Understanding Phaseshift's Framework

Phaseshift Technologies develops highly specialized and advanced alloys using state-of-the-art Computational Chemistry (CC) workflows coupled with Machine Learning (ML) models. This helps improve the ROI on materials R&D by making the process more time and resource-efficient while optimizing the end material for performance and cost. Phaseshift’s current material focus is on Amorphous Alloys (also known as Bulk Metallic Glass or BMG) – a class of highly-specialized alloys with high tensile strength, increased resistance to corrosion, and other unique material properties.


Alloys are solid metallic compounds that are developed by mixing different metals (in some cases non-metals) together to produce the desired set of material characteristics. Obtaining an alloy composition from a target set of properties is a form of an inverse design problem and is extremely difficult to solve. Traditionally, sequential trial-and-error has been used to develop new alloy compositions but due to a large combinatorial design space, it quickly becomes infeasible to effectively search through all possibilities in a realistic timespan. This traditional approach can cause two primary problems: 1) Failed attempts can lead to a large number of iterations before a potential candidate is identified, and 2) parts of compositional space are left unexplored leaving out the possibility of a better material candidate.


Phaseshift augments the R&D process with Machine Learning predictions and Quantum Chemistry modelling to address the aforementioned challenges, a description of which is as follows:

  1. Machine learning for exploring the design space: A combinatorial space of 3-5 elements can be as large as 100,000 to 500,000 unique compositions. Investigation of every single composition in this design space quickly becomes impractical through manual iteration. The incorporation of ML as a guide to identify highly promising regions in the design space accelerates the material discovery process. The design space, the database along with various descriptors, and the target metric for ML are defined based on the needs of the project and the type of alloy system to be explored.

  2. Quantum chemistry for material property modelling: Material candidates suggested by Machine Learning models are subsequently investigated through Density Functional Theory (DFT) and Molecular Dynamics (MD) simulations. DFT, based on the principles of Quantum Chemistry, is one of the most accurate ways of modelling material properties at an atomic scale. On the other hand, MD is able to model macroscopic properties. Incorporation of both the techniques allows for multi-scale modelling of a range of material properties, such as physical, mechanical, chemical, and electromagnetic.

  3. Positive feedback loop: The outcomes from the simulations and experimental testing of the materials is fed back into the ML models to iteratively improve the accuracy of predictions.

Coupling these approaches, Phaseshift has developed a workflow that optimizes the Materials R&D process for time, cost, and material performance.