A brain-like chip interprets 'neural network connectivity' in real time

The ability to analyze the brain's neural connectivity is emerging as a key foundation for brain-computer interface (BCI) technologies, such as controlling artificial limbs and enhancing human intelligence. To make these analyses more precise, it is critical to quickly and accurately interpret the complex signals from many neurons in the brain.

Dr. Jongkil Park and his team of the Semiconductor Technology Research Center at the Korea Institute of Science and Technology (KIST) have presented a new approach that mimics the brain's learning principles. The team engineered the principle of spike-timing-dependent plasticity (STDP), in which the brain adjusts the strength of connections based on the order of signal firing between neurons. This allows them to learn the connectivity in a brain's neural network in real-time without having to store the activity of all the neurons.

Conventional techniques involve storing neuronal activity data for a prolonged period and then using statistical methods to calculate the connection between neurons. This method requires enormous computation and time delays as the size of the neural network grows, making real-time analysis virtually impossible in environments where numerous signals occur simultaneously, such as the brain.

KIST researchers have devised a new learning algorithm that can significantly reduce the large memory required for hardware implementation of STDP. By eliminating the memory-consuming 'reverse lookup table', the technique enables STDP to be implemented in a scalable structure even on highly integrated neuromorphic hardware. As a result, the 'on-chip learning-based neuromorphic system' achieved up to 20,000 times faster processing speed while maintaining similar interpretation accuracy to existing conventional techniques.

Neuromorphic engineering is a new generation of artificial intelligence semiconductors that mimic the brain's neural network structure and learning behavior to emulate human cognition, and is a strategic area of investment for major industrialized nations such as the United States and Europe to secure technological hegemony. However, commercialization has been difficult due to the lack of a specific application field, or 'killer application,' that can be useful like the brain. In this situation, KIST researchers' 'real-time brain neural connectivity analysis' technology is an example of demonstrating the practical application of neuromorphic engineering, and is an important achievement that marks a turning point in the commercialization of next-generation AI semiconductors.

"This achievement marks an important turning point in the evolution of neuromorphic computing into a powerful tool for solving real-world problems," said Dr. Park Jongkil of KIST. "With its simple hardware structure and easy scalability, it can be applied to advanced AI fields such as autonomous vehicles and satellite communications by controlling devices with just a thought or copying specific brain functions, as well as analyzing complex sensor signals in real time, where time sequences and cause-and-effect relationships are critical."

###

KIST was established in 1966 as the first government-funded research institute in Korea. KIST now strives to solve national and social challenges and secure growth engines through leading and innovative research. For more information, please visit KIST’s website at https://www.kist.re.kr/eng/index.do

This research was supported by the Ministry of Science and ICT (Minister Bae Kyung-hoon) through the KIST Institutional Program and the National Research Foundation of Korea (2021R1A2C2092484). The results of this research were published in the latest issue of the international journal of IEEE Transactions on Neural Systems and Rehabilitation Engineering (IF: 5.2, JCR(%): 2.6 %).