23 / 2026-10-23 22:38:18

Physics-Guided Inverse Design of Multifunctional Phase-Change Composites for Advanced Thermal Management

physics-constrained machine learning,composite phase-change materials,amorphous alumina,carrier engineering,interface engineering

Efficient thermal management of next-generation power electronics requires materials that simultaneously offer rapid heat dissipation, high thermal-storage capacity, and reliable electrical insulation. However, these competing requirements pose a fundamental challenge to the rational design of multifunctional composites. Here, we develop a physics-constrained inverse design system (PHICS) that integrates multiscale physical modeling, machine learning, and multi-objective optimization to address these intrinsic trade-offs. By incorporating a directed acyclic graph (DAG) architecture, PHICS establishes physically consistent relationships spanning atomic-scale interactions, interfacial phonon and electron transport, mesoscale filler networks, and macroscopic material properties. Coupled with a diversity-enhanced genetic algorithm, this framework efficiently identifies Pareto-optimal material configurations and reveals physically interpretable design principles. In particular, the inverse-design results highlight the critical role of interfacial phase and thickness in decoupling thermal and electrical transport. Guided by these predictions, we fabricate vertically aligned graphite fiber/n-octacosane composites incorporating ultrathin amorphous Al₂O₃ interlayers (VA-GF@am-Al₂O₃/C₂₈H₅₈). The approximately 23 nm amorphous interlayer serves as both an efficient phonon bridge and an electron barrier, enabling enhanced thermal conduction (7.80 W m^–1 K^–1) and electrical insulation (0.093 S m^–1) while preserving substantial thermal-storage capacity (238 J g^–1). The agreement between computational predictions and experimental observations validates the effectiveness of this physics-guided design strategy. Beyond phase-change composites, this work establishes a generalizable paradigm for the cross-scale inverse design of multifunctional materials with competing performance requirements.



Fig. 1 Physics-constrained topological structure of PHICS.





References

[1] Ding, D.; Zou, M.; Huang, R.; et al. Natl. Sci. Rev. 2026, 13: nwag503.

[2] Lyu, T.; Xiong, J.; Zhang, T. Y.;. Natl. Sci. Rev. 2026, nwag615.