Ajou University Research Team Implements Core Technology for Random Number Generation Device

Ajou University announced on August 6 that a joint research team led by Professor Lee Hyungwoo of Ajou University has newly proposed a so-called two-level quantum system (TLQS) that minimizes interference from external variables in the discrete fluctuation of tunneling current.

This research was published last month in the global academic journal 'Nature Communications' under the title "Highly stable two-level current fluctuation in complex oxide heterostructures."

Kim Doyeop, a graduate student from the Department of Energy Systems at Ajou University, and Lee Jungwoo, a professor at Hongik University, participated as co-first authors. The research was led by Lee Hyungwoo, professor at Ajou University (Department of Physics and Graduate School of Energy Systems), Eom Gitae, professor at Gachon University (Department of Semiconductor Engineering), and Lee Sunwoo, professor at Inha University (Department of Computer Engineering), who served as corresponding authors. The research also involved the team of Professor Tula R. Paudel from South Dakota School of Mines and Technology in the United States and the team of Professor Yang Yongsoo from KAIST.

The joint research team implemented a highly stable two-level current fluctuation phenomenon using a complex oxide heterostructure (SrRuO3/LaAlO3/Nb:SrTiO3, SRO/LAO/Nb:STO) and developed a physical entropy source based on this phenomenon.

A random number refers to a number generated randomly within a defined range so that the next value cannot be predicted. Random numbers are essential in various fields such as encryption, security, simulation, and gaming to produce unpredictable values. In addition, random number data are crucial for big data training in artificial intelligence (AI) and machine learning. The use of high-quality random numbers is now considered essential for the efficient training of recent machine learning models.

In particular, a physical entropy source is a device that generates true random numbers, which are impossible for humans to predict or hack, by utilizing natural random phenomena. Conventional computers use software-based pseudo-random number generators (pseudo-random numbers), but physical entropy sources provide fundamentally higher security and reliability as they cannot be predicted or hacked. Furthermore, in the field of neuromorphic systems, which mimic the structure of the human brain, hardware-based physical entropy sources are indispensable for building hardware-based artificial neural networks, surpassing the limitations of software-based algorithms.

Existing systems based on random telegraph noise (RTN), which represent two-level current fluctuation phenomena, utilize the charge trapping of point defects within oxides. However, this phenomenon is closely related to the external environment, making it very unstable and difficult to maintain stable discrete signals?signals composed of only two types of values, such as 0 and 1?for extended periods.

To solve this problem, the research team intentionally induced the coexistence and interaction of two types of point defects?oxygen vacancies (VO) and Ti antisite defects (TiAl). Through this, they newly proposed a so-called two-level quantum system (TLQS) that minimizes interference from external variables in the two-level fluctuation of tunneling current.

When electrons are temporarily trapped in TiAl defects, the energy level of the surrounding oxygen vacancies (VO) changes instantaneously, which in turn induces discrete fluctuations in the tunneling current. This structure exhibited stable discrete current fluctuations lasting more than 169 seconds at room temperature and operated stably for over one year.

Furthermore, the research team verified the random number generation functionality by utilizing the discrete fluctuation characteristics of the two-level quantum system (TLQS) current signals. By binarizing the experimentally obtained analog current data into random sequences of 0s and 1s and evaluating their randomness, they demonstrated that the proposed TLQS can indeed generate excellent random number data.

The joint research team applied the random number data generated by TLQS to image super-resolution (VDSR) neural network training. They confirmed that the model using TLQS-based random number data achieved higher accuracy and faster training speed compared to models using conventional software-based random number generators (Numpy Random Generator, Python).

Image super-resolution (VDSR) neural networks are artificial intelligence technologies that restore blurry photos to clear images. To perform such advanced tasks, AI must be trained in advance by repeatedly showing numerous sample images and matching the correct answers, a process in which random numbers play a crucial role. Random numbers are used to shuffle the order of training data or to randomly determine the initial state of the neural network, enabling AI to learn more efficiently and function correctly in various situations. If the random numbers are predictable or biased, AI may produce distorted results, making the use of true random numbers extremely important.

Lee Hyungwoo, the professor at Ajou University who led this research, explained, "The two-level quantum system (TLQS) device we have implemented is compatible with silicon-based semiconductor technology (CMOS) used in computers and smartphones." He added, "Considering its excellent potential for device integration, this is a highly practical source technology for random number generators."

This research was supported by the G-LAMP project, the Mid-career Researcher Program, and the Basic Research Laboratory (BRL) Program of the National Research Foundation of Korea.

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