KAIST (Korea Advanced Institute of Science and Technology) announced that Dr. Kim Kyung Min’s research team has developed a groundbreaking semiconductor device called the “Neuransistor,” which mimics the characteristics of brain neurons.
Neuransistor combines “neuron” and “transistor,” representing a next-generation semiconductor device that operates on principles similar to how brain neurons process time-varying information.
Conventional digital transistor-based computers process input data sequentially, one piece at a time. This approach leads to increased computational steps and repeated memory access when handling time-series data, such as videos, resulting in slower processing speeds.
The Neuransistor operates by considering both the polarity (negative or positive) of voltage signals and the precise timing of signal inputs.
Unlike traditional transistors that only handle static signals of 0 or 1, the Neuransistor dynamically processes information by accounting for the timing, magnitude, and polarity of excitatory or inhibitory signals, much like actual brain neurons.
By incorporating signals’ timing and dynamic characteristics, the Neuransistor can process input data in parallel and instantaneously, demonstrating powerful performance in handling time-series data.
The research team controlled the responsiveness of the semiconductor device using a two-dimensional electron gas (2DEG) formed in a layered structure of two oxides, titanium oxide and aluminum oxide.
This innovation enabled the development of neuron-like semiconductor devices that selectively implement excitatory or inhibitory responses based on the polarity of the gate voltage.
Notably, the team successfully implemented a masking function that converts input signals into multi-dimensional outputs without complex preprocessing, facilitating the construction of a Neuransistor-based computing model known as Liquid State Machine (LSM).
Experimental results showed that the Neuransistor-based LSM achieved over ten times the data processing performance compared to traditional methods and had a faster learning rate.
Dr. Kim emphasized, “Implementing a signal processing method similar to the human brain in a semiconductor device is a significant breakthrough. This technology will be crucial in various fields, including brain-inspired artificial intelligence and predictive systems.”
This groundbreaking research was published in the prestigious international materials science journal Advanced Materials.
