Time-Resolved Photon-Arrival Readout Improves Nitrogen-Vacancy Center Spin Fidelity at Room Temperature

At room temperature, we enhance per-shot spin-state classification and multi-shot readout of nitrogen-vacancy (NV) centers by exploiting time-resolved photon-arrival statistics. Instead of integrating photoluminescence into a single count, we retain the spin-dependent dynamics arising from intersystem crossing. For 0.5–1.0 µs readout windows, sub-nanosecond time tags are binned and analyzed using lightweight supervised models such as gradient-boosted trees and one-dimensional convolutional networks. Calibrated class probabilities are incorporated into Bayesian updates or sequential probability ratio tests to enable adaptive stopping. The method requires no additional hardware and can be readily extended to other solid-state spin defects with spin-dependent photodynamics. We benchmark its performance against integrated-count and Poisson baselines, and evaluate robustness to drift and charge-state fluctuations. This strategy improves single-shot fidelity, reduces the number of repetitions needed for high-confidence decisions, and maintains low analysis latency, providing a deployable pipeline compatible with standard confocal setups.