Microsoft Uses AI And HPC To Analyze 32 Million New Materials

By Paul Smith-Goodson, Patrick Moorhead - February 9, 2024

In a partnership with the Pacific Northwest National Laboratory, Microsoft has used AI and high-performance computing to model 32 million new candidate materials to accelerate the search for more efficient rechargeable battery material. This project supports Microsoft’s broader corporate objective to compress 250 years of chemistry research into the next 25 years.

Full disclosure: Microsoft is a client of Moor Insights & Strategy, but this article reflects my independent viewpoint as an analyst.

Azure Quantum Elements

For this project, Microsoft researchers used something called Azure Quantum Elements, which is designed to accelerate scientific discovery. While it currently combines AI and traditional HPC, it will eventually use a Microsoft quantum supercomputer when one becomes available. Meanwhile, Azure Quantum Elements is playing a crucial role in battery lithium-ion research by increasing HPC cluster size and using AI for better reasoning. At the same time, Microsoft Copilot AI simplifies handling data, writing code and running simulations.

Azure Quantum Elements brings together the specific strengths of AI, high-performance computing and quantum computing.MICROSOFT

Azure Quantum Elements focuses on technical challenges created by the need for scale, speed and accuracy:

  • Scale is significant because it determines how many candidates can be screened in the search for a new molecule or material by scaling up to millions of candidate materials instead of just thousands. That level of scaling is important because researchers must model systems large enough to capture the complexity of defects or chemical heterogeneity inside a material.
  • Speed refers to the ability to simulate and analyze materials by speeding up some chemistry simulations by as much as 500,000 times. Providing a fast computational workflow allows researchers to work efficiently. The goal is to accelerate discovery by rapidly processing vast amounts of simulation data on many materials to more quickly identify promising candidates. Speed also allows a more efficient interaction between AI and HPC.
  • Accuracy isn’t entirely precise because some quantum mechanical aspects of chemical systems must be approximated. After all, classical computers can’t simulate these things, either. In the future, integrating quantum computing with AI and HPC will provide the necessary accuracy.

From 32 Million Candidates To One

The Microsoft Azure research team set out to create better solid-state electrolytes for use in lithium batteries. The team used ionic substitution to replace select atoms in 200,000 known crystals, using 54 atoms of potential electrolytes as substitutes. Through this process, the researchers created over 32 million new materials that had to be reduced to a much smaller and more manageable number for the eventual handoff to PNNL. Microsoft used AI acceleration to assay the materials for stability because traditional physics-based models—even running on HPC—weren’t fast enough for such a monumental task. In projects like this, AI is a fast and powerful tool for predicting material properties such as electrochemical stability, bandgap, electrochemical reactivity, energies and forces. Microsoft used AI to replace quantum chemical calculations in the HPC simulations, making this approach 15,000 times faster than traditional means.

Graphic showing the process of narrowing down material candidates from 32 million to a single new lithium-ion material candidate MICROSOFT

Using this process, the pool of materials was filtered down to half a million stable candidates. Those 500,000 were further screened for electrochemical stability using an AI-based screening process, resulting in 800 promising candidates. Although AI is fast and accurate, some residual errors can creep in for Schrödinger’s equation and quantum mechanics calculations. This is why the remaining 800 material candidates were processed a second time using a traditional physics based HPC pipeline for additional insights into the behavior of the materials.

In this phase, researchers used an AI-enabled screening pipeline to characterize the novel materials. The pipeline first utilized predictive modeling to rapidly assess candidates, followed by more accurate physics-based simulations for verification and, finally, molecular dynamics studies to evaluate fundamental dynamic properties and structural fluctuations. This phase narrowed the field to 18 candidate materials.

Microsoft gave six of the 18 candidates to PNNL researchers, who further narrowed the group down to a single material that contained 70% less lithium than current lithium-ion batteries. PNNL has since synthesized the material and used it to create a solid-state battery that is currently being tested.

Where This Research Could Lead

Both AI and HPC were vital technology components for this project. On the AI side, the researchers used a pipeline that Microsoft designed for molecular simulation and predicting energies and forces. HPC was used to drive simulations incorporating AI as well as traditional simulations that performed quantum chemistry calculations.

You can imagine the complexity of creating a new material using this process from beginning to end. To simplify it, AI in the form of a large language model copilot was used to remove many of the barriers that made the process difficult. It also eliminated the need for experts to understand what types of screening steps were needed and which calculations were necessary at each step in the process. The AI copilot also eliminated the need for scientists to figure out how to use and configure tools and how to assemble the complex workflows needed to conduct the necessary science.

It took only a week using Microsoft’s process with Azure Quantum Elements to create 32 million new candidate structures and to reduce them to 800 stable materials. Microsoft estimates that the same process would take 20 years to accomplish without AI acceleration.

It is staggering to think that the process will get even better over time. Azure Quantum Elements provides an onramp to quantum computing for experimentation using existing quantum hardware. It will also include priority access to Microsoft’s quantum supercomputer whenever it becomes available, perhaps by the end of the decade. When scaled quantum computing does come into play, it will deliver breakthrough accuracy for modeling the forces and energies within highly complex chemical systems. That will provide valuable insights currently unavailable using classic computers—and potentially all sorts of breakthroughs in materials science, pharmaceuticals and more. That’s why Microsoft’s project described here has implications well beyond a single new lithium material for batteries. The possibilities are limitless.

Paul Smith-Goodson
+ posts

Paul Smith-Goodson is the Moor Insights & Strategy Vice President and Principal Analyst for quantum computing and artificial intelligence.  His early interest in quantum began while working on a joint AT&T and Bell Labs project and, during 360 overviews of Murray Hill advanced projects, Peter Shor provided an overview of his ground-breaking research in quantum error correction. 

Patrick Moorhead
+ posts

Patrick founded the firm based on his real-world world technology experiences with the understanding of what he wasn’t getting from analysts and consultants. Ten years later, Patrick is ranked #1 among technology industry analysts in terms of “power” (ARInsights)  in “press citations” (Apollo Research). Moorhead is a contributor at Forbes and frequently appears on CNBC. He is a broad-based analyst covering a wide variety of topics including the cloud, enterprise SaaS, collaboration, client computing, and semiconductors. He has 30 years of experience including 15 years of executive experience at high tech companies (NCR, AT&T, Compaq, now HP, and AMD) leading strategy, product management, product marketing, and corporate marketing, including three industry board appointments.