Microsoft uses AI plus HPC to analyze 32 million new materials

Microsoft and Pacific Northwest National Laboratory are collaborating to use AI and high-performance computing (HPC) technology to model 3,200 new candidate materials to accelerate the development of high-efficiency rechargeable battery materials. . This collaborative project aims to support Microsoft's future development goals and incorporate 250 years of human chemistry research history into the data model to provide strong support for future scientific research.

Azure Quantum Elements

For this project, Microsoft researchers leveraged the Azure Quantum Elements platform, which is designed to accelerate scientific discovery. Currently, the platform uses artificial intelligence (AI) with traditional high-performance computing (HPC), but it aims to be compatible with Microsoft's quantum supercomputers in the future. In addition, Azure Quantum Elements also scales up HPC clusters and leverages AI for high-quality inference to play an important role in lithium-ion battery research. In addition, Microsoft's Copilot AI is also responsible for simplifying specific operations such as data processing, code writing, and simulation running.

Azure Quantum Elements focuses on solving the technical challenges posed by large-scale, high-speed, high-accuracy requirements:

• Scale: Processing scale is very important and will directly determine new molecules or the specific scope of the material discovery. To expand candidate materials from thousands to millions, the ability to scale is critical. Researchers must model systems large enough to accurately capture complex factors such as defects or chemical heterogeneity within the material.
• Speed: Speed ​​refers to accelerating specific chemical simulations 500,000 times to simulate and analyze the basic properties of materials. Faster computing processes help researchers work more efficiently. The goal of the Quantum Elements Platform is to accelerate discovery and efficiently identify promising candidate materials by rapidly processing large amounts of simulation data for a variety of materials. Speed ​​will also determine the efficiency of the interaction between AI and HPC.
• Accuracy: Accuracy does not emphasize complete accuracy. This is because some quantum mechanical effects in chemical systems can only obtain approximate results. Current classical computers are still unable to accurately simulate such scenarios, so further improvement in accuracy in the future requires the full integration of quantum computing with AI and HPC.

Find answers from 32 million candidate materials

The Microsoft Azure research team is exploring the ideal solid-state electrolyte for lithium battery manufacturing. The team replaced specific atoms in 200,000 known crystals through ion substitution and used 54 potential electrolyte atoms as replacement options. During this process, researchers created a total of more than 32 million new materials, but such a huge candidate library is too broad and needs to be further screened and reduced to a more manageable size before it can be handed over to Northwest National Laboratory. Considering that traditional HPC physics models are not enough to quickly solve such a large-scale problem set, Microsoft decided to use AI to accelerate the stability analysis of materials. In such projects, AI will become a fast and powerful tool option for predicting material properties such as electrochemical stability, band gap, electrochemical reactivity, energy and force. By using AI to replace quantum chemical calculations in HPC simulations, Microsoft successfully increased the screening speed to 15,000 times that of traditional methods.

Through this process, the material library has been initially screened, leaving 500,000 stable candidates. The 500,000 candidate materials were then further screened for electrochemical stability using an AI screening process, resulting in 800 promising candidates. Although AI algorithms are fast and accurate, there may be some errors due to the limitations of quantum mechanical calculations. Therefore, in order to further analyze the physical and chemical properties of materials, we need to use the HPC pipeline based on traditional physical effects to perform secondary processing on the remaining 800 candidate materials.

At this stage, the researchers used an AI screening process to characterize various new materials. The process begins with a rapid evaluation of candidate materials using predictive models, followed by more accurate validation with physical simulations, and finally evaluates their fundamental dynamic properties and structural fluctuations through molecular dynamics studies. By the time this stage was reached, the candidate materials had been narrowed down to 18.

Microsoft then selected six materials and gave them to researchers at Northwest National Laboratory, who ultimately selected a single material with 70% less lithium content, making it more ideal than current lithium-ion batteries.

The vast world has a lot to offer

Both AI and HPC play an important role in the project. The researchers leveraged Microsoft's pipeline designed for molecular simulation and energy/force prediction to implement AI research. HPC is responsible for supporting traditional simulation links, including tasks related to AI simulation results and quantum chemical calculations.

As you can imagine, the complexity of the discovery process of new materials and the huge amount of data processing. In order to simplify the process, AI auxiliary tools based on large language models can solve various difficulties and obstacles, while also replacing human experts in type screening and step-by-step calculation tasks. Scientists can quickly get help configuring tools and designing feature sets to dramatically accelerate complex processes in scientific research.

With the Microsoft Azure Quantum Elements platform, the creation of 32 million new candidate structures and the selection of 800 stable materials took just one week. Microsoft estimates that without the support of AI technology, it would take 20 years for pure manpower to complete such a screening process.

What is more worth looking forward to is that as time goes by, the execution efficiency of the entire process will become increasingly higher. The Azure Quantum Elements platform also reserves quantum computing experimental interfaces for existing quantum hardware. In this way, when Microsoft's quantum supercomputer is finally deployed, the platform will have priority access to quantum computing power. As large-scale quantum computing begins to play a practical role, it is believed that this technology will provide breakthrough accuracy guarantees for modeling force effects and energy in highly complex chemical systems. The resulting valuable insights that cannot be achieved by existing classical computers are expected to deliver more unprecedented new results in fields such as materials science and pharmaceuticals. Because of this, the impact of Microsoft's Quantum Elements project has gone far beyond the research scope of new battery lithium materials, and will definitely bring endless space for exploration to all walks of life.

The above is the detailed content of Microsoft uses AI plus HPC to analyze 32 million new materials. For more information, please follow other related articles on the PHP Chinese website!