Generative AI Is Reshaping Material Science

Posted: September 09, 2024

Material science has been the backbone of manufacturing innovation for decades, but discovering and developing new materials has traditionally been a bit of a slog. Scientists have had to test countless compounds and processes by hand, running experiment after experiment in the lab. It's been time-consuming, expensive, and sometimes, frankly, a crapshoot.

But generative AI is shaking things up in a big way.

Since the mid-2010s, chemists have combined machine learning with computer simulations to crunch through massive datasets and help make predictions about material properties.  Today, AI can predict thousands of new compounds with specific, required properties in a matter of hours.

As Ray Kurzweil, who's been in the AI game for over 60 years, wrote in The Economist last June: "AI is about to make the leap from revolutionizing just the digital world to transforming the physical world as well." He thinks this shift will bring countless benefits, but material science is where we might see the biggest impact.

How can generative AI speed up the chemistry processes?

Anyone who has spent some time in a chemistry lab knows how painful it can be to synthesize a new compound, purify it and get it to interact with other compounds. The process can take months, even years, and the result is not guaranteed. Now, generative AI can create new chemical structures based on specified parameters or desired properties in hours.

Here is a statistic that gives an idea of the magnitude of the AI potential: Scientists had discovered only about 20,000 stable inorganic compounds in all human history until Google's AI tool, Graph Networks for Materials Exploration (GNoME), discovered 380,000 new crystal structures predicted to be stable and promising candidates for experimental synthesis in a single night.

Generative AI helps chemists through all phases of the chemical discovery process.

Design - AI accelerates molecular design by rapidly exploring possible combinations and finding novel structures. This process involves iterative cycles that progressively refine molecular designs, resulting in a diverse array of promising molecules for further investigation.

Make - Chemical synthesis often represents the most time-intensive phase in the chemical discovery and materials development phase. AI platforms allow chemists to converse with AI to develop new synthetic routes for target molecules. AI algorithms can identify the most efficient and automatable pathways for chemical synthesis, and can assist in the identification and optimization of catalysts. Unlike human chemists, AI-driven synthesis ensures consistent, reproducible results across multiple iterations. Some advanced AI systems can even perform subsequent steps such as purification and analysis of synthesized compounds.

Test - AI combines current and historical data to predict things like bond lifetimes and light absorption efficacy. AI can simulate and predict molecular properties in virtual spaces at a pace that manual methods could never achieve, eliminating a lot of laborious experiments.

Smart materials for renewable energy production: The AI-driven solar revolution

Humanity has relied heavily on non-renewable, environmentally harmful fossil fuels for two centuries. Yet, the potential for clean energy is immense — as little as 0.02% of the incoming sunlight could satisfy all human energy needs.

Solar technology has made remarkable strides, with the cost per watt of capacity plummeting since 1975, when the first solar panels cost about $115.3 per watt. By 2010, this price was already $2.15 per watt, and today, a solar panel can cost as little as $0.50 a watt. However, solar energy has not yet become the dominant power source due to two main challenges:

1. Photovoltaic efficiency: Despite significant improvements, current materials remain too costly and inefficient to replace fossil fuels fully.
   
   
2. Energy storage: The intermittent nature of solar power necessitates extensive energy storage solutions, which current battery technology struggles to provide cost-effectively.

While the laws of physics suggest vast room for improvement, the sheer scale of chemical possibilities has resulted in slow progress through traditional research methods.

Since generative AI came on the scene, material discovery for photovoltaic and energy storage has boomed.

In January this year, Microsoft and the Pacific Northwest National Laboratory scientists queried their Azure Quantum Elements (AQE) AI platform to identify battery materials with reduced lithium content. For decades, scientists have tried to find alternatives to using this scarce element. AQE quickly suggested 32 million potential candidates. The AI then evaluated these materials for stability, narrowing the list to about 500,000. Further filtering assessed each material's energy conductivity, simulated atomic and molecular movements, and considered cost and availability, ultimately identifying 23 viable candidates, five of which were already known materials. This entire process, which would be unfeasible for humans to perform on a lab bench, took just 80 hours. AI systems such as AQE allow the chemical industry to accelerate processes and improve energy efficiency.

As AI algorithms become more advanced and discover optimal materials for photovoltaic and energy storage, large-scale solar projects could become not just viable, but transformative. This could lead to an era where solar energy is so abundant and cost-effective that it becomes almost free, marking a pivotal shift in global energy production and sustainability efforts.