Quantum Processor Achieves 98.4% Leakage Reduction with Integrated Units, Raising Error-Correction Fidelity to 99.2%

Quantum computers promise revolutionary computational power, but maintaining the fragile quantum states of qubits remains a significant challenge. Yuejie Xin, Sean L. van der Meer, and Marc Serra-Peralta, all from Delft University of Technology, alongside colleagues, now demonstrate a crucial step towards more reliable quantum computation by tackling a major source of errors, known as leakage. The team developed a novel leakage reduction unit that actively suppresses these errors during qubit measurement, achieving a remarkable 98. 4% removal fraction without slowing down the process. This breakthrough, which combines innovative hardware with advanced neural network decoding, successfully reduces logical error rates in quantum error correction experiments, paving the way for more stable and powerful quantum computers.

Surface Code Corrects Leakage Errors Effectively

This research focuses on improving quantum error correction, specifically using the surface code, and investigates methods to mitigate leakage errors, which occur when qubits unintentionally transition out of their computational state. The surface code is a promising approach due to its suitability for implementation on two-dimensional architectures. Researchers employed neural networks as decoders, enabling them to learn the most likely logical state from noisy qubit measurements and improve error correction performance. Both memory experiments, focusing on preserving quantum information over time, and stability experiments, assessing robustness against errors during operations, were conducted to provide a comprehensive evaluation of the system’s performance.

The team implemented leakage reduction units (LRUs) to actively suppress leakage errors by driving qubits back into their computational subspace, and integrating data from these LRUs into the neural network decoder further enhanced performance. Data balancing techniques, such as sample importance weighting, were used to ensure all error patterns are adequately represented during training. The decoder architecture utilizes a multi-stream bi-directional recurrent neural network (RNN), allowing for efficient processing of complex data, and combining memory and stability experiments provides a more complete assessment of logical qubit performance.

High Fidelity Leakage Reduction for Qubits

This research demonstrates a high-fidelity, zero-overhead leakage reduction unit (LRU) for superconducting qubits, effectively addressing a significant source of errors that limits quantum error correction. By integrating an active reset mechanism directly into the qubit measurement pulse, the team achieved 98. 4% leakage removal without compromising state assignment fidelity, which remained at 99. 2%. This LRU operates concurrently with standard qubit measurement, avoiding any additional time overhead.

Experimental validation incorporated the LRU into two quantum error correction benchmarks, demonstrating practical benefits for both memory and stability experiments. In repetition-code memory experiments, combining the LRU with a three-round readout protocol yielded the best performance, while in stability experiments, the LRU successfully mitigated the propagation of errors caused by leakage. These results establish a practical method for combating detrimental error channels in superconducting quantum processors and improving the scalability of near-term quantum error correction implementations. The authors acknowledge that microwave crosstalk may influence the performance of parallel LRUs and requires further characterization. Future work will focus on deploying this LRU-enhanced measurement across larger-scale quantum error correction codes and combining it with other leakage reduction techniques applicable to data qubits. Careful optimisation of system parameters to further enhance leakage removal efficiency also remains an important area for investigation.