This flagship project inverts the traditional in silico materials design paradigm, and uses practical, real world limitations and opportunities to screen out candidate materials that (for a variety of reasons) will never proceed beyond the computer. Culminating in a modular platform that can operating independently (with or without existing data sets) or as part of a consistent workflow, we aim to deliver the ultimate tool for more sustainable materials discovery.
For more information, contact Project Leader Dr Amanda Barnard.
Developing new random structure searching (RSS) algorithms that more efficiently sample millions of candidate materials, and use physics, chemistry and statistics to screen out those that are physically impossible, thermodynamically unstable, or require synthesis or processing methods that are commercially or scientifically impractical. This stage uses cheap, classical molecular simulation methods and theoretical methods to reduce the resources and investment needed at an early stage.
Incorporating important financial and social considerations that drive decision can be as important as the underlying materials science. Just because a material is physically possible, does not mean that it is economically feasible, or socially acceptable (non-toxic). Moreover, leveraging significant investment in existing infrastructure also guides research, regardless of scientific potential (consider the sunk costs in silicon-based electronics). This module uses the Earth audit, commodities prices, the real cost of fabrication and scale-up and an estimate of the technical risk to further eliminate candidate materials that are not sustainable. Integrating a specific business model into materials discovery and design will accelerate innovation.
Once we know what materials can be made, and which ones we can afford, the challenge becomes identifying the optimal candidate from the remaining options; finding the next “wonder material”. Predicting structure/property relationships using the right combination of accurate quantum mechanical methods and machine learning can draw out intrinsic structural patterns in the data, recognize samples that share structural similarity that may be particularly useful, and identify which structural features enhance functional properties (and which suppress them). This module of the platform will include popular techniques such as artificial neural networks, support vector machines and random forest that can identify materials that are preferable in a given technical context, and more sophisticated methods that use artificial intelligence to guide researchers though the final stages of this more pragmatic approach to materials design.