Research

Research Highlights

Home > Research > Research Highlights

Research

Research Highlights

Research Highlights

Professor Yang-Kyu Choi’s Research Team Solved Computing Challenges with Neuromorphic Neural Networks

Professor Yang-Kyu Choi’s Research Team Solved Computing Challenges with Neuromorphic Neural Networks

 

images 000075 photo1.jpg 13

<(from left) Professor Yang-Kyu Choi, ph.d. candidate Seong-Yun Yun, Professor Joon-kyu Han from Sogang University (KAIST alumnus)>
 
 
Professor Yang-Kyu Choi’s research team has built a miniature oscillatory neural network using only silicon materials and processes currently used in the semiconductor industry, implementing an edge detection feature and solving the graph coloring problem*.

 

*Graph coloring problem: A term used in graph theory, requiring different colors to be assigned to each vertex of a graph. This is similar to assigning frequencies to broadcasting stations to prevent overlap and the creation of areas with poor reception, and is widely applied in various fields.

 

The research team announced on the 3rd that they have developed a neuromorphic oscillatory neural network that mimics the interactions of biological neurons using silicon varistor components.

 

With the arrival of the big data era, artificial intelligence technology has made significant progress. One of the neuromorphic computing methods, the oscillatory neural network (oscillatory neural network), is an artificial neural network that mimics the interaction of neurons. The oscillatory neural network uses the connection operations of oscillators, which are the basic units, and performs calculations using oscillations rather than the magnitude of signals, thus offering advantages in terms of power consumption.

 
 
images 000075 image1.jpg 12
 
< Figure 1. The oscillatory neural network using varistors and its applications >

 

 

The research team developed the oscillatory neural network using silicon-based oscillators. By connecting two or more silicon oscillators using capacitors, the oscillation signals interact with each other and synchronize over time. The research team implemented edge detection, a feature used in image processing, with the oscillatory neural network and solved one of the challenges, the vertex coloring problem.

 

Furthermore, this research has the advantage of being immediately applicable to mass production from a manufacturing perspective, as it built the oscillatory neural network using only silicon materials and processes currently used in the semiconductor industry, instead of complex circuits or materials and structures with low compatibility with existing semiconductor processes.

 

The research, led by Seong-Yun Yun, a doctoral student, and Professor Joon-Kyu Han from Sogang University, stated, “The developed oscillatory neural network can be used as neuromorphic computing hardware capable of calculating complex computing challenges, and is expected to be useful in resource allocation, new drug development, semiconductor circuit design, and scheduling,” highlighting the significance of the research.

 

The study, co-authored by Seong-Yun Yun and Professor Joon-Kyu Han, was published in ‘Nano Letters’, in its 24th volume, issue 9, on March 2024, and was selected as a supplementary cover article.

 
images 000075 image2.jpg 9
Photo Caption: < Figure 2. The image selected as a supplementary cover article for Nano Letters >

(Paper title: A Nanoscale Bistable Resistor for an Oscillatory Neural Network) (https://pubs.acs.org/doi/full/10.1021/acs.nanolett.3c04539). 

This research was conducted with the support of the Korea Research Foundation’s Next-Generation Intelligent Semiconductor Technology Development Project and the National Semiconductor Research Laboratory Support Core Technology Development Project.