Catalysts that could recycle carbon dioxide into clean fuel. Compounds that could boost the power of batteries. Proteins that could transform how we treat disease.
Every day, scientists hunt for new molecules to tackle the world’s toughest challenges. But that search is mindbogglingly complex. Researchers must sift through countless combinations of nature’s building blocks to find the best solution. The number of possible stable molecules and materials is believed to far surpass the number of atoms in the known universe.
To help scientists explore more of that vast search space, Microsoft has announced Azure Quantum Elements. It is a new system that aims to accelerate chemical and materials science today through the scale of Azure High Performance Computing (HPC) and the speed of AI.
And it includes tools that will help scientists prepare for a future where a scaled quantum computer could accurately model the most complex molecules – a task the largest imaginable classical computers could not solve. In that future, customers will be able to tap the system’s quantum supercomputing technology to run simulations with unprecedented accuracy, Microsoft said.
“Our goal is to compress the next 250 years of chemistry and materials science progress into the next 25,” said Microsoft CEO Satya Nadella during the announcement.
Chemistry directly touches more than 96% of all manufactured goods, so the system’s potential impact is huge, said Jason Zander, executive vice president of Strategic Missions and Technologies. Helping scientists accelerate their research could help countless businesses save time and money – and be a catalyst for widespread innovation in our everyday lives.
“With Azure Quantum Elements, scientists can navigate limitless possibilities and narrow them down to the most promising candidates with unprecedented speed,” Zander said. “It’s critical for the planet if we want to have an impact on the biggest issues facing humanity, and it’s critical for most businesses.”
Azure Quantum Elements, which will be available in private preview on June 30, delivers that speed through proprietary software tailored to the needs of chemical and materials scientists and built on Microsoft’s investments in AI, HPC and future quantum technologies, he said.
Researchers can use Azure Quantum Elements to explore more materials with the potential to scale from thousands of candidates to tens of millions and to speed up certain chemistry simulations by 500,000 times, Zander said.
The system can also boost researcher productivity through automated workflows and the new Copilot in Azure Quantum, which lets researchers use natural language to do things like find and visualize data or quickly develop, configure and run simulations.
Azure Quantum Elements has been adopted by industry leaders including BASF and Johnson Matthey.
“Chemistry is in everything,” said Ansgar Schaefer, a vice president at BASF who leads the company’s quantum chemistry research. “To be able to improve products and processes, it’s really about understanding the chemistry behind them on a microscopic level. And the more complex the challenge, the more computing power is required. [Azure Quantum Elements] is a tool that gives us additional required capacity to help advance completely new research approaches and increase the efficiency and speed of development.”
Much of chemistry can be thought of as a molecular game of musical chairs. When atoms get together to form new molecules, their electrons rearrange themselves to make and break the bonds between them.
Scientists rely on computers and intricate calculations to simulate those interactions. In this virtual lab, they can explore different molecular configurations before investing time and resources in experiments.
At Johnson Matthey, an early Azure Quantum Elements customer, scientists use those simulations to discover and develop sustainable technologies like hydrogen fuel cells and catalysts that can slash emissions from cars and trucks.
“If you think about the materials discovery process, you can go into the lab and tinker with different chemicals to try and make something, or you can use fundamental physics to screen the entire periodic table,” said Misbah Sarwar, research lead for Johnson Matthey’s physical and chemical modeling team. “Modeling lets you home in on the kind of materials that the experimentalists should be focusing on and, equally as important, the types they should avoid.”
But there is a catch: the more accurate you want to be, the slower and costlier the computation. That is why computing resources can be a bottleneck in the creation of new chemicals and compounds, Sarwar said.
Her team recently removed that bottleneck by moving its research from local clusters into the Azure HPC cloud. Its scale has delivered a two-fold speedup for certain quantum chemistry calculations, accelerating their delivery of insights to their colleagues in the lab.
Speed isn’t the only gain, she added. Her team can now simulate much larger – and therefore much more realistic – chemical systems. That helps their research into products like catalytic converters, the devices that destroy harmful gases before they escape from a car’s tailpipe.
Johnson Matthey scientists are exploring how certain molecules can improve the catalyst’s performance. But adding multiple types of these molecules makes it increasingly hard to model their effect on the catalyst as well as each other. With the scale of the cloud, they can now untangle those complex interactions, Sarwar said.
“I felt like I was working on a backlog all the time, and now I can get these projects moving and make an impact on experimental work that’s happening now as opposed to coming in a bit later in the development cycle,” Sarwar said.
BASF, a renowned chemical company, wants to use Azure for demanding calculations that push the limits of its custom supercomputer.
“When you need additional computing power by an order of magnitude on a short timescale, that’s only possible through cloud resources,” Schaefer said.
BASF is using the HPC capabilities of Azure Quantum Elements to further expand its research and development capabilities and accelerate innovation to bring more sustainable products to market faster, Schaefer said. The extra computing muscle will help to get a finer-grained picture of vast and intricate chemical reaction networks or the complex relationships between the composition and performance of materials. That in turn could speed the development of products like catalysts and battery materials.
Stephan Schenk, BASF’s product manager for high performance computing (HPC), thinks the cloud can effectively put a supercomputer in the hands of every chemist. He has spent 25 years watching HPC evolve and become more widespread.
“But the software and the environments required are a significant barrier to companies and research institutions. While we at BASF have a dedicated team of people with the right skills to operate the ecosystem, many others don’t,” Schenk said. “With the Azure cloud, you instantly get the full-fledged environment that has some of the most common tools. That lowers the barrier to entry and will significantly contribute to the democratization of these tools.”
…these models learn the language of nature, which is quantum physics.
Chemical reactions occur across a wide range of timescales. Some are measured in nanoseconds, others in decades.
Modeling reactions at a molecular level becomes increasingly difficult the longer they get, said Matthias Troyer, Microsoft technical fellow and corporate vice president of quantum. Traditional computational methods are too slow and costly to simulate step-by-step anything that takes longer than a blink of an eye.
Azure Quantum Elements includes AI models and code developed in collaboration with Microsoft Research’s AI4Science team designed to accelerate simulation. Chemists can use these foundation models to simulate molecular models more efficiently, with more complexity and on longer timescales than existing methods, Troyer said.
Different models can help speed up the various stages of research. For example, a lightweight AI model could be used at the start of a project to quickly pare down a list of initial candidates. If a scientist is only looking for molecules with a particular boiling point, the model could give a quick thumbs up or down on millions of candidates by running a relatively simple simulation. As they get closer to moving from the computer to the lab, they can run the slower, more detailed simulations on only the strongest candidates.
The machine learning algorithms were trained on large datasets containing information about various chemical systems. Scientists can feed the trained model the types and locations of atoms in a molecule, and the AI will predict their energy and forces – the key ingredients for most molecular simulations. AI can find those ingredients in a fraction of the time compared with traditional methods.
These AI models hint at a future where designing a new molecule could be as easy as asking Bing Image Creator to paint a picture, Troyer said.
“Just like large language models are trained on human language, these models learn the language of nature, which is quantum physics,” he said. “If we train an AI to understand quantum physics and quantum chemistry, then that AI can help us predict the properties of materials, it can help us design molecules and materials.”
But it will take a quantum computer before AI becomes truly fluent.
Troyer and other scientists believe that solving the problems of chemical and materials science will be the first ‘killer app’ for quantum computing. And that makes sense.
“A quantum computer works based on the same principles as how nature works,” he said. “So it is natural that the native application of a quantum machine is to simulate the behavior of electrons in a molecule or a material.”
No quantum computer exists today that could solve problems of this scale. But Azure Quantum Elements offers tools that can run on today’s classical computers in the cloud, and can help scientists prepare for that quantum future, Troyer said. These tools can determine whether a quantum computer would be needed for a particular problem. It could turn out that the researcher’s problem could be solved today on a classical computer. If so, the tools will suggest the best methods and algorithms to use.
If the problem does require a quantum computer, the resource estimator will determine how many qubits – the basic building blocks for quantum computation – are required. Either way, the researcher will gain crucial information to help develop and refine algorithms ready to run on tomorrow’s scaled quantum computers, Troyer said.
The ultimate dream would be to accurately predict and understand chemical reactions at the atomic level. That could bypass years of slow and costly trial and error in the lab and unleash a wave of transformative new materials and compounds.
Troyer cautions that much challenging work remains before that future arrives. Microsoft is focused on engineering a quantum supercomputer now and just achieved the first of six milestones on its roadmap. It also introduced a new performance metric to provide an objective understanding of the speed and reliability of a quantum supercomputer. The reliable Quantum Operations Per Second (rQOPS) metric measures how many reliable operations a quantum computer can perform accurately and consistently within a one-second timeframe.
Microsoft’s goal is that its first quantum supercomputer will deliver one million rQOPS, Troyer said. But today every known quantum computer performs at the same rQOPS value: Zero.
As the industry works to push that value upwards, Troyer hopes Azure Quantum Elements will empower all chemists to achieve more.
“Chemistry is hard,” he said. “The challenge chemists face is having to search through trillions of potential candidates, and they don’t have time to learn complicated new tools. We want to accelerate their work with a platform that every chemist in every lab can use.”
Top image: A Johnson Matthey employee applies a catalyst coating to a supporting structure – a key component of a catalytic converter. (Credit: Johnson Matthey)
Related
Leave a Reply