Robust and fast post-processing of single-shot spin qubit detection events with a neural network.

Establishing low-error and fast detection methods for qubit readout is crucial for efficient quantum error correction. Here, we test a neural network to classify a collection of single-shot spin detection events, which represent the readout signal of our qubit measurements. This readout signal contains a stochastic peak, for which a Bayesian inference filter including Gaussian noise is theoretically optimal. Hence, we benchmark our neural network, trained with synthetically generated measurement traces, versus this latter algorithm. Comparing both processing methods on our experimental single-shot traces, we find similar performance in terms of the detection error and the post-processing speed. Notably, the neural network turns out to be more robust to fluctuations in the signal offset, length and delay as well as in the signal-to-noise ratio. This higher robustness manifests in a slightly improved visibility of Rabi-oscillations of the spin qubit compared to the Bayesian filter. A training of the network with several million experimental data sets, did not provide a further gain in performance compared to training with synthetically generated data, hence reducing overhead for the application of the neural network. Our contribution thus represents an example of the beneficial role which software and hardware implementation of neural networks may play in scalable spin qubit processor architectures.