In simulating properties of molecules and materials, computational chemistry has the potential to accelerate scientific discovery for a wide variety of applications by reducing the need for costly and drawn-out experiments. But current computational methods are inadequate for exploring the vast and complex space of possible materials—thus limiting their effectiveness and performance.

Azure Quantum Elements, a Microsoft platform built with artificial intelligence (AI) and cloud-powered high-performance computing (HPC), offers scientists an opportunity to examine immense regions of the chemical universe at unprecedented speeds. In one recent demonstration, Microsoft researchers used Azure Quantum Elements to rapidly search tens of millions of possible materials for use as a battery electrolyte—the conductive medium that facilitates the transit of ions between electrodes. After narrowing this virtual field in just a week, they turned to a major US National Laboratory to select and test the most promising candidates. They even identified one particularly interesting material that could reduce dependency on lithium, whose increasing scarcity is a critical concern for future energy-storage applications.

This entire process took just nine months, including creating a proof-of-concept battery with the newly synthesized candidate. Even though the team is still characterizing and optimizing these candidate materials, this project demonstrates how the quick analytical capabilities of Azure Quantum Elements can help chemists and materials scientists get projects off the ground faster than ever and focus on the creative work of developing innovative and functional tools and products.

Streamlining discovery

When Microsoft announced Azure Quantum Elements in June 2023, CEO Satya Nadella declared that the goal was to “compress the next 250 years of chemistry and materials science progress into the next 25.”

This is no simple feat, says Nathan Baker, product leader of Azure Quantum Elements at Microsoft and a former chemist. “Computational chemistry is hard,” he explains. “It’s hard because it needs high-end computing, and that high-end computing is hard to procure, it’s hard to scale up, and it’s hard to maintain.” The goal of Azure Quantum Elements is to make these capabilities more broadly available and usable for researchers who may not be deeply familiar with cutting-edge computing.

The Azure Quantum Elements team achieved this goal with a multipronged approach. The first component is cloud-based HPC, which provides the flexible and scalable processing power that scientists need to analyze materials and chemical reactions at ever-larger scales. These HPC capabilities are wedded to a sophisticated toolbox of AI models, which have been trained on a dataset of millions of materials simulations and massive amounts of published scientific data from the chemical literature.

AI-assisted analysis can eliminate certain time-consuming calculations by using this training material to generate data-informed predictions about critical properties of a particular substance or chemical formulation—for example, parameters that influence electrochemical behavior. Of course, there are still many things that AI cannot predict and are better resolved with computer simulations grounded in real-world physical and statistical principles. “I believe firmly that there’s never going to be a replacement for simulation, because simulation doesn’t just give you an answer. It gives you insight, and that insight drives the design,” Baker says. But even in this context, AI can help derive physical models that lead simulations that are more efficient and accurate. By combining AI and HPC, the Microsoft team estimates that researchers can speed up key aspects of their analytical workflows by up to 500,000-fold compared with conventional computational methods.

Azure Quantum Elements also supports a wide variety of popular computational chemistry tools. Connecting all these processes can be technically daunting for researchers focused on laboratory work rather than on computational methods. To further simplify these processes, Microsoft also released Copilot in Azure Quantum Elements, an AI-powered interface that relies on the same large language model framework underlying systems like ChatGPT. Users can input queries to the platform in plain English—or whatever the scientist’s language of choice may be—on a particular chemistry problem and receive an equally clear response from the algorithm. Copilot can also cue up the algorithmic tools required to perform the analysis. “That really removes barriers for people,” Baker says.

Turbocharging battery development

The battery electrolyte development effort began as a proving ground to demonstrate the efficiency and utility of Azure Quantum Elements to the chemists and materials scientists on the Azure Quantum team. Materials development involves much winnowing down of candidate materials to home in on the most promising. By starting with a bigger pool in the early stages, scientists can potentially stack the odds in favor of success. Azure Quantum Elements’ analytical capabilities allowed Microsoft researchers to begin at a truly massive scale, with a starter set of 32.6 million possible materials.

The first round of AI-assisted screening narrowed that number to half a million potential electrolytes. Most of these were culled during an assessment of properties related to the material’s ability to facilitate the movement of ions, such as redox potential and band gap. Filtering 32.6 million starting points down to 800 semifinalists took just 80 h of computing. This represents a dramatic time savings relative to conventional computational methods, with which “screening that many materials would have taken two decades,” Baker says. The options were then narrowed to 150 via physics-based simulations that analyzed each material’s molecular-scale properties and behavior.

At this point, Microsoft reached out to a major US National Lab team to get its perspective on the fruits of Azure Quantum Elements’ labors. The team saw opportunities for further refinement and suggested additional screening parameters, including eliminating materials that are excessively chemically reactive or that incorporate conductive but costly elements such as platinum or gold. After applying these parameters and further considering the list of finalists, the team converged on four materials that appeared to be sufficiently cost effective, conductive, and stable for use as a battery electrolyte.

Most modern lithium-ion batteries rely on electrolytes based on lithium salts. The flow of lithium ions between the battery’s electrodes results in the generation of current, whereas applying a charge to the battery reverses that ionic flow, which regenerates the battery’s current-generating capacity. Lithium is extremely lightweight and electrochemically active, making it a valuable material for batteries. Unfortunately, the demand for lithium currently outstrips production capacity, and lithium is generally extracted via environmentally damaging mining processes. As a result, there is considerable interest in reducing lithium dependence.

One material proposed by Azure Quantum Elements offered such an opportunity, achieving good conductance while using about 70% less lithium than existing electrolytes. Remarkably, this feature was not even among the parameters applied during screening. The lithium ions are substituted with positively charged sodium or potassium, both of which are abundant and far easier to obtain than lithium. After synthesizing this material and experimentally validating the properties predicted by Azure Quantum Elements, the US National Lab team assembled a simple proof-of-concept battery using this electrolyte and showed that it could generate a stable charge at both room temperature and under high heat.

Spurring chemical creativity

This isn’t the end of the story. Although the low-lithium electrolyte exhibited reasonable conductance, it was not able to outperform other state-of-the-art materials. Further testing and optimization of this material will thus be necessary, and several other candidates identified by Azure Quantum Elements have yet to be assessed by the US National Lab.

“We got a hit, and now we need to optimize that lead,” says Baker, who draws parallels to the process that the pharmaceutical industry uses to turn vast chemical libraries into safe and effective drug candidates. He also sees lessons that could lead to greater efficiency in future projects of this nature, including earlier use of more aggressive filtering measures based on desirable material parameters to narrow the field sooner.

Nevertheless, the results of this project highlight opportunities for researchers in a wide range of industries to dramatically accelerate their early-stage R&D efforts. Microsoft has already attracted many commercial partners, including chemical industry leaders like Johnson Matthey and AkzoNobel, to explore Azure Quantum Elements’ capabilities. Baker is particularly excited about the potential to use this platform to model chemical reactions and design catalysts that efficiently execute those reactions—an important capability in manufacturing, energy generation, environmental remediation, and other applications.

The Azure Quantum Elements platform is designed to interface seamlessly with emerging quantum computing capabilities. These computing technologies are still under development and are primarily being used in an exploratory fashion. But Baker predicts that as quantum computing matures, it will enable researchers to confidently assess the ground-truth accuracy of their models and thereby develop even more computationally efficient and successful processes for materials discovery.

Azure Quantum Elements already has the potential to be a game changer. “Scientific discovery is too slow—the Edisonian approach doesn’t scale the way we need it to scale,” Baker says. As more and more researchers use the platform, he is eager to see how they leverage the power of AI and HPC in their work. “Watching this get out into industry and have an impact is thrilling,” he says. “Just seeing the creativity of customers with the platform is really exciting.”