Spiking neural networks for streaming qubit readout
This study addresses the limitations of conventional matched filtering in superconducting qubit readout, which struggles with crosstalk and transient noise while lacking real-time streaming estimation capabilities. To overcome these challenges, this work proposes a chunked streaming quantum state classification method based on spiking neural networks (SNNs). By leveraging quantization-aware training (QAT) and HLS4ML synthesis techniques, the model is efficiently deployed on an FPGA platform to enable low-latency hardware inference. Experimental results demonstrate that a single inference completes before the arrival of the subsequent data chunk, providing dynamically evolving state estimates with high accuracy that approach those of full-trajectory artificial neural networks (ANNs). This work establishes the potential of SNNs for real-time quantum control and scientific inference applications.