Loss Biasing Tackles Quantum Computer Errors by Swiftly Eliminating Faulty Atoms

Scientists at University of Strasbourg and CNRS, in collaboration with Macquarie University, have demonstrated a new technique for quantum error correction that restores the potential for scalable, fault-tolerant quantum computing. Laura Pecorari led the research, which found that conventional error correction methods can be hampered by the speed and accuracy of modern neutral-atom processors. The investigation reveals that faster quantum error correction cycles inadvertently amplify specific errors, hindering performance and introducing complexities not previously accounted for in theoretical models. To overcome this, the team introduces ‘loss biasing’, a technique which rapidly eliminates problematic quantum states, effectively converting errors into a more manageable form. This approach enables sharply faster error correction cycles with reduced hardware requirements and provides a key set of tools for building practical quantum computers, addressing a critical bottleneck in quantum computer development.

Loss biasing substantially reduces correlated errors and enhances quantum error correction

Following implementation of loss biasing, logical error rates dropped from approximately 10−3 to levels comparable with state-of-the-art erasure-conversion protocols. This represents a significant improvement, bringing the technology closer to the threshold required for fault-tolerant quantum computation. A tenfold gain in quantum error correction performance was achieved by suppressing correlated errors, which previously hindered fault tolerance; conventional approaches struggled with their build-up as cycle times shortened. These correlated errors arise from the interconnected nature of qubits and the non-instantaneous nature of quantum operations, leading to dependencies between errors that are difficult to model and correct. Loss biasing proactively converts spurious Rydberg excitations, temporary high-energy states causing errors, into atom loss via mid-circuit ionization, simplifying the error signal and restoring optimal logical error scaling. Rydberg states are created when atoms are excited to a higher energy level using lasers, and these states are particularly susceptible to unwanted interactions that introduce errors. The process of ionization removes the atom from the computation, effectively erasing the error.

Surface code simulations, utilising code distances from three to nine, revealed that near-perfect autoionization, achieving approximately 100% success, yielded a logical error probability consistent with fault-tolerant scaling. This demonstrates the potential for building large-scale quantum computers with a low probability of errors. Zero ionization, however, resulted in sharply degraded performance, highlighting the importance of the loss biasing technique. Detailed analysis of single surface code plaquettes showed that perfect ionization, whether applied to all qubits or solely to ancilla qubits, eliminated correlated “hook” errors, indicators of non-Markovian noise, completely. Non-Markovian noise refers to errors that are correlated in time, meaning that the probability of an error occurring at one time depends on whether an error occurred at a previous time. Hook errors are a specific type of non-Markovian error that can be particularly difficult to correct. Simulations also reveal that even partial inter-gate ionization, converting most but not all problematic atoms, can achieve near-optimal performance, and applying it solely to ancilla qubits after each gate is sufficient to preserve fault tolerance; in particular, achieving 75 or 50% ionization success after each gate produced logical error probabilities comparable to those obtained with perfect ionization applied mid-cycle. This suggests that the technique is robust and can tolerate some degree of imperfection in the ionization process.

Rydberg excitation removal via mid-circuit ionisation for correlated error mitigation

Loss biasing actively reshapes the nature of errors within neutral-atom quantum processors, addressing challenges arising from increasingly rapid error correction cycles. The technique centres on swiftly converting unwanted Rydberg excitations, temporary high-energy states introducing errors, into atom loss via mid-circuit ionization; a problematic atom is effectively removed from the computation. This isn’t simply deleting information, but transforming errors into what is termed erasure-like noise, akin to deleting letters from a message instead of scrambling them. Erasure-like noise is easier to correct than scrambling, as the location of the error is known. Neutral-atom qubits are created by trapping individual atoms using lasers, and their quantum state is manipulated using pulses of light. The Rydberg states are used to create strong interactions between the atoms, which are necessary for performing quantum computations.

The technique tackles error correction in neutral-atom quantum processors as cycle times increase, revealing that faster cycles can amplify specific errors like Rydberg excitation hopping, hindering the decay of residual Rydberg population and degrading performance. Rydberg excitation hopping refers to the transfer of excitation energy between atoms, which can introduce errors into the computation. Proactively eliminating these troublesome states prevents errors from propagating and amplifying, a phenomenon described as non-Markovian correlated errors. These errors are particularly problematic because they violate the assumption of Markovian noise, which is commonly used in quantum error correction algorithms. Unwanted Rydberg excitations were converted into atom loss, transforming errors into a more manageable erasure-like noise during implementation. The speed of this conversion is crucial, as it must be faster than the rate at which errors propagate.

Loss biasing offers a pathway to simplified quantum error correction through controlled atom loss

To tackle a key problem, the amplification of errors in faster quantum processors, the technique swiftly converts quantum errors into atom loss. Restoring predictable error reduction is vital for building quantum computers capable of tackling complex problems, such as drug discovery and materials science, though simulations suggest that loss-aware decoding will ultimately deliver optimal performance; key experimental validation of this decoding process remains necessary. Loss-aware decoding refers to algorithms that take into account the fact that some qubits have been lost during the computation. Nevertheless, this offers a valuable route towards practical quantum error correction, reducing the hardware demands of maintaining stable qubits, a major obstacle to building useful quantum computers and allowing for shorter, more efficient error correction cycles. Maintaining qubit stability requires precise control of the atoms and shielding them from external noise. This method restores predictable error reduction as computational complexity increases, circumventing issues arising from the amplification of errors in rapidly cycling processors, and offers a pathway towards scalable, fault-tolerant quantum computing with sharply faster error correction cycles. Establishing effective loss-aware decoding strategies, capable of fully utilising this error management approach, now represents the key next step in realising its potential. Further research will focus on optimising the ionization process and developing more sophisticated decoding algorithms to maximise the performance of this promising technique.

The research demonstrated that converting unwanted quantum errors into atom loss, a process called loss biasing, restores predictable error reduction in neutral-atom processors. This is important because faster quantum processors can amplify errors, hindering the development of stable quantum computers. By transforming errors into a more manageable form, loss biasing enables shorter quantum error correction cycles with reduced hardware requirements. The authors suggest that further optimisation of the process, alongside development of loss-aware decoding, will be crucial for realising the full potential of this technique.