AI Semiconductor Resembling Human Brain Reduces Power and Increases Performance

[Financial News] An AI semiconductor that learns with low energy like a human brain and responds autonomously has been developed by a domestic research team. This device significantly reduces power consumption compared to existing AI chips and can process information more precisely, making it a crucial turning point for next-generation AI technology.
On the 14th, according to POSTECH (Pohang University of Science and Technology), a POSTECH research team has created the world's first '3-terminal ECRAM device'. Utilizing hexagonal tungsten trioxide (h-WO₃) single crystal nanowire with atoms arranged in an orderly manner, this material is hundreds of times thinner than a human hair, allowing current to flow stably and consistently. It implements a learning method by creating three electrodes that control and flow current, exchanging signals in multiple directions like actual brain neurons.
In particular, this device increases its conductivity when receiving repetitive electrical stimuli (pulses), which is very similar to the 'integrate-and-fire' mechanism where neurons fire information above a certain stimulus threshold. Previously, the function of integrating and firing signals of 'neurons' and the function of adjusting signal strength and learning of 'synapses' had to be implemented with separate circuits, but they succeeded in simultaneously implementing them in a single ECRAM device.
Currently, AI technology is rapidly advancing, but massive power consumption is a major issue. Recently, research on 'neuromorphic computing' technology, which mimics the structure of the actual brain, has been actively conducted. However, existing 'ECRAM1)' devices mainly used amorphous structures with disorderly arranged materials, limiting precise operation like actual brain neurons.
Professor Kim Se-Young of POSTECH said, "This research will significantly enhance the integration and energy efficiency of neuromorphic hardware, reducing the circuit complexity of AI semiconductors and becoming a turning point for implementing efficient computing systems like the brain."
This research, conducted by Professor Kim Se-Young of the Department of Materials Science and Engineering and Semiconductor Engineering at POSTECH, and master's student Lee Jun-Yong, was published in the international journal 'Small' in the field of electronic materials. It was also supported by the Ministry of Trade, Industry and Energy's public-private joint investment semiconductor advanced manpower training project for the development of high-performance ECRAM compatible with CMOS processes for implementing high-density storage class memory and deep learning accelerators, and the development of neuromorphic and in-memory computing chip implementation technology through co-optimization of the Tiki-Taka algorithm and high-performance synapse devices. 

jiany@fnnews.com Reporter Yeon Ji-an