Physics-informed neural network based topology optimization for thin-film evaporation in hierarchical structures

Abstract

Thin film evaporation through hierarchical structures is a promising approach for thermal management in electronics and photonics. However, identifying the optimal hierarchical structure for efficient thermal management remains an ongoing challenge. This study presents a coupled framework that integrates classical SIMP-based thermal topology optimization with a pretrained physics-informed neural network (PINN) for data-driven verification to final optimal hierarchical structures. The objective is to minimize thermal compliance in evaporative structures while ensuring physical fidelity. The findings suggest that topologically optimal structures are mostly in the form of branched structures with solid density of $\approx$ 0.5. These structures could achieve high critical heat flux (CHF) at much lower superheats compared to traditionally studied structures. In addition, even for optimal structures, higher density of solid–liquid contact line directly correlates to higher CHF values. This hybrid approach not only enhances computational efficiency but also bridges the gap between simulation and real-world physical behavior, paving the way for validated thermal design in advanced cooling systems.