The next generation of microelectronics relies on radical improvements in transistor switching performance to advance computing power. However, conventional semiconductor technology has hit a physical limit known as the "Boltzmann tyranny," which restricts the energy efficiency of traditional transistors and stalls progress in high-performance electronics. To address this key bottleneck, a research team at The Hong Kong Polytechnic University (PolyU) has engineered a novel tunnelling field-effect transistor (TFET) utilising two-dimensional (2D) nanomaterials. The breakthrough brings this long-awaited experimental technology closer to commercial reality, offering a fundamental building block for energy-efficient computing and next-generation artificial intelligence (AI) chips.

Colorectal polyp detection from endoscopic images is critical for the early diagnosis of colorectal cancer. However, traditional deep learning methods often suffer from limited generalization when deployed across datasets containing different polyp morphologies. This work aimed to investigate whether vision-language foundation models can facilitate zero-shot generalization across multiple polyp datasets without target-domain fine-tuning.

The three-year project will explore how industry-informed research, entrepreneurship training and early engagement can better prepare undergraduates for careers in biotechnology and the life sciences. Students will gain access to research, internships, industry experiences and AI-enabled learning while expanding workforce opportunities, particularly for transfer and first-generation students. The project also will identify practices to strengthen career readiness and develop approaches that can be adapted by institutions nationwide.

A new artificial intelligence (AI)-based system shows how natural damage to traffic signs can mislead deep neural network (DNN)-based vision systems. The framework, AdvWT, uses a generative model to learn realistic damage patterns and then optimize them to trigger misclassification. Tests across multiple AI architectures showed strong attack success, transferability and physical-world robustness. The approach can identify vulnerabilities and improve AI generalization to real-world damage.