Hydrogen Instead of Oxygen... Changing the Operating Principle of AI Semiconductors [Reading Science]

DGIST Develops Neuromorphic Device
Capable of Simultaneous Processing and Memory Storage

Daegu Gyeongbuk Institute of Science and Technology (DGIST) has developed an artificial intelligence (AI) semiconductor device capable of self-learning and memory, utilizing the movement of hydrogen ions instead of the conventional 'oxygen vacancy (defect)' principle that has been central to traditional memory semiconductors. Unlike conventional resistive memory devices that rely on defect migration, this new device precisely controls hydrogen ions electrically, presenting a new possibility for next-generation neuromorphic semiconductors.

DGIST announced that a research team led by Senior Researcher Lee Hyunjun and Research Fellow Noh Heeyeon at the Division of Nano Technology has developed a 'two-terminal-based AI semiconductor device' that controls the injection and extraction of hydrogen ions (H+) using electrical signals.

Schematic diagram of a two-terminal hydrogen-controlled artificial intelligence semiconductor device. When a positive (+) voltage is applied, hydrogen moves from the supply layer to the semiconductor, increasing conductivity. When a negative (-) voltage is applied, hydrogen moves back to the storage layer, decreasing conductivity. This process varies according to the strength and frequency of the voltage, enabling the implementation of analog-type multi-level resistance. Provided by the research team

Recently, artificial intelligence has required rapid processing of massive datasets, but the traditional computer architecture separates processors, which handle calculations, from memory, which stores data. This separation causes significant slowdowns and increased power consumption during data transfers. To address this, neuromorphic semiconductors, which can perform processing and storage simultaneously like the human brain, are attracting attention as next-generation AI hardware.

Precise Control of Hydrogen Ion Movement via Electric Field

Until now, oxide-based memory devices have typically adjusted conductivity and stored information by utilizing the movement of oxygen vacancies (defects). However, this approach has limitations in ensuring long-term stability and uniformity between devices.

The DGIST research team instead focused on lightweight and highly mobile hydrogen ions. When an electric field is applied, hydrogen ions are injected into the semiconductor, increasing conductivity. When the voltage is reversed, hydrogen ions exit, decreasing conductivity. By precisely controlling this hydrogen movement, the team implemented a new memory operation mechanism that adjusts electrical resistance and stores information.

In handwritten digit recognition experiments using this device, the research team achieved over 97% recognition accuracy, confirming its potential for practical use as AI hardware.

Research team photo. From left to right: Hyunjun Lee, Heeyeon Noh, Shinbum Lee, Myungjae Lee from DGIST, and Jiyong Woo from Kyungpook National University. Provided by DGIST.

High-Density 'Two-Terminal Vertical Structure'... Potential for Neuromorphic Chips

This technology is especially significant because it has been realized in a 'two-terminal vertical structure' that is both highly integrable and simple to manufacture. While this structure is advantageous for high-density AI chips, there have been few examples of precisely controlling hydrogen movement in a vertical structure to enable AI functionality.

The device developed by the research team operated stably in more than 10,000 repeated cycles and maintained its memory state for over 1 million seconds even when powered off. In addition, its gradual analog conductivity transitions enabled learning and memory functions similar to synapses in the human brain.

Lee Hyunjun, Senior Researcher at DGIST's Division of Nano Technology, stated, "This research is meaningful in that it proposes a resistive switching mechanism based on 'hydrogen migration,' which is different from conventional oxygen vacancy-based memory," adding, "It presents a new direction for the development of next-generation low-power AI semiconductors."

Noh Heeyeon, Research Fellow at DGIST's Division of Nano Technology, commented, "This is a case of electrically controlling hydrogen atoms moving between stacked semiconductor layers with high precision," and noted, "Hydrogen-based memory devices have potential to evolve into highly efficient neuromorphic semiconductor technology in the future."

This research was supported by the Mid-Career Researcher Program of the Ministry of Science and ICT and the National Research Foundation of Korea, as well as by DGIST's institutional research program. The teams of Professor Lee Shinbum and Senior Researcher Lee Myungjae at DGIST, and Professor Woo Jiyong at Kyungpook National University participated as co-researchers. The results were selected as a cover paper in the international journal ACS Applied Materials & Interfaces, which specializes in materials and interfaces.

This content was produced with the assistance of AI translation services.