Quantinuum, DeepMind Scientists Use AI to Minimize Tricky T Gates in Step Toward Practical Quantum Computing

Insider Brief

• The Google DeepMind and Quantinuum partnership made progress on using artificial intelligence for developing quantum computers.
• The team said they have created a new techniques that automates the optimization of quantum circuits, specifically focusing on reducing the number of T gates, which are essential, but challenging to implement.
• The DeepMind-Quantinuum partnership could provide the type of scientific prowess needed for solving other problems in “quantum AI”.

Google DeepMind and Quantinuum released their findings on research that demonstrates how artificial intelligence (AI) can significantly advance the development of quantum computers. The team also suggests this is a step toward integrating AI and quantum computing.

According to the scientists, who published their findings on the pre-print server ArXiv, their new approach automates the optimization of quantum circuits, focusing on reducing the number of T gates, or π/8 gates, which are implemented in quantum circuits as operations that add a specific phase to the state of a qubit. These gates are pivotal yet challenging to implement due to their high resource demands.

T gates, essential for achieving universal fault-tolerant quantum computers, have long been considered the most ‘expensive’ gates in quantum computing, both in terms of time and resources. The universality of quantum computers—a measure of their ability to perform any calculation—is crucial to create what some experts term, “quantum practicality,” or quantum computers that are competitive with classical computing systems at certain tasks.

The recent research introduces the AlphaTensor-Quantum method, an AI-based solution that employs deep reinforcement learning to minimize the T count—or the number of T gates used in a quantum circuit. This advancement not only slashes the resources required for quantum circuit implementation but also signifies the first large-scale application of AI for T count reduction. The AlphaTensor-Quantum algorithm has outperformed existing state-of-the-art T-count optimization methods and equaled the best human-devised solutions across various applications, suggesting a potential shift away from manual or hybrid approaches towards fully automated quantum circuit optimization.

This breakthrough is particularly timely as quantum processors become increasingly capable, highlighting the role AI systems can play in writing efficient code for quantum computations. Moreover, it underscores the utility of AI models in leveraging the computational power of Quantinuum’s H-Series systems, which are among the most powerful in the world based on quantum volume and other metrics.

The implications of this research are vast, promising a significant reduction in the costs associated with quantum computing. In standard benchmark sets of quantum circuits, the AlphaTensor method has reduced costs by 37%, and by 47% in circuits relevant for elliptic curve cryptography. This cost reduction is applicable across nearly all quantum computing platforms, given their reliance on T gates for achieving universality.

A real-world application of this technology can be seen in quantum chemistry, where the model has matched human expertise in minimizing the T count for simulations, such as that of the FeMoco molecule crucial for nitrogen fixation in fertilizer production.

Scheduled for publication in Nature Communications, the paper offers a view of not just the practical value of minimizing T count but also the broader potential of AI in enhancing quantum computing capabilities.

Deep Partnership, Expansive Potential

While the findings that show the unique technological partnership between AI and quantum computing are intriguing enough, the real breakthrough might be in the crafting of the partnership between these AI and quantum computing leaders and the crafting of a collaboration uniquely qualified for exploring some of the sticky issues facing quantum AI. In fact, this research marks the first collaboration of its kind between Google DeepMind and a commercial quantum company outside Google. According to information provided by Quantinuum, after recognizing the potential of DeepMind’s AlphaTensor AI system in addressing the optimization of T gates, Quantinuum initiated the collaboration with the team at DeepMind.

Both the opportunities and problems associated with quantum AI should give the teams plenty of opportunities for further collaborations. While the theoretical potential of combining quantum computing and AI is immense, the practical feasibility and scalability of creating such systems are subjects of intense debate within the quantum community. A collaboration between two heavyweights of their industries — DeepMind in AI and Quantinuum in quantum — may be necessary to address the significant hurdles in the way of tapping quantum AI for practical uses.

The success in tackling the optimization of quantum circuits shown in this work — one of  the key challenges of the merger of quantum computing and AI — could put the Google-Quantinuum partnership in a perfect position to investigate the remaining challenges, particularly as the collaboration and quantum computing, itself, continue to evolve.

For deeper analysis of the work, please read the paper here.

The research team included: Francisco J. R. Ruiz, Johannes Bausch, Matej Balog, Mohammadamin Barekatain, Francisco J. H. Heras, Alexander Novikov, Bernardino Romera-Paredes, Alhussein Fawzi, and Pushmeet Kohli, all of Google DeepMind. Tuomas Laakkonen and Konstantinos Meichanetzidis, all associated with Quantinuum in Oxford and Nathan Fitzpatrick, of Quantinuum in Cambridge. John van de Wetering is part of the Informatics Institute at the University of Amsterdam.