Fast and accurate qubit-state assignment is essential for feedback, calibration, and error correction in quantum processors. In superconducting platforms, frequency-multiplexed readout makes this task intrinsically multivariate as measured traces can encode crosstalk, qubit-state relaxation events, and other transient nonidealities that are not fully captured by conventional matched filtering. Here, we introduce spiking neural network (SNN) discriminators for superconducting qubit readout. By processing the measurement window in successive time chunks, the networks exploit temporal structure and update classification scores as data arrive, rather than waiting until the end of the readout window. The spiking networks outperform matched-filter discrimination and approach the accuracy of a full-trace artificial neural network. Beyond reaching the performance of artificial neural networks, the key advantage of SNNs is that they provide a streaming, time-resolved estimate of the qubit state that evolves as the readout signal is acquired. Using quantisation-aware training and hls4ml synthesis, we further demonstrate that each FPGA inference update can be completed before the next readout chunk arrives. These results establish spiking neural networks as a promising route to low-latency, real-time qubit readout on FPGA hardware, with broader implications for time-critical quantum-control and scientific-inference applications.
Fast and accurate qubit-state assignment is essential for feedback, calibration, and error correction in quantum processors. In superconducting platforms, frequency-multiplexed readout makes this task intrinsically multivariate as measured traces can encode crosstalk, qubit-state relaxation events, and other transient nonidealities that are not fully captured by conventional matched filtering. Here, we introduce spiking neural network (SNN) discriminators for superconducting qubit readout. By processing the measurement window in successive time chunks, the networks exploit temporal structure and update classification scores as data arrive, rather than waiting until the end of the readout window. The spiking networks outperform matched-filter discrimination and approach the accuracy of a full-trace artificial neural network. Beyond reaching the performance of artificial neural networks, the key advantage of SNNs is that they provide a streaming, time-resolved estimate of the qubit state that evolves as the readout signal is acquired. Using quantisation-aware training and hls4ml synthesis, we further demonstrate that each FPGA inference update can be completed before the next readout chunk arrives. These results establish spiking neural networks as a promising route to low-latency, real-time qubit readout on FPGA hardware, with broader implications for time-critical quantum-control and scientific-inference applications.
Measuring a qubit state is a fundamental but error-prone task in quantum computing. As the number of qubits increases, scalable readout demands multiplexed qubit measurement, which often suffers from crosstalk and other nonidealities. The conventional approach of qubit state detection often struggles with scalability and noise resilience. Neural network (NN)-based discriminators have emerged as a promising alternative, offering superior performance for inferring qubit states from multiplexed signals. However, their deployment on hardware platforms like Field programmable gate arrays (FPGA) is hindered by computation complexity, resource constraints, and latency. In this work, we present an end-to-end methodology for designing and deploying quantized NN-based multi-qubit state discriminators on FPGAs. We demonstrate that implementing a fully connected neural network accelerator for multi-qubit readout is advantageous, balancing computational complexity with low latency requirements without significant loss in accuracy. We significantly reduce the FPGA footprint and latency of the neural network by employing quantization-aware training (QAT) to quantize weights, activations, and inputs, and leveraging automated mapping flows for efficient, hardware-aware design-space exploration and optimization. The hardware accelerator performs frequency-multiplexed readout of five superconducting qubits in 26.7 ns, on a radio frequency system on chip (RFSoC) ZCU111 FPGA. These modules can be implemented and integrated into existing FPGA based quantum control and readout platforms, making them ready for experimental deployment.
Abstract Fast, high-fidelity single-shot readout is a prerequisite for real-time feedback in superconducting-qubit processors, yet the matched filter that remains the field standard discards the temporal structure of the dispersive readout signal and degrades in the presence of energy decay, leakage, and slow parameter drift. We study neural-network discriminators that can run inside the readout feedback loop on a field-programmable gate array (FPGA). Using a physically motivated simulation of dispersive readout that includes cavity ring-up, T1 decay during measurement, leakage to the second excited state, slow IQ drift, and finite signal-to-noise ratio, we benchmark four classical baselines (boxcar integration, matched filtering, linear discriminant analysis, and a Gaussian mixture model) against a floating-point convolutional neural network (CNN), a quantization-aware CNN trained with QKeras, and a new hybrid architecture, the Trace–IQ Fusion Network (TIF-Net), that fuses a learned temporal feature extractor with an integrated-IQ skip path. TIF-Net reachesan assignment fidelity of 0.9962 with only 686 trainable parameters, roughlyan order of magnitude fewer than the 7194-parameter CNN, while exceedingthe matched-filter baseline by 2.3 percentage points. The architecture is constructed to be synthesis-safe: the two pooling operations are realised as fixed-weight convolutions, which we show avoids a scaling pathology of the pooling primitives in our high-level-synthesis (HLS) toolchain. A weight bit-width sweep shows that both networks retain near-peak fidelity down to 4-bit weights, indicating a small multiplier budget for on-board deployment.
Superconducting qubits have emerged as a premier platform for large-scale quantum computation, yet the fidelity of state readout is often hindered by random noise and crosstalk, especially in multi-qubit systems. While neural networks trained on labeled data have shown promise in reducing crosstalk effects during readout, their current capabilities are limited to binary discrimination of joint-qubit states due to architectural constraints. Here we introduce a time-resolved modulated neural network capable of full-state tomography for individual qubits, enabling detailed time-resolved measurements like Rabi oscillations. This scalable approach, with a dedicated module per qubit, mitigated readout error by an order of magnitude under low signal-to-noise ratios and substantially reduced variance in Rabi oscillation measurements. This advancement bolsters quantum state discrimination with neural networks, and propels the development of next-generation quantum processors with enhanced performance and scalability.
Measuring a qubit state is a fundamental yet error-prone operation in quantum computing. These errors can arise from various sources, such as crosstalk, spontaneous state transitions, and excitations caused by the readout pulse. Here, we utilize an integrated approach to deploy neural networks onto field-programmable gate arrays (FPGA). We demonstrate that implementing a fully connected neural network accelerator for multi-qubit readout is advantageous, balancing computational complexity with low latency requirements without significant loss in accuracy. The neural network is implemented by quantizing weights, activation functions, and inputs. The hardware accelerator performs frequency-multiplexed readout of five superconducting qubits in less than 50 ns on a radio frequency system on chip (RFSoC) ZCU111 FPGA, marking the advent of RFSoC-based low-latency multi-qubit readout using neural networks. These modules can be implemented and integrated into existing quantum control and readout platforms, making the RFSoC ZCU111 ready for experimental deployment.
This work presents QubiCML, a novel FPGA-based system engineered for real-time, in-situ discrimination of two- and three-level quantum states (quitrits) in superconducting quantum processors. Accurate and low-latency identification of quantum states is a critical enabler for scaling quantum computing, particularly as higher-dimensional systems such as qutrits are explored to increase information density, circuit depth, and gate expressivity. QubiCML integrates a lightweight, multi-layer neural network directly into FPGA hardware to classify the qutrit states∣g⟩,∣e⟩, and∣f⟩with high fidelity and minimal latency. Traditional state discrimination methods rely on offline post-processing, resulting in delays incompatible with the short coherence times of superconducting qubits. QubiCML overcomes this bottleneck by performing all inference operations within the FPGA fabric itself, achieving full quantum state discrimination within 66 ns (with softmax) or 40 ns (without softmax). The system supports fast reconfiguration via BRAM-stored parameters, enabling compatibility with varying quantum hardware and dynamic calibration environments. Additionally, we propose FPGA-friendly techniques for data normalization and fixed-point quantization to eliminate the overhead of floating-point and division operations—further optimizing the architecture for real-time deployment. We evaluate QubiCML on in-house superconducting hardware across multiple qubits and qutrits. The system achieves up to 97 % average accuracy for qutrits and up to 99 % for two-level systems, using readout windows as short as 650-750 ns. These results establish QubiCML as one of the first practical implementations of real-time qutrit state discrimination, marking an essential step toward mid-circuit measurements, dynamic feedback, and advanced error correction in higher-dimensional quantum computing.
Single-shot dispersive readout of superconducting qubits is a very important part of real-time quantum feedback and quantum error correction (QEC). Conventional discriminators such as thresholding and matched filtering are optimal under Gaussian noise assumptions, yet performance degrades in the presence of non-Gaussian effects in particular when including energy relaxation during readout and leakage to non-computational states. Neural network based discriminators have been proposed to mitigate these limitations. There is still a lack of characterization of the scenarios in which they offer a significant benefit and the mechanisms that underlie that advantage. This paper shows a controlled simulation study of six discrimination methods that get more detailed as the model gets more complex: thresholding, matched filtering, linear discriminant analysis, a multilayer perceptron operating on integrated in-phase and quadrature (IQ) features, one-dimensional convolutional neural network operating on the full time-resolved measurement trace, and a hybrid CNN. By systematically varying physically relevant parameters, we map classifier performance as a function of the ratio of readout duration to qubit relaxation time, which governs the strength of non-Gaussian effects. We demonstrate that neural network-based discriminators offer minimal or no benefits in Gaussian-dominated contexts where linear approaches are almost optimal, yet produce substantial enhancements when non-Gaussian effects are present. Specifically, a hybrid architecture combining a small temporal CNN branch with a direct skip connection for the integrated IQ features achieves assignment fidelity of about 0.98 in the stress-test scenario, which represents a significant improvement over the matched filter. This hybrid architecture, at only 686 trainable parameters, is smaller than the pure CNN yet dominates it in easy settings (where the pure CNN underperforms linear baselines by ≈6 percentage points, pp) while retaining its advantage in hard settings. Additionally, confusion-matrix analysis reveals that this enhancement results from the capacity of temporal models to leverage within-window dynamics that are unavailable to classifiers utilizing integrated observations. Rather than recommending the hybrid as a universally superior method, we characterise the regime in which it helps: it trails the matched filter by only ≈1.2 pp in Gaussian conditions while providing a + 5.39 pp advantage in the stress-test regime, which makes it a regime-robust default wherever the operating point is uncertain or leakage is a meaningful error channel. These results provide a predictive understanding of when neural-network-based readout is beneficial and offer practical guidance for selecting discrimination strategies in superconducting qubit systems subject to real-time constraints.