Empowering Materials Innovation Through AI Advancements
Advanced artificial intelligence and robotics tools are accelerating efforts to develop new, much-needed materials.
● Google DeepMind researchers announced this week that a new artificial intelligence model has identified more than 2.2 million virtual objects.
● Of these predictions, 381,000 new substances were considered stable, making them prime candidates for further experiments in a laboratory setting.
The development of new materials is crucial to the development of various technologies such as batteries, solar cells, semiconductor chips and other devices necessary for the next generation of electrical grids and computing systems.
● Various countries, including the United States, China and India, are investing heavily in materials science and engineering. According to a report by the Georgetown Center for Security and Emerging Technology, artificial intelligence and materials science have been the biggest recipients of US federal grants to industry in the past six years.
● Over the past century, the traditional method for discovering new materials has involved replacing elements in the molecular structure of existing stable materials. However, this approach has limitations in terms of cost, time and the possibility of detecting radically different structures.
● DeepMind was made possible by leveraging existing data from the Materials Project at Lawrence Berkeley National Laboratory (LBNL) and other databases.
● The AI, called Graph Networks for Materials Exploration (GNoME), uses two deep learning models to represent atoms and bonds in a molecule as a graph.
● GNoME has demonstrated the ability to predict structures beyond what it was trained on, demonstrating its ability to discover materials that would otherwise be difficult for human scientists to identify. While predicting the stability of a potential structure does not guarantee its ability to be manufactured.
The use of AI-guided robots at LBNL has shown promising results in autonomous crystal manufacturing.
However.
● Installation and testing of new materials remains expensive and time-consuming. in addition to.
● The lack of explanation for the decisions made by AI models poses a challenge in gaining new knowledge from their predictions.
● Developments in artificial intelligence and machine learning are beginning to impact this field, with the development of self-driving laboratories and the automation of some scientific experiments.
This intersection between software and hardware is expected to drive the next wave of scientific progress.