Quantum Algorithms Meet AI Chips: A Breakthrough in Simulation

December 4, 2023
Quantum Algorithms Meet AI Chips: A Breakthrough in SimulationQuantum Algorithms Meet AI Chips: A Breakthrough in Simulation

The synergy between AI and quantum technologies (AQ), and the potential possibilities they unlock in molecular simulation, is becoming stronger – thanks, in part, to the power of GPUs. Originally created to support increasingly immersive video games, these same powerful graphics chips are now ushering in a new era in computing centered around simulation and AI.

Recently, SandboxAQ announced a collaboration with NVIDIA to predict chemical reactions in applications such as drug discovery, advanced battery design, and clean energy. As part of the collaboration, we will leverage NVIDIA’s quantum platforms to direct the quantum-mechanical simulation of systems underpinning modern chemistry, biology, and material science using tensor networks and machine learning. We will also explore adding new capabilities to NVIDIA CUDA libraries to co-design this next stage of hardware and software to drive advanced AI applications.  

A Quantum Leap in Simulation

Thanks to their ability to handle dense mathematics and high-bandwidth memory, GPUs have become the workforce behind many of our AQ solutions, including molecular simulation and quantum chemistry – both of which are indispensable tools for driving innovation in the fields of drug discovery and materials science.

For example, the molecular and atomic processes occurring in catalysis, protein-drug binding, or electro-chemistry are described by the theory of quantum mechanics and can, in principle, be predicted by solving their underlying mathematical equations using computers. However, the complexity of these equations can grow so quickly, due to the number of atoms involved, that solving them becomes a hopeless task, even when using the largest classical computers on earth.

Quantum computers can simulate atomic and molecular systems using the language of quantum mechanics directly, thereby circumventing the dramatic increase in complexity. However, realizing this potential will take many more years, or even decades, of research.

While the community awaits the advent of error-corrected quantum computers, recent breakthroughs in machine learning, quantum information theory, and hardware development have shown the potential to revolutionize our abilities to simulate quantum systems on classical hardware today. Our collaboration with NVIDIA will explore the use of novel simulation methods, such as tensor networks, and deep learning for quantum chemistry applications. 

Tensor networks are a class of very powerful algorithms developed in the context of quantum many-body physics — and they excel at solving the type of computational problems that quantum chemists often face. Tensor networks have become the de-facto, state-of-the-art approach for many challenges in computational quantum mechanics, ranging from modeling properties of low-dimensional solids (e.g. for understanding high-temperature superconductivity) to quantum chemistry for large transition metal complexes. More recently, tensor networks have had a range of uses, such as low-rank machine learning, generative modeling, and classification as solvers for nonlinear differential equations like the Navier-Stokes equation for fluids or even for numerical analysis.

Tensor network algorithms typically exploit a fundamental property of nature called locality. In layman’s terms, locality means that far-separated parts of a system, such as two distant atoms in a long molecule, do not influence each other in a meaningful way. Utilizing this property enables the simulation of much larger molecules than is achievable with other approaches. In more technical terms, tensor networks can represent quantum states with an area law of entanglement, with deep connections to the theory of quantum information and quantum computation.

As a computational tool, tensor networks share many similarities with algorithms developed in the areas of machine learning and artificial intelligence, with the development of tensor network methods for machine learning being an active research area. Vice versa, the recent breakthroughs in NLP, computer vision, and generative modeling have sparked a surge of global research efforts to understand how powerful machine learning models like recurrent neural networks or transformer-based architectures have the potential to simulate challenging quantum mechanical systems. 

Our collaboration with NVIDIA will explore how tensor networks, deep learning models like transformers, and combinations and variants thereof can be used to solve the most challenging problems in quantum chemistry and beyond. By leveraging NVIDIA’s latest hardware developments, we can better push quantum simulation to an unprecedented speed and scale. The structure of tensor network algorithms will allow us to optimally leverage distributed hardware accelerators like GPUs. For example, in a recent publication by our research team, we demonstrated how Google’s tensor processing units (TPUs) can be used to perform complex, high-dimensional optimizations of over 600B parameters in less than a day, representing the world’s largest tensor network calculation to date.

AI has already demonstrated its value in innovating effective therapies for existing conditions. For more challenging, “undruggable” conditions like cancer or Alzheimer’s disease, which have little data available for AI to leverage, researchers are turning to quantum technologies, such as those used by our AQBioSim division. Digitally modeling chemical compounds and simulating their interactions with human receptors at the molecular level delivers previously unattainable insights and new data that AI models can use to predict a compound’s efficacy and commercial viability. This greatly accelerates the drug discovery process, lowers R&D costs and risk, and gets potentially life-saving therapies to patients faster.

The same molecular modeling technologies and approaches are being used in materials science to solve major technical challenges. For example, global demand for lithium-ion batteries for consumer electronics, EVs, and energy storage is skyrocketing, prompting manufacturers to develop batteries that are safer, durable, and more powerful. Before they can be mass produced, batteries must be rigorously tested for performance and longevity. But the only way to test how many years a battery will last is to actually test it for years. 

Recently, a leading battery materials and technology company, NOVONIX, partnered with SandboxAQ to enhance its industry-leading battery testing and prototyping technologies with AQ. Leveraging early-stage testing data, our AI and quantum technologies can simulate a battery’s performance under various electrochemical conditions and accurately predict its potential lifespan in weeks instead of years. This enables manufacturers to make better-informed decisions about battery chemistry, materials, processes, cell design, recovery, and technologies at every stage of R&D and manufacturing.

A Quantum Leap in Demand and Adoption

Right now, generative AI applications and large language model training are driving the global demand for GPUs and specialty AI chips from NVIDIA and other manufacturers. 

However, as we progress deeper into the quantum era, commercial and public sector demand for quantum technologies will also stimulate semiconductor sector growth – possibly even after the availability of error-corrected, fault-tolerant quantum computers. As a recent Wall Street Journal story pointed out, GPUs transformed AI. Now they’re here for quantum.

Battery testing is just the beginning for opportunities with simulation. Using quantum molecular simulation to create and test new materials will impact numerous industries, including construction, manufacturing, transportation, chemical, defense, energy, and other sectors – and make products more eco-friendly as well. Quantum simulation will also be used to help tackle larger global and societal challenges, such as climate change.

The potential for AQ to transform industries is only limited by our imagination. None of this would be possible without the powerful data signal processing capabilities of GPUs, which are enabling quantum mechanical equations and accelerating AI. 

To learn more about how GPU-powered AQ solutions are delivering breakthrough innovations across industries, visit our quantum simulation page. 

No items found.