Latest Research News
Development of Artificial Synaptic Semiconductor Device Based on New 2D Materials
- Date : 22-10-10
- Views : 465
- Utilization of new two-dimensional materials as thin as a single atomic layer- Development of semiconductor devices that operate at low energy like human synapses
The Korea Institute of Science and Technology (KIST, President Yoon Seokjin) announced that the research team of Dr. Joon Young Kwak at the Artificial Brain Convergence Research Center developed a new two-dimensional insulator material synthesis technology with a new element composition ratio, as well as a high-performance and low-power artificial synaptic semiconductor device using this new material, through joint research with Professor Ki-beom Kang's team at the Korea Advanced Institute of Science and Technology (KAIST) and Dr. Taek-mo Jeong's team at the Korea Research Institute of Chemical Technology (KRICT).
With the recent increase in the proportion of video and image data, the processing of unstructured data is drawing attention as a key factor in the development of future artificial intelligence (AI) systems. In line with this trend, to overcome the excessive power consumption and limited information processing performance of the current widely used von Neumann computing structure, a “neuromorphic system” that can process and learn information with high efficiency and low power consumption is emerging as a next-generation semiconductor system.
Neuromorphic systems mimic the human brain to increase computing performance while reducing power consumption. To implement this, it is necessary to develop high-performance next-generation semiconductor devices that can precisely simulate “synapses” that regulate the connection strength between neurons according to the input signals. Silicon-based semiconductor devices, which are predominantly used at present, consume much more energy than biological synapses, and have physical limitations in simulating a highly integrated system similar to a real nervous system.
For this reason, research is actively being conducted to realize high-performance artificial synaptic devices by applying the properties of materials such as oxides and organic/inorganic materials. In addition, newly emerging two-dimensional materials are very thin at the atomic level, which gives them a great advantage in high integration of semiconductor devices. They have superior performance compared to existing silicon materials, such as fast switching speed and charge transfer speed, due to their unique characteristics.
The joint research team developed a synaptic device based on a new 2D insulator material and a heterojunction structure of a 2D semiconductor, enabling electrons to move efficiently even at low energy. Using these physical characteristics, they succeeded in developing an artificial synaptic device that shows uniform synaptic connection strength change and operates with an energy of about 15 fJ, which is similar to the actual energy consumption of human synapses. In addition, synaptic connection strength can be maintained for a short or long time depending on the number and intensity of external stimuli, enabling more precise simulation of human brain functions. The research team attempted artificial intelligence learning based on the developed high-performance two-dimensional artificial synaptic device, and the classification accuracy of handwritten digit image data (MNIST) was about 88.3%, confirming the possibility of application to actual neuromorphic systems.
Dr. Kwak of KIST said, “As the importance of research on high-efficiency new materials that can be used as substitutes for silicon in the development of next-generation semiconductors is growing, synaptic devices based on the heterojunction structure of semiconductors and the new two-dimensional insulator material presented in this study should have excellent competitiveness in implementing high-level neuromorphic hardware that can accurately simulate brain behavior."
This research was carried out with the support of a KIST institutional research program, the Next-Generation Intelligent Semiconductor Technology Development Project of the National Research Foundation of Korea, and the New Concept PIM Semiconductor Leading Technology Development Project of the National Institute of Information and Communications Technology Evaluation. The research results were published in the latest issue of the international journal Advanced Materials (IF: 32.086).
[Core Figure (Main)] Characteristics of the low-power, high-performance artificial synapse (left) and image classification learning accuracy test (right) of the new 2D-material-based artificial synaptic device developed by the research team.
[Reference figure] Synthesis technology developed by the research team and structure and analysis of the new 2D material.