A high performance CALPHAD data engine for scalable ICME and AI-assisted materials design

Abstract CALPHAD calculations provide the thermodynamic and kinetic basis for integrated computational materials engineering, but point-wise coupling becomes inefficient when simulations, optimization loops and AI-assisted workflows repeatedly query large material-state spaces. We report a runtime CALPHAD data-delivery framework that reduces redundant equilibrium and property evaluations through managed thermodynamic state-space reuse, finite-element cell clustering and adaptive time control for kinetic and microstructure updates. Benchmarks in the Co-Cr-Fe-Mn-Ni system show numerical consistency with direct point-wise CALPHAD calculations, with an approximately 11-fold speedup for first-run high-throughput calculations and four-order-of-magnitude acceleration for repeated queries after state-space reuse is enabled. Runtime coupling to laser powder bed fusion simulations links local thermal histories to thermophysical properties and solidification descriptors, while an AI-assisted cryogenic alloy-design workflow shows how deterministic CALPHAD calculations can screen language-model-proposed alloys and conditions for testing a design hypothesis. These results demonstrate a scalable route for database-consistent CALPHAD delivery in process modeling, microstructure prediction and physics-grounded materials design.