Extracting Photon-Number Information from Superconducting Nanowire Single-Photon Detectors Traces via Mean-Derivative Projection

Photon-number resolved detection with superconducting nanowire single-photon detectors (SNSPDs) attracts increasing interest, but lacks a systematic framework for interpreting and benchmarking this capability. In this work, we combine principal component analysis (PCA) with a new readout technique to explore the photon-number resolving capabilities of SNSPDs and find that the information of the photon number is contained in a single principal component which approximates the time derivative of the average response trace. We introduce a new confidence metric based on the Bhattacharyya coefficient to quantify the photon-number-resolving capabilities of a detector system and show that this metric can be used to compare different systems. Our analysis and interpretation of the principal components imply that photon-number resolution in SNSPDs can be achieved with moderate hardware requirements in terms of both sample rate (5 GSample/sec) and analog bandwidth (3 GHz) and could be implemented in an FPGA, giving a highly scalable solution for real-time photon counting.

💡 Research Summary

This paper addresses the long‑standing challenge of extracting photon‑number information from superconducting nanowire single‑photon detectors (SNSPDs), which are traditionally binary (“click” or “no‑click”) devices. By combining principal component analysis (PCA) with a novel read‑out scheme, the authors demonstrate that the photon‑number content of an SNSPD electrical pulse is essentially captured by a single principal component that closely matches the time derivative of the mean response trace.

The experimental setup uses a commercial fiber‑coupled SNSPD (ID Quantique ID281) operated at 2 K. Pulses from a 1551 nm, 23 ps laser are attenuated to produce mean photon numbers µ ranging from 0.003 to 2.55, logarithmically spaced across 20 settings. For each µ, 100 000 traces are recorded with a 3 GHz‑bandwidth digitizer at 5 GS/s, yielding 512‑sample waveforms (including 20 pre‑trigger samples). In total, 2 million traces constitute the dataset.

Applying PCA to the full 512‑dimensional waveforms yields 512 eigenvectors ordered by explained variance. The first principal component (PC1) accounts for >90 % of the variance for all µ and exhibits a shape virtually identical to the numerical derivative of the ensemble‑averaged trace. Physically, this reflects the observation that multi‑photon events produce steeper rising edges, which can be approximated as a small time‑shift Δt of the single‑photon pulse shape V₁(t). Under a linear approximation, the projection of a measured trace Vₙ(t) onto the mean derivative dV₁/dt satisfies

 P(Vₙ) ≈ C Δtₙ,

where C is a constant. Consequently, the projection value directly yields an estimate of the time shift, which in turn maps monotonically to the photon number. The authors validate this relationship experimentally, showing clean, horizontally separated clusters in the PC1 projection space for different µ.

To quantify the detector’s photon‑number‑resolving capability, the paper introduces a confidence metric based on the Bhattacharyya coefficient. For consecutive photon‑number peaks G|n⟩(t) and G|n+1⟩(t), the metric is defined as

 C|n⟩→|n+1⟩ = 1 − ∫√{G|n⟩(t) G|n+1⟩(t)} dt,

which ranges from 0 (complete overlap) to 1 (perfect separation). The histograms of the projected values are modeled as a sum of exponentially‑modified Gaussian (EMG) functions, each representing a photon‑number peak, with relative weights Pₙ(µ) following a zero‑truncated Poisson distribution. Fitting the µ = 0.003 data (where >99.8 % of events are single‑photon) yields a clean EMG for G|1⟩(t). Subsequent µ values are fitted by adding EMGs for G|2⟩ and a residual for G|3+⟩. Using these fits, the authors obtain confidence values C|1⟩→|2⟩ = 0.81 ± 0.01 and C|2⟩→|3+⟩ = 0.73 ± 0.01, indicating robust photon‑number discrimination up to at least three photons.

A key practical contribution is the demonstration that high‑performance photon‑number resolution does not require ultra‑high sampling rates. By down‑sampling an open‑source dataset (originally 128 GS/s) to 4.92 GS/s and applying a modest 10 MHz–2 GHz band‑pass filter, the authors show that the photon‑number information still collapses onto PC1, preserving horizontal cluster separation. This confirms that a modest 5 GS/s digitizer with 3 GHz analog bandwidth—readily implementable on field‑programmable gate arrays (FPGAs)—is sufficient for real‑time photon‑number extraction.

In the discussion, the authors argue that their mean‑derivative projection provides a physically intuitive and computationally efficient alternative to earlier methods that relied on rise‑time fitting, amplitude thresholds, or time‑to‑digital conversion on multiple edges. The proposed Bhattacharyya‑based confidence metric offers a standardized, physically meaningful figure of merit that can be used to compare disparate detector architectures, bias conditions, or read‑out electronics.

Overall, the paper delivers a comprehensive framework: (1) a clear physical interpretation that the first PCA component equals the mean derivative of the SNSPD pulse, (2) a linear projection method that maps this component to photon number via an effective time shift, (3) a robust statistical confidence metric grounded in the Bhattacharyya coefficient, and (4) evidence that modest hardware (5 GS/s, 3 GHz) suffices for scalable, FPGA‑based, real‑time photon‑number resolution. These advances are poised to impact quantum communication, linear‑optical quantum computing, quantum metrology, and any application where photon‑number‑resolved detection is essential.