RESEARCH

We investigate complex physical mechanisms and derive optimal designs by conducting coupled multiphysics simulations—encompassing fluid dynamics, heat transfer, and mass transfer—within Micro-Electro-Mechanical Systems (MEMS) and microfluidic platforms.

In particular, we are pioneering research that integrates Physics-Informed Neural Networks (PINNs) and deep learning technologies with high-fidelity data accumulated from conventional finite element method (FEM) analyses and experiments. By embedding physical governing equations into the machine learning process, we overcome the inherent limitations of black-box AI models, achieving groundbreaking improvements in both prediction accuracy and computational efficiency.

Based on these advanced analytical and modeling methodologies, our current research focuses on the following three core application areas:

• Thermal Management in AI Chips: To address the extreme heat generation issues in highly integrated AI semiconductors, we utilize PINNs embedded with heat diffusion equations. We optimize next-generation cooling packaging structures by predicting heat transfer and fluid flow trajectories from the internal chip to the heat sink in real time.

• Cultured Meat Process Optimization: For the mass production of cultured meat, we precisely model oxygen transfer rates, fluid shear stress, and nutrient distribution inside large-scale bioreactors using multiphysics simulations. By coupling these analyses with AI technologies, we design optimal process conditions to maximize cell proliferation yield over long-term culture periods of 30 days or more.

• Droplet Microfluidics: We analyze the mechanisms of droplet generation, coalescence, and flow instability in microchannels, such as T-junctions, using deep learning techniques. By constructing predictive models for droplet behavior based on hydrodynamic variables, we enhance the reliability and precision of targeted drug delivery and high-throughput single-cell analysis systems.