Abstract
The speed of a fault-tolerant quantum computer is dictated by the reaction time of its classical electronics, that is, the total time required by decoders and controllers to determine the outcome of a logical measurement and execute subsequent conditional logical operations. Despite its importance, the reaction time and its impact on the design of the logical microarchitecture of a quantum computer are not well understood. In this work, we build, for a surface code based architecture, a model for the reaction time in which the decoder latency is based on parallel space- and time-window decoding methods, and communication latencies are drawn from our envisioned quantum execution environment comprising a high-speed network of quantum processing units, controllers, decoders, and high-performance computing nodes. We use this model to estimate the increase in the logical error rate of magic state injections as a function of the reaction time. Next, we show how the logical microarchitecture can be optimized with respect to the reaction time, and then present detailed full-system quantum and classical resource estimates for executing utility-scale quantum circuits based on realistic hardware noise parameters and state-of-the-art decoding times. For circuits with $10^{6}$--$10^{11}$ $T$ gates involving 200--2000 logical qubits, under a $\Lambda=9.3$ hardware model representative of a realistic target for superconducting quantum processors operating at a 2.86 MHz stabilization frequency, we show that even decoding at a sub-microsecond per stabilization round speed introduces substantial resource overheads: approximately 100k--250k additional physical qubits for correction qubit storage in the magic state factory; 300k--1.75M extra physical qubits in the core processor due to the code distance increase of $d$ to $d+4$ for extra memory protection; and a longer runtime by roughly a factor of 100.
The Classical Bottleneck in Fault-Tolerant Computing
Fault-tolerant quantum computers (FTQC) are inherently hybrid systems. As the quantum processing unit (QPU) operates, it continuously generates "syndromes"—classical data indicating where physical errors may have occurred. A classical decoder must process this data in real-time to inform the QPU of the necessary corrections.
The primary challenge is latency. If the classical decoding process is slow, logical qubits must remain idle while waiting for instructions. During this idling period, they continue to accumulate physical errors, necessitating a larger "code distance" (more physical qubits per logical qubit) to maintain the desired logical error rate. While many theoretical models assume a negligible "reaction time," this research demonstrates that classical latency is a dominant factor in determining the physical scale and execution speed of utility-scale quantum computers.
Modeling the Reaction Time Loop
To quantify these effects, the authors developed a comprehensive model for the reaction time, defined as the total duration from the completion of a quantum measurement to the execution of a conditional operation based on that measurement's result.
The model incorporates several critical components:
- Communication Latency: This was experimentally characterized using high-performance hardware, including an OPX1000 controller and an NVIDIA Grace Hopper server.
- Decoding Latency: The authors utilized parallel space- and time-window decoding techniques. These methods decompose massive decoding tasks into smaller, overlapping segments that can be processed concurrently.
- Microarchitecture Optimization: Using the TopQAD software suite, the team estimated the resources required for a surface-code-based architecture. They specifically analyzed how the Magic State Factory (MSF) and the core processor must be scaled to accommodate varying latency levels.
Key Insights: The 10µs Myth and the Idling Penalty
A significant finding of this work is the debunking of the "10µs myth." Many previous resource estimates assumed a total reaction time of 10 microseconds. However, this study shows that communication overhead alone—moving data between the QPU, controller, and decoder cluster—can consume this entire budget, leaving no time for the actual decoding algorithms.
Furthermore, the research highlights a sequential bottleneck. While many decoding tasks are parallelizable, the measurement of "correction qubits" in certain operations (like $\pi/8$ rotation gadgets) is inherently sequential. This creates a fundamental speed limit for the entire system.
The "Idling Penalty" describes the relationship between latency and code distance. For every factor of $\Lambda$ (the error suppression factor) that the reaction time increases, the code distance must typically increase by 2 to preserve logical integrity. This results in a "step-up" effect where physical qubit requirements jump significantly at specific latency thresholds.
Resource Estimates and Architectural Impacts
The study evaluated two representative utility-scale algorithms: a Fermi-Hubbard model for ground-state energy estimation and an $\alpha$-conotoxin NMR spectral prediction algorithm. The hardware parameters were based on a realistic superconducting qubit model ($\Lambda = 9.3$) with a 2.86 MHz stabilization frequency.
For a machine supporting 200–2000 logical qubits, the impacts of latency are profound:
- Physical Qubit Overhead: Even with sub-microsecond decoding, the system requires an additional 100k–250k physical qubits for correction storage in the MSF. The core processor requires 300k–1.75M extra qubits as the code distance increases from $d$ to $d+4$.
- Runtime Penalty: Real-world latencies lead to execution times approximately 100 times longer than an idealized zero-latency scenario.
- Infrastructure Scale: A 10-million-qubit QPU would necessitate a decoder cluster of roughly 15,000 nodes supported by a high-speed network with thousands of channels.
Limitations and Future Directions
This study focuses specifically on the rotated surface code. Other error-correcting codes, such as Low-Density Parity-Check (LDPC) codes, may exhibit different sensitivities to latency and different scaling behaviors. Additionally, the results are optimized for superconducting qubits; systems with slower gate times, such as trapped ions, might be less sensitive to sub-microsecond classical delays. Future research should focus on developing codes with superior rates and distances compared to the surface code, as well as more efficient parallel decoding algorithms to mitigate the substantial runtime slowdowns identified here.
Summary updated