한국재료연구원(KIMS, 원장 최철진) 에너지·환경재료연구본부 김용훈 박사 연구팀이 포항공과대학교 황현상 교수 연구팀와 공동 연구를 통해 3차원 초고집적 뉴로모픽 시냅스 소자를 구현하는 데 성공하며, 향후 2차원 소재를 이용해 3차원의 인공지능 반도체를 상용화 할 수 있는 가능성을 열었다.

▲Schematic diagram of a vertical-responsive ECRAM synapse semiconductor device (left) and a three-dimensional multi-vertical-responsive ECRAM synapse semiconductor (middle), and a transmission electron microscope (TEM) cross-sectional image (right)
Materials Research Institute, World's First Development of 3D Ultra-High-density Neuromorphic Synapse Device
The Korea Institute of Materials Science (KIMS, President Choi Cheol-jin) has succeeded in producing an ion barrier layer without a transfer process by grafting a synthetic method that directly grows two-dimensional materials at low temperatures below 200℃, opening up the possibility of commercializing three-dimensional artificial intelligence semiconductors in the future.
The Korea Institute of Materials Science announced on the 1st that the research team led by Dr. Yonghoon Kim of the Energy and Environmental Materials Research Division succeeded in implementing a 3D ultra-high density neuromorphic synaptic device through joint research with the research team led by Professor Hyunsang Hwang of Pohang University of Science and Technology.
This technology is a technology that achieves three-dimensional ultra-high integration of vertical-responsive ECRAM (Electrochemical random-access memory) devices by simultaneously applying a van der Waals two-dimensional nanomaterial (tungsten nickel selenide, WS2) synthesized at low temperatures (<200°C) to the drain electrode and ion barrier layer.
In order to implement synaptic elements, a technology is needed to precisely control the movement of ions, but if this control is difficult, it is difficult to obtain the desired resistance value.
Our research team controlled ion movement through a barrier layer made of two-dimensional nanomaterials between the channel and the ion electrolyte layer.
As a result, high-performance synaptic characteristics such as linearity, symmetry, and durability could be implemented.
We also confirmed that we achieved a high handwriting (MNIST, Modified National Institute of Standards and Technology) pattern recognition accuracy of 95.22%.
Previously, two-dimensional materials such as graphene were used to produce a barrier layer to control ion movement in the electrolyte layer in ECRAM devices.
This required a transcription process, making it difficult to implement high-density and high-yield synaptic semiconductor devices.
To improve this, our research team successfully created an ion barrier layer without a transfer process by adopting a synthetic method that directly grows two-dimensional materials at low temperatures below 200℃.
In addition, we verified the independent weight update characteristics of the multi-stacked 3D vertical-responsive ECRAM, demonstrating its potential for fabricating ultra-high-density 3D-based synapse arrays.
The research team succeeded in developing key materials and process technologies to perform large-scale calculations required for implementing artificial intelligence semiconductors at ultra-high speed and ultra-low power through the development of 3D ultra-high-density neuromorphic synapse devices.
“Research on van der Waals two-dimensional semiconductor materials has been ongoing for over 10 years, but we have not yet reached commercialization,” said Yonghoon Kim, a senior researcher at the Materials Research Institute and the principal investigator. “Based on the results of this study, we will continue our research on large-area synthesis of two-dimensional semiconductor materials, commercialization of electronic devices, and application to semiconductor processes.”
This research was conducted through the basic project of the Korea Institute of Materials Science and the Leader Research of the National Research Foundation of Korea with the support of the Ministry of Science and ICT. In addition, the results of this study were published in the world-renowned academic journal 'Advanced Functional Materials (IF: 19.0/ First author: PhD candidate Lee Kyu-min, research student Hwang Seung-kwon)' published by Wiley.