Insider Brief
- Microsoft Quantum and QOLAB proposed a framework for assessing whether logical qubits can scale into fault-tolerant systems capable of useful quantum computations.
- The paper says scalable logical qubits should be evaluated across reliability, scale, capability and performance, rather than through qubit counts or isolated error-rate results.
- The framework calls for repeated error correction, universal fault-tolerant operations, low-latency feedback and a credible route to operating hundreds or thousands of logical qubits.
A framework from Microsoft Quantum and QOLAB proposes that quantum computing’s next progress reports should be judged by more than a rising count of logical qubits.
The paper, posted on arXiv by Microsoft Quantum researchers Matthias Troyer and Chetan Nayak and QOLAB’s John Martinis, introduces the term “scalable logical qubit” for an error-corrected qubit that can support long computations, execute a full set of fault-tolerant operations and be replicated into systems containing hundreds or thousands of logical qubits.
The proposal is ultimately aimed at a more demanding common standard for evaluating an increasingly crowded field of quantum error-correction demonstrations.
The industry’s use of the term, “logical qubit,” has broadened a bit over the past few years. Companies and research groups have shown that encoding information across multiple physical qubits can improve memory, suppress certain errors or outperform an individual physical qubit under defined conditions. According to the researchers, the problem is that, while those are meaningful steps, they may not necessarily demonstrate a system that can run a long and useful quantum program.
In light of this trend, the paper argues that a logical qubit should be assessed across four connected measures — reliability, scale, capability and performance. A result that is strong on one measure, such as an improved error rate in a small system, may still face major obstacles in operating many logical qubits, carrying out universal operations or correcting errors rapidly enough during a computation.
For an industry that often communicates progress through individual figures such as physical-qubit counts, logical error rates or code distances, the team suggests that this approach could offer a more complete test of whether a machine has a credible route to practical fault-tolerant quantum computing.
Beyond a Protected Quantum Memory
Quantum computers use qubits, which are fragile units of information that can be disturbed by environmental noise, imperfect control and measurement errors. Quantum error correction is designed to address that problem by encoding the information of one logical qubit across many physical qubits.
Rather than directly measuring an unknown quantum state, which would disrupt it, error-correction systems repeatedly measure auxiliary qubits for clues about errors. Classical computers then process those measurements, known as error syndromes, and determine what correction or adjustment is needed.
The researchers report that this repeated correction process is central to the definition of a scalable logical qubit. A system cannot rely only on one-time error detection, post-selection or after-the-fact data processing if it is intended to support deep calculations.
Post-selection, for example, can discard experimental runs in which an error is detected. That can improve the quality of the retained results, but it becomes inefficient as a calculation grows longer because the likelihood of an entirely error-free run falls quickly. Error-mitigation techniques can also improve estimates from noisy hardware, the researchers write, but they generally require extra sampling and become difficult to extend indefinitely.
Those tools may remain useful alongside error correction. But the paper suggests they cannot replace repeated quantum error correction for calculations requiring very low failure rates across billions of operations.
The team estimates that the lower end of useful quantum computing could require more than 100 high-quality logical qubits and error rates around one failure in 10 billion operations. More demanding applications, such as large-scale chemistry calculations and cryptanalysis, could require more than 1,000 logical qubits and error rates of one in 1 quadrillion operations or better.
Those estimates offer some idea why small demonstrations of protected quantum memory should not be confused with a system capable of running full-scale algorithms.
Four Measures of Progress
Under the proposed definition, a scalable logical qubit must satisfy four conditions:
- It must be sustained during computation through repeated quantum error correction.
- It must support a universal set of fault-tolerant operations, including operations that allow a program to respond to measurement outcomes.
- It must belong to a code family and hardware architecture in which adding physical qubits predictably reduces the logical error rate.
- It must have a credible route to replication at the scale required for applications.
The latter point addresses how a hardware platform may demonstrate a promising error-correction result with a small group of qubits, but operating hundreds or thousands of such logical qubits requires control electronics, calibration, interconnects, decoding systems and physical hardware that must also grow without undermining performance. This is considered a persistent challenge in quantum computing.
The researchers group these challenges into four dimensions — reliability, scale, capability and performance.
Reliability concerns whether logical error rates are better than the error rates of the underlying physical qubits and whether they can be reduced further by increasing the resources devoted to the code. The paper points toward logical error rates in the range of one in 1 trillion to one in 1 quadrillion as a target for utility-scale systems, depending on the workload.
Scale concerns the ability to operate hundreds or thousands of logical qubits at once without losing fidelity, creating major correlated errors or slowing operations to an impractical degree.
Capability refers to what the logical qubits can actually do. Breaking down capability, the paper outlines a progression from preparing and measuring logical states to storing them, performing a full set of Clifford operations, adding non-Clifford operations needed for universal quantum computing and allowing measurement-conditioned control flow.
That final capability is particularly important. Some quantum algorithms require the machine to measure part of a calculation, interpret the result and rapidly decide what operation to perform next. That depends on a classical decoder that can process error-syndrome data in real time and feed instructions back to the quantum hardware with low enough delay.
Performance covers the speed and eventual cost of fault-tolerant operations. A system may achieve a low error rate, but it could still be too slow or too resource-intensive to be useful for a commercial or scientific workload.
Trade-offs and Limits
The researchers conclude in the paper that no single measure entirely captures fault-tolerant progress.
Allocating a fixed supply of physical qubits across more logical qubits may increase system capacity but deliver less error suppression per logical qubit. Choosing an error-correction code that uses fewer physical qubits may increase the time required to perform logical gates. A more accurate decoder may require more classical computing resources and introduce delays that reduce the benefit of better correction.
Compilation creates another trade-off. More sophisticated software can reduce the number of gates, routing steps or special resource states required for an algorithm. But that optimization can increase offline compilation work or create tighter timing requirements if decisions must be made during execution.
The researchers say that these choices should be evaluated according to the target application rather than treated as a race for a single record. A quantum computer intended for a particular chemistry problem may have different requirements from one designed for cryptanalysis or quantum simulation.
The framework also places cost more explicitly in the discussion. The researchers bring up the Defense Advanced Research Projects Agency’s view of utility-scale quantum computing as the point at which the value of a computation exceeds its cost. Under that approach, the relevant question becomes how much reliable computation a machine can produce per unit of hardware and time.
As for limits, the paper offers definitions, questions and an evaluation framework, but it does not ultimately establish a new universal benchmark, report an experimental result or settle which error-correction code or hardware platform will prevail. That will likely be the goal of future research.
Future work will also need to develop a proposed standard that can be applied uniformly. Hardware approaches differ substantially in qubit type, connectivity, measurement speed, control requirements and error characteristics. Some platforms may make progress on fast feedback before they can show large-scale replication, while others may demonstrate favorable scaling properties before they offer a complete set of fault-tolerant operations.
