A semiconductor device that can show significant potential in neuromorphic applications has been developed by domestic researchers, and its wide-ranging use in various fields of artificial intelligence (AI) hardware is expected in the future.

Professor Kwak Jun-young's team from the Department of Convergence Electronic Semiconductor Engineering at Ewha Womans University and Professor Kang Ki-beom's team from the Department of Materials Science and Engineering at the Korea Advanced Institute of Science and Technology recently announced that they have jointly developed a zirconium oxide (ZrO2) memristor-based switching element for low-power neuromorphic computing.

This research is expected to play an important role in various application fields such as next-generation artificial intelligence neuromorphic computing, probabilistic computing, and reservoir computing.

Memristor-based switching elements have the characteristic of switching to a low resistance state when power is applied and returning to a high resistance state when power is turned off, making them essential for implementing computing hardware.

On the other hand, existing volatile memristor devices have limitations in practical applications due to the difficulty in controlling ion movement and low reliability during repeated operations.

To address this, the research team presented a new technology to precisely control the formation of conductive filaments using the crystal structure of zirconium oxide.

By introducing a rapid heat treatment process, the complex multilayer structure was replaced by forming a filament path in a single oxide layer, and the durability of the device was significantly improved by stably providing a migration path for silver (Ag) ions.

Crystallized ZrO2-based memristors have demonstrated excellent switching performance with high reproducibility, fast switching speed, and low-power operation.

This technology is also effective in complex pattern recognition tasks.It enables rational analysis and holds significant promise, especially in neuromorphic applications such as reservoir computing.

This study is expected to have wide-ranging applications in various fields of next-generation artificial intelligence hardware, and has confirmed excellent performance in voice and image recognition.

In speech recognition simulations, it achieved a high accuracy of 97.4%, suggesting that it could be expanded to various metal oxides in the future.

The results of this study were published in the internationally renowned academic journal 'InfoMat' as a cover paper with the title 'Crystallinity-controlled volatility tuning of ZrO2 memristor for physical reservoir computing'.

Meanwhile, this research was conducted with the support of the Next-Generation Intelligent Semiconductor Technology Development Project of the National Research Foundation of Korea and the KIST Institutional Project.