disulfide. This is a three-terminal neural network device capable of storing multi-level data in a manner similar to neurons.

Through this research, the team not only successfully implemented biological characteristics such as short-term and long-term properties but also developed an ultra-high-efficiency device that operates 10,000 times faster than human synapses while consuming very little energy.

Professor Kwon Hyuk-jun from the Department of Electrical Engineering and Computer Science stated, "This research is an important stepping stone for next-generation computing architectures that require low power consumption and high-speed computation by developing high-performance neuromorphic hardware using next-generation materials such as two-dimensional channels and ferroelectric hafnium oxide." He added, "It is expected to be applied to various applications incorporating AI and machine learning in the future."

This research was conducted with Song Chong-myeong, a master's and integrated Ph.D. student, as the first author and Professor Kwon Hyuk-jun as the corresponding author. The research results were published online on February 20 in the world-renowned journal Advanced Science (IF: 17.521).

It was also supported by the Ministry of Science and ICT's Basic Individual Research Program, the Next-Generation Intelligent Semiconductor Technology Development Project, and the Nanomaterial Technology Development Project.

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