Bringing an end to Mix, Heat, Test, Repeat

How PhaseTree is able to rapidly accelerate materials discovery to deliver the next generation of advanced materials

Throughout human history, our progress has been defined by the materials we master. From the Stone Age to the Iron Age to the Silicon Age, new materials have consistently unlocked new technological paradigms. Today, as we face global challenges like climate change and resource scarcity, the need for material innovation has never seemed more urgent. 

We need lighter, stronger alloys for more fuel-efficient vehicles and aircrafts. We need more efficient solar cells and battery components for the green energy transition. We need more sustainable alternatives to the way we currently use concrete and steel, which account for a significant portion of global carbon emissions. These future-defining technologies all depend on the creation of new and improved materials.

Traditionally, the process of discovering a new material has been a slow, expensive, and often serendipitous journey of trial and error. This analogue approach typically involves a researcher having an idea for a new chemical composition, which is then physically synthesized in a lab. 

That small sample is then subjected to a plethora of tests to determine its properties. This cycle of "mix, heat, test, repeat" is done, often for years, inching towards a desired outcome. The entire process, from initial concept to market-ready material, can easily take between 10 and 20 years, involving immense investment in lab equipment and personnel. It is a system built on a foundation of educated guesswork, where breakthroughs are hard-won and the pace of innovation struggles to keep up with global demand.

But what if we could test thousands of potential materials without ever stepping into a lab?

This is the promise of computational materials science. By using powerful computational simulations, we can replace the costly and lengthy physical trial-and-error process with a far more efficient, virtual one. Before a single gram of material is ever synthesized, we can model its atomic structure and predict its properties—its strength, its conductivity, its resistance to heat or corrosion, and more. This in-silico approach allows us to rapidly screen vast libraries of potential candidates, discarding dead-ends and identify the most promising compositions for further, targeted physical testing. This shift from physical to virtual experimentation drastically reduces the development timeline from over a decade to a matter of months, sometimes even weeks, while simultaneously lowering costs.

The PhaseTree Approach: Seeing the Forest and the Trees

The challenge with simulation is capturing the full picture. The properties that make a material useful—like its strength or how it fractures—don't come from a small atom structure, but from the collective behavior of millions or even billions of atoms. This is known as the "mesoscopic scale." Many simulation methods can either see the precise, quantum-level interactions of a few atoms or the bulk behavior of a material, but they struggle to connect the two. 

It’s like trying to understand how a city functions by only studying a single brick, or by only looking at a satellite map. You miss the crucial information at the scale of buildings and neighborhoods.

PhaseTree's platform is built to bridge this critical gap. Our technology performs multiscale simulations that connect the fundamental, atomic-level interactions all the way up to the mesoscale phenomena that determine real-world material performance. This allows us, for the first time, to predictably design new materials deterministically from a conceptual idea. By creating a reliable two-way flow of data between the atomic and mesoscale, we can build a more complete and accurate picture of a material's potential than ever before. This connection between the different scales furthermore enables us to harness a powerful cycle for AI self-learning.

The Science Behind: Multiscale Modeling

Multiscale modeling is a computational technique that uses the results from a simulation at one length or time scale as input for a larger one. In materials science, this means we can use highly accurate but computationally expensive quantum mechanical calculations (which model the behavior of electrons and atomic nuclei) to understand the fundamental forces between a small number of atoms.

This information is then used to train a robust model for a larger group of atoms, which is then again used to further scale up, and so on, up to the mesoscale. This "upscaling" of information is what allows PhaseTree to simulate the critical, collective atomic behaviors that give a material its unique macroscopic properties, bridging the gap between quantum mechanics and real-world engineering.

The quantum level computations rely on Density Functional Theory (DFT) to calculate system energies of various small-scale atomic structures, using different atomic concentrations. These calculations provide critical insights into material behavior, including total energy, electronic structure, and magnetic properties.

With this DFT “ground truth” dataset, PhaseTree trains an advanced Cluster Expansion (CE) model to efficiently map the structure-property relationship of materials with crystal structures, enabling rapid statistical sampling of billions of structures. This allows for the scaling up of atomic structures into the thousands, without breaking the structure-property relationship, and without the computational cost associated with it. Something that would otherwise take months to compute can be generated in seconds.

Thanks to the ultrafast CE framework, our Monte Carlo (MC) module supports the simulation of millions of atoms in a simulation cell, covering up to the micrometer scale while retaining atomic resolution of material structure and evolution. Our proprietary module also includes specialized analysis tools that track atomic connectivity, magnetic properties, segregation, and short-range order, as well as defect formation and migration, during MC simulations to provide deep insights into the materials at different temperatures, pressures, compositions, and timescales.

The Road Ahead: AI and the Future of Materials

As we look to the future, we are extremely excited as we near the closing of the otherwise pervasive gap between the atomic- and the mesoscale, with our addition of Phase-field modelling. Bridging this gap will enable an entirely new paradigm of two-way learning between atomic properties and microstructure evolutions, revealing things like fracturing, ductility, and tensile strength, derived from e.g. grain boundary formation. This presents a critical advancement in materials science between how materials behave in simulations relative to in real-world applications.

The field of materials discovery is rapidly advancing towards greater integration with artificial intelligence. A key advancement in this area is the development of Machine Learning Interatomic Potentials (MLIPs). Derived from neural-network AI models trained on vast datasets from quantum mechanical calculations, these models learn to predict various mathematical forces between atoms. At PhaseTree, we see this as one of the greatest catalyst for further materials innovation, as it significantly expands the top of the “discovery funnel” with computed thermodynamically stable materials candidates. 

As the space of potential compositions to review grows exponentially, we are confident that the need for multiscale simulation, helping with candidate targeting and property optimisation, will grow along with it.