SCIENCE CHINA Materials•2026•DOI: 10.1007/s40843-026-4467-x
Conventional eye-movement interaction systems depend on video capture, infrared tracking, and image recognition, which impose inherent constraints on accuracy, response latency, and stability. This study introduces an eyelash-guided signal interaction system based on a triboelectric nanogenerator (PF-TENG) using PDMS-FDTS thin films. The system employs eyelash movements as interactive inputs, eliminating the need for complex optical acquisition devices. A CNN-LSTM hybrid neural network classifies distinct eyelash movement patterns with a classification accuracy exceeding 98.5%. The PF-TENG device exhibits ultra-flexibility and transparency, enabling seamless integration onto eyeglasses without obstructing the user's field of view. Experimental validation demonstrates real-time monitoring of ocular states for driving fatigue detection, accurately identifying fatigue signs and enhancing application potential in intelligent driving. The system offers a natural, comfortable input modality and significant advantages for human-machine interaction, with broad prospects in eye-movement control and intelligent transportation.
SCIENCE CHINA Materials•2026•DOI: 10.1007/s40843-025-3553-7
Gesture interaction has emerged as a highly effective interface for intelligent human-computer interaction, attributed to its intuitive interaction modality and multi-dimensional control capabilities. However, traditional gesture interaction devices often depend on predefined encoding rules, which substantially limit interaction efficiency and degrade user experience. This study introduces an innovative intelligent finger ring interaction system based on a triboelectric nanogenerator utilizing PDMS/SrTiO3 composite thin film (PS-TENG). The system maps freehand writing gestures directly to textual information input, thereby eliminating the need for complex gesture encoding schemes and offering a user-friendly, low-learning-curve input method. By integrating a deep learning model, the system achieves recognition accuracies of 98.21% for English letters, 96.87% for Arabic numerals, and 96.44% for Chinese characters. Furthermore, it supports secure and encrypted data transmission and enables wireless interaction for gaming control. These findings indicate that the intelligent finger ring interaction system possesses significant potential for practical applications in information input and wireless control.
SCIENCE CHINA Materials•2025•DOI: 10.1007/s40843-025-3428-6
Alternating current electroluminescent (ACEL) devices with a pyramidal conical structure luminescent layer were fabricated and integrated with a convolutional neural network (CNN) to construct an image recognition system for robotic arm applications. The ACEL device serves as a flexible, low-power, homogeneous light source, enabling multispectral imaging that mitigates ambient light interference. Images captured by an ESP32-CAM module are processed by a deep learning model, achieving a recognition accuracy of 96.7% for seven distinct shapes (rectangle, triangle, circle, star, butterfly, flower, snowflake). The system demonstrates high brightness, high contrast, and flexibility, addressing limitations of traditional image recognition systems that rely on hand-designed features and are susceptible to illumination variations. This work validates the potential of ACEL-based multispectral imaging for robust environment perception in dynamic scenarios, offering a pathway toward more efficient and reliable robotic vision systems.
SCIENCE CHINA Materials•2025•DOI: 10.1007/s40843-025-3498-5
Amorphous gallium oxide (a-Ga2O3) suffers from low carrier concentration and limited mobility, impeding its use in neuromorphic computing. This study fabricates Zn-doped Ga2O3 (ZGO) two-terminal artificial synaptic devices via radio-frequency magnetron sputtering (RFMS) under oxygen-free conditions. Compared to undoped Ga2O3, the ZGO device exhibits a 106-fold increase in excitatory post-synaptic current under 254 nm illumination, with response intensity positively correlated to optical pulse parameters. Under light pulse modulation, the devices demonstrate dynamic transitions from short-term plasticity to long-term plasticity, including paired-pulse facilitation and a learning-forgetting-relearning process. Electrical and optical energy consumptions of synaptic events are as low as 28 fJ and 2 nJ, respectively. Mechanism analysis attributes the persistent photoconductivity effect in ZGO thin films to abundant oxygen vacancies. A multi-layer perceptron simulation based on ZGO devices achieves 90.74% accuracy in handwritten digit recognition and maintains 76.18% accuracy under 50% noise. Zn doping provides a new material design approach for Ga2O3-based neuromorphic devices, demonstrating potential for future neuromorphic computing applications.