A research team from the Department of Electronic Engineering at The Chinese University of Hong Kong (CUHK) has successfully developed chip-scale reconfigurable physical unclonable functions (PUFs), also known as “digital fingerprint” technology, based on carbon nanotubes. This innovation offers high-level security for smart devices, particularly against attacks targeting artificial intelligence (AI) and machine learning systems. It holds promising potential for applications in autonomous vehicles, such as robotics, drones and unmanned aerial vehicles, and the Internet of Things (IoT) systems. The research findings have been published in the well-known international journal Nature Communications.

Turning randomness into a defence against attacks

The rise of edge intelligence – where devices analyse data locally rather than relying on cloud servers – is resulting in increased vulnerability to hacking, physical cloning and reverse engineering. To counter these security risks, hardware-based authentication and trust protocols are required to combat counterfeiting, prevent unauthorised access and secure the communications between edge devices. PUFs, also known as the “fingerprints” of hardware, are  physical objects whose operation cannot be cloned in a physical sense by creating identical hardware. For a given input and conditions (i.e. challenge), PUFs are made possible by exploiting tiny, uncontrollable variations from the hardware fabrication process to generate unique, unclonable responses when challenged, providing a physically defined digital fingerprint output (i.e. response) that serves as a unique identifier.

Traditional PUFs are built on silicon technologies, but they face limitations in entropy, reconfigurability or resilience against modern AI and machine learning-driven attacks. A research team led by Dr Liu Yang, Dr Pei Jingfang and Professor Hu Guohua from the Department of Electronic Engineering at CUHK has developed a new approach using the random distribution of carbon nanotubes in large-scale reconfigurable memory device fabrication. This technique enables both physical unclonability and dynamic reconfigurability, two critical features in enhancing edge device security. Its reconfigurability allows PUFs to generate challenge-response pairs dynamically without recalling the chips, which is especially valuable in resource-constrained edge environments. This scalable, solution-based fabrication approach is also crucial for widespread PUF deployment in edge intelligence.

Innovations from materials science to electronic circuits demonstrate high security to resist attacks and cracking

Carbon nanotubes, a one-dimensional nanomaterial with high carrier concentration, high mobility and tunable electronic properties, lay the foundation for this innovation. Dr Liu Yang said: “By designing and fabricating carbon nanotube memory arrays, the team prototyped PUFs and demonstrated that they could be programmed into over 10¹³ possible configurations, representing unprecedented reconfigurability for PUFs.” The PUFs achieved ideal randomness, uniqueness and reliability, outperforming prior solutions. Moreover, the reconfigured PUFs also demonstrated great performance in repeated primitive generation operations, outperforming other reports.

By exploiting the idea of physical unclonability, the carbon nanotube PUFs demonstrated exceptional resilience, proving that they provide high security. Dr Pei Jingfang said: “The experimental results showed that even advanced AI and machine learning algorithms could only achieve about 50–60% success rate in attacking the PUFs, essentially no better than random guessing. In addition, brute-force attacks would require an estimated 10¹⁶ years for successfully cracking a 108-bit PUF primitive. The demonstrations underscored the robustness of the PUFs in resisting attacks and cracking.”

With this resilience and high security, as a proof of concept, the researchers designed a key-exchange protocol to secure self-driving vehicular communication networks. Embedded in a self-driving vehicular network model in Hong Kong’s Central district, the key-exchange protocol ensured vehicular communication with fast authentication, low computational overhead and minimal latency.

Towards large-scale integration and deployment

Professor Hu Guohua said: “While our carbon nanotube PUF shows these promising results, the lab-to-fab gap needs to be bridged. Future efforts will focus on adapting carbon nanotube PUF fabrication to industrial-scale photolithographic processes, achieving complementary metal-oxide-semiconductor (CMOS) integration for the embedding of analogue front ends and digital logic on a single chip.” In the future, it will be necessary to integrate multi-channel data converters and microprocessors into the system, combining hardware and algorithms to promote their application across various domains.

The full research article can be accessed here:

Nature Communications https://doi.org/10.1038/s41467-025-63739-x

Source: CUHK CPRO’s press release

The Chinese University Hong Kong (CUHK)’s InnoHK Hong Kong Centre for Logistics Robotics (HKCLR) has unveiled Hong Kong’s first self-locally developed, AI-powered robotic platform. It comprises two core components, the LY 1 quadruped robot and the dual-arm embodied AI system powered by vision language model. This platform integrates high mobility with intelligent manipulation capabilities, showcasing advanced technologies in perception, decision-making and hardware execution. This is highly versatile and can be applied across various sectors such as logistics, retail and industrial automation, highlighting Hong Kong’s innovative capabilities in robotics and embodied intelligence, and marking a new chapter for local innovation and technology. 

Promote smart city development through robotic technology

Professor Sun Dong, Secretary for Innovation, Technology and Industry of the Hong Kong Special Administrative Region of the People’s Republic of China, said: “HKCLR is a R&D centre under AIR@InnoHK, focusing on the development of applied robotics and AI technologies. I am delighted to witness the launch of Hong Kong’s first self-locally developed quadruped robot and humanoid dual-arm manipulation platform based on advanced AI models. This technology and product launch is closely aligned with the HKSAR Government’s strategy of focusing on developing the AI and robotics industries, demonstrating the results of close collaboration among the Government, industry, academia and research sectors.”

Professor Sham Mai-har, CUHK’s Pro-Vice-Chancellor (Research) and Chairperson of Board of Directors of HKCLR, said: “Since its inception, HKCLR has been closely collaborating with prominent international and local research teams, achieving continuous breakthroughs in key technologies. This latest embodied AI research outcomes not only inspire the industry but also foster cross-sector collaboration, facilitating the industry towards a more efficient and intelligent future.”

Hong Kong’s first locally developed AI quadrupedal robots and humanoid dual-arm platform

The LY 1 quadruped robot is a compact and efficient robotic platform designed for high mobility in complex environments. Its innovative differential drive system enhances structural efficiency and load capacity, while reinforcement learning-based motion control and visual navigation algorithms enable autonomous movement across rugged and challenging terrain.

The VLM-powered dual-arm embodied AI system integrates multimodal perception with a vision language model, enabling semantic understanding and environment perception. It can adapt its responses to real-world situations, providing intelligent assistance across household, industrial and logistics settings. Through semantic sorting and multi-robot coordination, the system enhances logistics and warehouse efficiency, while enabling flexible handling, precision assembly and seamless human-robot collaboration in industrial operation. This contributes to the advancement of smart city development.

These two systems form a unified platform which enhances operational efficiency in warehouse and retail environments. The dual-arm system can automatically perform tasks such as goods picking, sorting and transport, reducing manual labour demand and increasing operational throughput. It also supports shelf restocking, order fulfilment and collaborative functions in supermarkets and unmanned retail stores.

Professor Tsang Hon-ki, Dean of Engineering at CUHK, said: “The research achievements of CUHK’s InnoHK Hong Kong Centre for Logistics Robotics in embodied intelligence have provided practical solutions for industrial upgrades through automation and intelligent technologies, and brings new momentum to the sustainable development of smart cities. The CUHK’s Faculty of Engineering will continue to support HKCLR to promote the wider applications of robotics and artificial intelligence technologies, and to reinforce the position of Hong Kong, China as a major international hub for smart logistics and robotics innovation.”

Professor Liu Yunhui, Director of HKCLR, and Choh-Ming Li Professor of Mechanical and Automation Engineering in the Faculty of Engineering at CUHK, said: “HKCLR has made technological breakthroughs in quadrupedal robots, dual-arm manipulation and embodied AI, providing cutting-edge technologies for Hong Kong. These innovations can be applied to robotic inspection, warehouse automation, intelligent sorting, delivery, and services, contributing to the sustainable development of smart cities locally and globally.”

Source: CUHK CPRO’s press release