In the transient thermal management of high-power electronic devices, composite phase-change materials (CPCMs) exhibit unique advantages. However, traditional material development has long been constrained by inherent trade-offs among competing performance metrics, where heavy reliance on trial-and-error experiments. In recent years, AI has introduced a data-driven paradigm for materials discovery. Nevertheless, existing machine learning methods predominantly rely on end-to-end "black-box" models, which often neglect underlying physical mechanisms, and struggle with multiscale inverse design under conflicting multi-objective constraints. To address these critical challenges, the physics-constrained inverse design system PHICS has been developed. By integrating physics-causal graph modeling, multiscale mechanism embedding, and multi-objective Pareto optimization, PHICS establishes a closed-loop research and development paradigm from microscale design to macroscale performance validation.
The directed acyclic graph (DAG) is introduced as the core topological architecture for physical causal modeling. The PHICS framework explicitly decouples the multiscale hierarchy into a transparent causal chain: microscale composition → mesoscale interfaces → multiphase networks → macroscale functional properties. This allows the framework to retain the high fitting efficiency of data-driven models while strictly adhering to fundamental physical conservation laws and boundary constraints, ensuring superior extrapolation and generalization fidelity. In the presence of intrinsic performance conflicts, the optimal solutions of composites mathematically form a complex, non-convex Pareto front. Conventional weighted scalarization methods easily become trapped in local optima and fail to capture the complete boundary. PHICS tightly couples the physics-causal surrogate model with the non-dominated sorting genetic algorithm II (NSGA-II), enabling high-throughput evaluations within seconds and high-precision inverse parameter inversion of optimal microstructural parameters across high-dimensional, non-convex target spaces.
Guided by the inverse design of PHICS, an ultrathin amorphous alumina (am-Al 2 O 3 ) interface transition layer was precisely introduced into graphite fiber/n-octacosane CPCMs. Acting simultaneously as a "phonon transmission bridge" and an "electron insulation barrier", this layer enables a concurrent leap in both thermal conductivity and electrical resistivity while preserving high latent heat storage density, thereby effectively breaking the long-standing "seesaw effect" among competing properties.
This work not only delivers a new methodology for the development of high-performance thermal management materials, but the established paradigm of "physics-constrained modeling + multi-objective inverse optimization" also paves a pathway for the intelligent inverse design of next-generation functional composites.
This study is supported by the Centre for Heterogeneous Integration and Production (CHIP) under SEAM@InnoHK, the third InnoHK research cluster of the Hong Kong Special Administrative Region. Led by Prof. Shih-Chi Chen, Prof. Jianbin Xu, and Prof. Ni Zhao from The Chinese University of Hong Kong, CHIP is dedicated to advancing next-generation electronics through innovations in semiconductor equipment and materials processing, powered by heterogeneous integration technologies.
National Science Review
Computational simulation/modeling