| Abstract Scope |
Computational materials design through ICME workflows fundamentally requires reliable phase stability predictions, yet many critical material systems lack validated CALPHAD databases. This gap is acute for frontier materials —ultra-high-temperature ceramics, refractory high-entropy alloys, and high-entropy diborides — where experimental thermodynamic data are scarce and traditional assessment is prohibitively slow. We introduce PhaseForge, an open-source pipeline that generates CALPHAD-ready thermodynamic databases from universal machine-learned interatomic potentials (MLIPs) via automated SQS generation, property calculation, and TDB parameterization, achieving >1000x speedup over DFT. The companion MaterialsFramework supports 16+ MLIPs for systematic benchmarking. We further present PhaseForge+, which exploits the Jansson derivative formalism to compute semi-analytical gradients of CALPHAD residuals, enabling conjugate-gradient optimization 1–3 orders of magnitude more efficient than MCMC. We demonstrate the framework on refractory alloys and high-entropy diborides, rapidly bringing underexplored materials into actionable ICME workflows. |