1. (Design) Bidirectional AI Sensor Design via Physics-Guided Neural Networks
We are building a Physics-Guided Neural Network (PGNN) platform that couples governing sensor physics, capacitive, piezoresistive, and triboelectric with data-driven training. The framework operates bidirectionally: forward prediction estimates sensor performance (sensitivity, linearity, response speed, hysteresis, stability) from design parameters (material, structure, operating principle), while inverse design returns optimized geometries and materials for a target specification. A plug-in module architecture lets each physics block (pressure, strain, vibration sensor) be swapped without retraining the entire network, making the platform readily extensible to wearable healthcare, robotics & electronic skin, human-machine interaction, and smart-clothing applications.
Figure 1. PGNN-based bidirectional sensor design platform
2. (Fabrication) Versatile 2D / 3D Nanotransfer-Printing and All-Inorganic Nanoyarn Platform
We have developed a versatile nanotransfer-printing (nTP) toolkit that orthogonally controls the two competing forces governing nanostructure transfer. mold release (F_mold↓) and substrate adhesion (F_substrate↑). On the F_mold↓ side, we combine solvent-injection-assisted, mold-stripping-based, and mold-dissolution-based release. On the F_substrate↑ side, we engineer adhesive-based, thermoplasticity-based, Laplace-force-based, and nanowelding-based pickup. Hybrid combinations of these mechanisms enable solvent-free, room-temperature 2D and 3D nanostructure printing on arbitrary surfaces — flexible, stretchable, hydrophobic, curved, and even biological (Nature Communications 2023, 2026). The same platform is extended to the world-first all-inorganic nanoribbon yarn (Au, Pd, Pt, Bi₂Te₃, WO₃, SnO₂, NiO, In₂O₃, CuO, etc), providing textile-form-factor building blocks for the downstream sensor and energy devices.
Figure 2. Versatile 2D/3D nanotransfer-printing platform
Figure 3. SEM images of all-inorganic nanoribbon yarn
3. (Sensing & AI Processing & Application) AI-Embedded Multimodal Sensors for E-tongue, XR, and Digital-Human
The designed and fabricated devices are deployed as multimodal sensor systems, including SERS substrates, soft pressure/strain/bending arrays, and gas/bio nanoyarns, and paired with on-device deep learning (CNN, transformer, physics-guided super-resolution networks) that systematically boosts the four critical sensor metrics: selectivity, spatial resolution, response time, and power consumption. Chemical, biological, and physical signals are fused into two application pillars: (i) Food/Plant monitoring (E-tongue), where SERS substrates with AI spectral analysis monitor freshness, nutritional quality, and microbial contamination of meat, fish, and fruit, and on-leaf electrodes track plant impedance and physiological status for smart-farm deployment; and (ii) Human motion monitoring, where skin-conformal soft sensor arrays with on-device deep learning capture real-time posture and gesture for extended reality(VR/AR), physical-twin, and physical-AI applications.
Figure 4. AI-embedded multimodal sensor system