Thermally Controlled Conductive Filaments Enable Both Hardware Security and In-Memory Computing on a Single RRAM Array

Research Published in the Prestigious International Journal Advanced Functional Materials

A research team led by Professor Kim Sungjun and Professor Hwang Sungmin of the Department of AI Semiconductor Engineering at Korea University Sejong Campus has developed a next-generation resistive random-access memory (RRAM)-based semiconductor device capable of implementing both hardware security through Physically Unclonable Functions (PUFs) and in-memory computing (IMC) within a single memory array.

The research represents a significant advancement for next-generation edge AI and Internet of Things (IoT) semiconductors, where both robust security and high computational efficiency are essential. Conventional RRAM-based security devices utilize the intrinsic randomness of memory cells to generate security keys. However, irreversible damage caused during the electrical forming process has limited the reuse of the same memory array for high-precision computing. In contrast, in-memory computing requires highly uniform devices and precise conductance control, making the integration of hardware security and in-memory computing within a single device a longstanding technical challenge.

To overcome these limitations, the research team proposed a Thermally Pre-formed Filaments (TPFs)-RRAM platform, which replaces conventional electrical forming with a thermal annealing process that induces oxygen migration inside the device to create conductive filaments—the microscopic current pathways responsible for resistive switching. Rather than partially damaging the device using high-voltage electrical forming, the proposed approach employs thermodynamic reactions to freely reconfigure device functionality without causing physical degradation.

The research findings were published in Advanced Functional Materials (Impact Factor: 19.9; top 4.2% in the Journal Citation Reports). The paper “Thermally Pre-Formed Reconfigurable Resistive Random-Access Memory Crossbar Arrays: A Dual-Mode Platform for Robust Physically Unclonable Functions and In-Memory Computing” (Advanced Functional Materials, 2026, e76481), was co-first-authored by graduate students Kwon Seongbin and Lee Taewon from the Department of AI Semiconductor Engineering. The study presents a next-generation secure AI hardware platform capable of supporting both hardware security and in-memory computing within a single RRAM array.

The team successfully generated PUF-based security responses using an RRAM crossbar array and demonstrated compliance with the National Institute of Standards and Technology (NIST) SP 800-22 randomness test suite. After completing the PUF operation, the same array was reconfigured through an initialization process, enabling precise control of multiple conductance states and demonstrating the feasibility of neuromorphic in-memory computing that mimics biological synaptic behavior.

The study demonstrates that the inherent randomness of RRAM can be utilized not merely as device variability but as a valuable physical resource for hardware security, after which the same device can be transformed into an AI computing memory array. The technology is expected to provide an important foundation for future commercialization of on-device AI and AIoT systems—including autonomous vehicles, smart homes, and wearable devices—that require both hardware security and AI computation within a single platform.

The research was conducted by graduate students Kwon Seongbin (co-first author), Lee Taewon (co-first author), and Kim Dohyung (co-author) from the Department of AI Semiconductor Engineering, under the supervision of Professors Kim Sungjun and Hwang Sungmin, who served as corresponding authors.

Kwon and Lee commented, “Rather than viewing the randomness inherent in RRAM devices as a source of instability, we sought to utilize it as a physical resource for hardware security. Our study demonstrates that thermally controlled conductive filaments make it possible to reconfigure the same memory array for in-memory computing even after generating PUF-based security keys.”

Professors Kim and Hwang stated, “This research presents a new paradigm for implementing both strong security authentication and high-efficiency AI computing on a single hardware platform, addressing key requirements for next-generation edge devices. Because the platform can be reconfigured without physical damage, we believe it will become a key technological breakthrough for the commercialization of secure and efficient intelligent hardware.”

This research was supported by the Ministry of Science and ICT and the Institute of Information & Communications Technology Planning & Evaluation (IITP) through the University ICT Research Center (ITRC) Program (IIPT-2026-RS-2026-25520273), and by the Ministry of Education and the National Research Foundation of Korea (NRF) through the Basic Science Research Program (RS-2023-00248731).