MatterGen Diffusion Model for Inorganic Material Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for designing crystalline materials are limited by the need for screening-based approaches, which are constrained by the size of known stable materials and cannot efficiently explore materials guided by desired properties.
Innovation Solution
The development of a generative model called MatterGen, which uses a diffusion process to generate stable inorganic materials across the periodic table. This model can be conditioned to steer the generation towards desired properties, chemistry, and symmetry conditions, using a large energy-compatible training dataset and a fine-tuning scheme.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If screening-based approaches are used to design crystalline materials, then the process is constrained by the size of known stable materials, but the method is simple to implement
Solution Approach 1:
The patent replaces traditional screening-based mechanical search methods with a diffusion-based generative model that uses probabilistic diffusion and denoising processes to explore materials space, enabling property-guided exploration beyond known stable materials
Solution Approach 2:
The patent changes the fundamental parameters of the design process by using conditional generation where desired properties, chemistry, and symmetry conditions are input as conditional parameters that steer the diffusion process toward target material characteristics
2Productivity
If a diffusion model is trained to generate material structures by noising atom types, coordinates, and lattice, then the generation of stable diverse materials is improved, but the training complexity and computational resources increase
Solution Approach 1:
The patent segments the material structure into three independent components that are noised separately: atom types, fractional coordinates, and lattice parameters. This segmentation allows the diffusion model to learn and generate each component independently while maintaining overall structural stability
Solution Approach 2:
The diffusion model is designed to be universal by training on a large dataset of stable periodic materials and then using the same trained model to generate diverse novel materials across different chemical systems and property spaces, achieving multi-functionality in material discovery
3Manufacturing precision
If conditional generation is used to steer material generation towards desired properties, then the precision of property matching is improved, but the model complexity increases
Solution Approach 1:
The patent implements feedback through conditional generation where desired properties, chemistry, and symmetry conditions are fed into the diffusion model as conditional inputs, and the model adjusts its generation process to match these target specifications, achieving precise property matching
Data Source
AI summary
Examples are disclosed that relate to a generative model for generating inorganic material candidates, such as crystalline structures. One example provides a method, comprising training an unconditional generative model using a dataset of stable periodic material structures, the unconditional generative model comprising a diffusion model. The training comprises learning the diffusion model to iteratively noise the stable periodic material structures of the dataset towards a random periodic structure by noising atom types of atoms in the periodic material structure, noising fractional coordinates of the atoms in the periodic material structure, and noising a lattice of the periodic material structure. The method further comprises using the trained unconditional generative model to generate a material structure by iteratively denoising an initial structure sampled from a random distribution.