AI Cuts Quantum Computing Steps for Complex 144-Qubit Codes

Michael Doherty and colleagues at University College London, in a collaboration between University College London and Quantinuum, have developed a method to prepare stabilizer states using fewer quantum operations than currently possible. Their approach uses artificial intelligence to optimise the process of converting a graph representation of the quantum state into an actual quantum circuit. This advancement enables the synthesis of code states for larger quantum error correcting codes, including the 144-qubit gross code, exceeding the capabilities of existing methods. Results indicate their AI-driven techniques can reduce the number of two-qubit gates required by up to a factor of 2.5, while maintaining a shallow circuit depth, representing a key step towards practical quantum computation Efficient quantum circuit synthesis enables preparation of 144-qubit error correcting codes Two-qubit gate counts for synthesising quantum circuits have been reduced by up to a factor of 2.5, a substantial improvement over prior methods limited to approximately 20 qubits. This breakthrough sharply expands the scale of achievable quantum computations, enabling the preparation of stabilizer states for a 144-qubit quantum error correcting code; previously, synthesising states of this size was intractable. Dr. Patrick Draper and Dr. Koushik Pavan from the University of Toronto developed QuSynth, a new method combining reinforcement learning and Monte Carlo tree search to efficiently navigate the complex process of converting a graph representation of a quantum state into a physical circuit. The technique successfully synthesised code states for both the 23-qubit Golay code and the larger 144-qubit gross code, demonstrating scalability and achieving reductions in two-qubit gate counts while maintaining low circuit depth, an important metric for minimising errors. QuSynth tackles the complex task of building stabilizer states, specific, well-defined configurations of qubits used for quantum error correction, by initially representing them as graph states; visualise this as a map of roads showing how different cities, the qubits, are linked. Systematic simplification of this ‘map’ occurs through a process called graph decimation, effectively removing connections to create a more efficient circuit, relying on applying two-qubit Clifford gates at each step to reshape the graph, an approach locally equivalent to preparing arbitrary stabilizer states with the same gate count and depth. Stabilizer states are crucial for quantum error correction because they allow errors to be detected and corrected without destroying the quantum information. These states possess properties that make them resilient to certain types of noise, a significant challenge in building practical quantum computers. The choice of error correcting code, such as the Golay or gross code, dictates the level of protection against errors and the overhead in terms of qubit requirements. The gross code, with its 144 qubits, offers a higher degree of error correction but demands significantly more resources to implement. Prior to this work, synthesising the necessary quantum circuits for such large codes was computationally prohibitive, hindering progress in fault-tolerant quantum computation. Clifford gates are particularly important in this context as they form a universal set of gates for preparing stabilizer states and are relatively easy to implement on current quantum hardware. The core of QuSynth’s efficiency lies in its intelligent search algorithm. The method begins with a graph state representing the desired stabilizer state. This graph state is essentially a blueprint for the quantum circuit. The algorithm then iteratively applies two-qubit Clifford gates to simplify the graph, reducing the number of connections (edges) while preserving the underlying quantum state. At each step, the algorithm faces a choice of which Clifford gates to apply. A naive approach would involve trying all possible gates, but this quickly becomes computationally intractable as the number of qubits increases. Instead, QuSynth employs a combination of reinforcement learning and Monte Carlo tree search. Reinforcement learning allows the algorithm to learn from experiences, identifying gate choices that lead to more efficient circuits. Monte Carlo tree search helps to explore the vast search space of possible gate sequences, focusing on promising paths and avoiding unproductive ones. The combination of these techniques allows QuSynth to efficiently navigate the complex landscape of possible circuits and find solutions with minimal two-qubit gate counts and shallow depths. Scaling synthesised stabiliser states for improved quantum error correction Preparing quantum states for error correction represents a vital step towards building practical quantum computers, though the computational resources needed to actually run QuSynth itself remain unclear. Achieving synthesis alone doesn’t guarantee a usable quantum state, raising concerns about the practicality of scaling this approach further; acknowledging QuSynth’s computational demands remain unclear is sensible, as building the circuits is only half the battle. Nevertheless, this method demonstrably synthesises larger states than previously possible. The computational cost of running QuSynth, particularly for very large qubit numbers, is an area for future investigation. While the method reduces the complexity of the resulting quantum circuit, the process of finding that circuit can still be demanding. Further optimisation of the search algorithm and potential parallelisation strategies could help to mitigate these computational costs. Moreover, the fidelity of the synthesised states, how closely they match the ideal state, needs to be carefully assessed. Imperfections in the quantum gates and noise in the system can introduce errors that degrade the performance of the error correction code. Characterising and mitigating these errors is crucial for realising the full potential of this approach. This method demonstrably expands the scale of stabiliser states, a key building block for quantum error correction, that can be practically synthesised. Fewer two-qubit gates, prone to error, and maintaining low circuit depth constitute a significant advance. This development establishes a new benchmark in preparing quantum states for error correction, streamlining the process and reducing the computational burden by representing these states as graph states and employing artificial intelligence to optimise circuit construction. Future work will likely focus on verifying the reliability of these complex quantum configurations and assessing the fidelity of these larger synthesised states. The expansion of achievable stabiliser states, the building blocks of strong quantum computation, is a key step forward. The ability to synthesise a 144-qubit code state opens up possibilities for exploring more sophisticated error correction schemes and ultimately building more robust and scalable quantum computers.

This research contributes to the broader effort of translating the theoretical promise of quantum computation into a practical reality. The researchers successfully created a method for preparing complex quantum states, known as stabiliser states, using up to 144 qubits. This matters because fewer two-qubit gates, reduced by up to 2.5 times compared to previous methods, mean less potential for error in quantum computations. The technique, employing a combination of graph states and artificial intelligence, expands the scale of states achievable for quantum error correction. Future work will concentrate on verifying the reliability of these larger states and improving the efficiency of the synthesis process itself. 👉 More information🗞 Fast stabilizer state preparation via AI-optimized graph decimation🧠 ArXiv: https://arxiv.org/abs/2603.17743 Tags: