Lawrence Livermore National Laboratory (LLNL) has been selected to lead a project that will receive $4.1 million in funding from the U.S. Department of Energy Advanced Research Projects Agency-Energy (ARPA-E) as part of the Quantum Computing for Computational Chemistry (QC3) program.
QC3 seeks to develop and apply quantum algorithms to accelerate simulations of chemistry and materials science to advance commercial energy applications ranging from superconducting power lines, advanced batteries, engineered rare-earth magnets and breakthrough catalytic systems.
LLNL will develop quantum and machine learning-accelerated software tools and apply them to discovering ultra-strong, lightweight magnets that are crucial for electronic motors, generators and high-performance information technology. The core innovation is a hybrid classical-quantum algorithm that can accurately predict material performance.
The result could have a huge impact on how America uses energy.
"Anytime you want to convert energy between electrical forms and mechanical forms, like in wind turbines, electric vehicles or hydro power, you need to have a magnet that mediates that process," said LLNL scientist and project lead Ilon Joseph. "If we can do much better calculations of magnetic materials science, we can find new kinds of magnetic materials that can power our energy technology."
New magnet materials could circumvent China's critical material supply chain and offer improvements in terms of weight, strength, robustness and resistance to corrosion.
Even slight enhancements could also decrease the resources needed to power artificial intelligence (AI) and information technology (IT). Much of the energy consumption in AI and IT comes from writing and erasing information stored in memory. For MRAM-based chips, which store data using magnetic states, reading and writing requires flipping the magnetization of tiny thin-film magnets. Because AI and IT are predicted to dominate U.S. electricity consumption by the end of the decade, magnetic memory that takes less energy to flip - even by 20% - would lower energy costs significantly.
To discover these new magnetic materials, the team at LLNL is combining various fields of expertise. Researchers at the Laboratory created some of the most advanced codes in the world for simulating electronic structure and realistic materials at the atomic scale. Those tools currently run on El Capitan, the most powerful supercomputer in the world.
"We will connect our state-of-the-art electronic structure simulation code running on high-performance computing systems, such as El Capitan, and offload hard quantum aspects of the problem to quantum frameworks," said LLNL scientist Alfredo Correa Tedesco. "Of course, making those quantum resources work is the most challenging part - but it is also where we have the most to gain in terms of capabilities."
Adapting these materials simulations to run on a quantum computer will offer even better performance. The magnetic spins present in a material represent a many-body quantum system, and, while modeling them with a classical computer is challenging, modeling them with a quantum computer is efficient - a natural fit.
However, almost none of the algorithms that we use on today's classical computing hardware will be good for quantum computers. LLNL's main task lies in the translation from the classical to the quantum algorithm. For example, Joseph has a track record of developing efficient quantum algorithms for solving the partial differential equations needed to simulate fluids and plasmas. He will focus on developing and implementing efficient quantum algorithms for the direct simulation of quantum magnets.
For useful quantum calculations, the scientists will need to focus on quantum error correction, which is essential to obtain a realistic calculation that beats a classical computer. With many physical qubits - on the order of 10,000 - they plan to group them together and create enough redundancy to generate 100 so-called "logical qubits". While some of the physical qubits might be wrong, the error correction protocol ensures that the physical calculation comes together to form a correct solution in terms of logical qubits.
That requires significant quantum hardware that, as of today, is not yet available. The LLNL researchers expect to begin working with a prototype from their hardware partner, one of the leaders in the field of neutral atom computing, in about a year. Then they'll have the remaining two years of the project to make their algorithm work, tying the results of the quantum computation to a machine-learning algorithm that will flag magnetic materials with the potential to transform the energy landscape.
"This is a project that's almost on the edge of the impossible. We're on the cusp," said Joseph. "But even if we fail, if we can prove we are on the path to making a quantum computer that can do these calculations within the next 2-3 years, that will be a major victory."
