**- Professor Ji-Hyun Yeom’s Team from the Department of Materials Science and Engineering**
– Development of quantum dots with directional, magnetic, and optical properties
– Results in up to 30% power savings when applied to neuromorphic devices that mimic the brain
[Herald Economy = Reporter Bon-Hyuk Gu] KAIST announced on the 25th that a research team led by Professor Ji-Hyun Yeom from the Department of Materials Science and Engineering has developed CFQDs, a type of quantum dots that possess chirality and magnetism, reacting asymmetrically to light for the first time in the world. They successfully implemented ChiropS, an AI neuromorphic device that mimics the structure and workings of the low-power human brain.
The photo synaptic transistor utilizing the chiral quantum dots developed by Professor Yeom’s team integrates various functions, such as polarization distinction, multi-wavelength recognition, and electric erasure, into a single device. It is a core technology for implementing high-speed, intelligent, low-power AI systems and can be applied to optical encryption, secure communications, and quantum information processing in the future.
Chiral magnetic quantum dots are synthesized by introducing chiral organic substances L- or D-cysteine into Ag2S-based inorganic nanoparticles. They have the characteristic of reacting differently according to the polarization direction (circular polarization) of light.
The research team fabricated a synapse transistor structure by stacking an Ag2S layer with chiral magnetic quantum dots and an organic semiconductor pentacene on silicon. This device shows long-term memory characteristics (LTP) when exposed to light and has an electric erasure function that resets with electric pulses, successfully creating an artificial mechanism that learns and adapts like the brain.
Additionally, when short light pulses (laser light) are repeatedly applied, a gradually increasing multi-level state is formed due to accumulating current, indicating that synaptic weight adjustment for AI learning, similar to the brain, and multiple learning are possible.
The research team produced a 2×3 device array and confirmed that the response current of each device was distinctly separated when light of different polarizations and wavelengths was applied. The system can detect and process up to 9 pieces of information in parallel through 6 channels, yielding at least 9 times more information processing than existing methods.
Moreover, the device can operate with up to 30% less power compared to existing computing technologies.
Professor Ji-Hyun Yeom emphasized, “Since a single device can process multiple polarizations and wavelengths and also integrate the function of resetting via electric signals, it could be an innovative platform for realizing low-power, high-precision AI systems.”
The research results were published in the international journal ‘Advanced Materials’ on April 7.