Physics-Informed Neural Network for Fast single NV^- center characterization

Rapid characterization of many single NV centers is important for perfecting growth and processing conditions and identifying high-performance NV centers for sensing.  Unfortunately, it is bottlenecked by long measurement times and model-dependent fits. Ramsey interferometry measurements of coherence lifetime and the hyperfine coefficients form a good benchmark for quantum sensing; however, these quantities are difficult to extract with short acquisition times. We introduce a physics-informed neural network that jointly denoises and estimates parameters directly from few-shot Ramsey traces. The model is pre-trained on semi-quantum–mechanical simulations with realistic noise modeling, then fine-tuned on experimental data using an uncertainty-aware objective. Given a Ramsey signal with a short acquisition time compared to conventional approaches, the network returns a denoised signal, a calibrated estimate of the detuning associated with nearby hyperfine interactions, and confidence intervals. Across multiple NV centers, predictions agree with references, with integration times up to 50 times higher within experimental uncertainty. The approach enables rapid screening of candidate NV centers for sensing, robust operation under mild miscalibration, and seamless integration into adaptive measurement loops.