Phase Lab: a cloud-native CALPHAD-to-data platform for accelerated alloy design and AI/ICME/MGE workflows

Artificial intelligence (AI)-assisted alloy design, integrated computational materials engineering (ICME), and materials genome engineering (MGE) increasingly require CALculation of PHAse Diagrams (CALPHAD) calculations to be executed at scale and stored as reusable, traceable data. However, conventional CALPHAD workflows are often desktop-centered and tool-specific, which limits reproducibility, high-throughput reuse, and direct integration with machine-learning pipelines. Here we present Phase Lab, a cloud-native CALPHAD-to-data platform designed for alloy designers, CALPHAD practitioners, and AI/ICME/MGE researchers who require scalable computation together with standardized data products. Rather than introducing new physical models, Phase Lab standardizes the full calculation workflow, including browser-based configuration, application programming interface-driven task orchestration, server-side computation, metadata capture, provenance tracking, and structured export. The platform supports thermodynamic, kinetic, thermophysical, solidification, and mechanical-property calculations, while storing database versions, model selections, input conditions, solver settings, and execution records. Point-to-point benchmark calculations showed agreement with Thermo-Calc for representative ternary systems, and batch tests achieved a 100% completion rate for 20,000 ternary tasks with an average runtime of 0.006933 s per task. For 9-component and 12-component systems, completion rates of 99.32% and 98.14% were obtained, respectively, with average runtimes of 1.122 and 2.19 s per task. Validation against literature and experimental data across Fe-, Ni-, Al-, and Cu-based systems further demonstrated the applicability of the platform. For Ni-based superalloys, the room-temperature yield-strength and ultimate-tensile-strength models achieved R2 values of 0.858 and 0.824, respectively. These results indicate that Phase Lab can serve as a practical data-generation platform for dataset assembly, surrogate modeling, active learning, and ICME/MGE-oriented alloy screening.