Inorganic Synthesis Reaction Condition Prediction with Generative Machine Learning

Abstract

Data-driven synthesis planning with machine learning is a key step in the design and discovery of novel inorganic compounds with desirable properties. Inorganic materials synthesis is often guided by chemists' prior knowledge and experience, built upon experimental trial-and-error that is both time and resource consuming. Recent developments in natural language processing (NLP) have enabled large-scale text mining of scientific literature, providing open source databases of synthesis information of synthesized compounds, material precursors, and reaction conditions (temperatures, times). In this work, we employ a conditional variational autoencoder (CVAE) to predict suitable inorganic reaction conditions for the crucial inorganic synthesis steps of calcination and sintering. We find that the CVAE model is capable of learning subtle differences in target material composition, precursor compound identities, and choice of synthesis route (solid-state, sol-gel) that are present in the inorganic synthesis space. Moreover, the CVAE can generalize well to unseen chemical entities and shows promise for predicting reaction conditions for previously unsynthesized compounds of interest.

Predicting synthesis recipes of inorganic crystal materials using elementwise template formulation

“…The predicted temperatures from the model reproduce real temperatures qualitatively with a mean absolute error (MAE) of 121.7 °C, which outperforms the MAE (∼140 °C) of previous results. 27,28 Nevertheless, a wide range of temperatures (300-1600 °C) used to synthesize a target crystal with a limited number of data points is potentially contributing to the relatively large MAE observed here.…”

“…heating temperature and time) to synthesize the target materials using text-mined meta-datasets. [26][27][28][29] However, outcomes from generative models used in the latter studies 29 can contain thermodynamically unstable precursors and do not generally present priority among the results, still remaining a question of which results should be tried experimentally rst. In the same vein, they do not inform the measure of condence for predicted reactions, requiring an additional process by domain experts to screen or rank the generated reaction recipes.…”

Flexible, model-agnostic method for materials data extraction from text using general purpose language models

“…More complex information such as synthesis recipes [18][19][20][21][22][23] have also been extracted with automated NLP-based methods. Although not complete due to the relatively low recall, databases of that size are useful for training machine learning models, [24][25][26][27][28][29][30] and would be very time consuming or impossible to extract with virtually any other method than full automation. Other recent examples of databases created in a similar way include photovoltaic properties and device material data for dye-sensitized solar cells, 31 yield strength and grain size, 32 and refractive index.…”

Novelty Detection in the Design of Synthesis of Energy Storage Materials: a Case Study of Garnet-Structured Solid Electrolytes

“…Materials informatics was emphasized as an efficient way towards the rational synthesis of new compounds and new properties. It was efficiently applied in the design of the materials with desired characteristics [5,6,7,8,9], in the design of synthesis and for the autonomous laboratories [10,11,12,13], for natural language processing to obtain and analyze experimental data in chemistry and materials science [14,15,16,17,18], for modeling the microstructure [19,20], for the analysis of the output of physicochemical methods of characterization [21,22], for inverse design of materials [23,24] and in many other areas of their application.…”