Photon Number-Resolved Avalanche Detectors

• Number-resolved avalanche detectors are specialized photodetectors that leverage avalanche multiplication to quantify discrete photon numbers with high sensitivity.
• They operate by capturing analog avalanche amplitudes or statistical multiplexed responses, enabling precise mapping of photon-induced events.
• Current architectures, including fast-gated APDs and multiplexed SPAD arrays, provide practical tools for quantum optics diagnostics and photonic state characterization.

A number-resolved avalanche detector is a photodetector that leverages avalanche multiplication processes to achieve sensitivity not only to the arrival of one or more photons, but to the discrete number of photons absorbed or photo-generated within a well-defined temporal or spatial interval. This distinguishes such devices from conventional single-photon avalanche diodes (SPADs) or Geiger-mode avalanche photodiodes (APDs), which operate fundamentally as binary click detectors. Number-resolved avalanche detectors preserve or extract coarse-grained or even quantized information about absorbed photon number—either via analog avalanche amplitude, statistical multiplexing, or advanced discrimination of avalanche development dynamics. This capability enables new diagnostics for optical quantum sources and in principle provides resources for photonic quantum information processing.

1. Physical Principles and Detector Architectures

The essential operating principle of a number-resolved avalanche detector is that the output observable (typically the avalanche amplitude, integrated charge, or number of independent clicks in a multiplexed system) retains a well-defined, monotonic dependence on the number of incident photons, up to a certain nonlinearity and noise-induced limit.

Two broad classes are identifiable:

• Analog-amplitude number-resolved avalanche detectors: These exploit the proportionality between the number of primary carriers and the early-stage avalanche current or voltage before full saturation. An explicit example is the use of self-differenced, fast-gated APDs, where the amplitude distribution of unsaturated avalanche events separates into peaks reflecting different photon-number classes (Thomas et al., 2010, Dynes et al., 2011, Yuan et al., 2010).
• Multiplexed or pseudo-number-resolved architectures: These combine many binary avalanche elements (e.g., as in multi-pixel photon counters or time-multiplexed detectors) such that the number of pixels or bins fired correlates with input photon number. Segmented SPADs coupled along a waveguide with engineered splitting ratios are a canonically studied example (Nehra et al., 2017, Eraerds et al., 2010, Kröger et al., 2017).

Both categories can be implemented in semiconductor (Si, InGaAs) APDs or in superconducting nanowire architectures (SNAPs), although the underlying amplification and timing statistics differ significantly (Cheng et al., 2016, Zhao et al., 2014).

2. Avalanche Multiplication Statistics and Limitations

The statistical properties of the avalanche multiplication process are determinative for number resolution. The gain in an APD or SPAD initiated by $M$ primary carriers is stochastic, with variance governed by the excess noise factor $F = \langle M^2 \rangle / \langle M \rangle^2$ or, more generally, by a branching–process noise parameter $A$. Early results (e.g., (Yuan et al., 2010)) have shown that in the early temporal regime and under strong excess bias, the multiplication noise can be so low ($\mathfrak{F} \approx 1.07$) that not only 1-photon but 2- and 3-photon locally seeded avalanches are well-resolved. In the more general stochastic theory (Windischhofer et al., 2020), the relative variance scales as $1/A$; for $M$ initial $e$–$h$ pairs in pure single-species-multiplying material, $A = M$, so separation $\propto 1/\sqrt{M}$.

However, as the avalanche evolves toward global saturation, amplitude information is rapidly lost. For Geiger-mode SPADs (including most SiPMs and quenching avalanche cells), the avalanche diverges and is subsequently quenched by the circuit, producing an output pulse essentially binary with respect to initial photon number. True amplitude-based number resolution requires access to the linear, unsaturated, and low-noise gain regime—a nontrivial constraint for ordinary APDs.

Furthermore, in segmented or multiplexed detectors the mapping is statistical and saturating: as the number of detected photons approaches or exceeds the number of available modes or pixels, the output distribution compresses and one-to-one photon-number resolution is impossible (Kröger et al., 2017, Nehra et al., 2017).

3. Device Realizations and Quantitative Performance

Several key architectures embody number-resolved avalanche detection:

• Self-differencing, fast-gated APDs: Biasing InGaAs or Si APDs with ultrafast ($F = \langle M^2 \rangle / \langle M \rangle^2$0GHz) gates and extracting the avalanche amplitude during the first hundreds of picoseconds enables up to four discrete photon-number peaks to be resolved in amplitude histograms, with reported detection efficiencies as high as 73.8% at 600 nm and dark-count probabilities below $F = \langle M^2 \rangle / \langle M \rangle^2$1 per gate (Thomas et al., 2010). Avalanche probability per absorbed photon can exceed 90%. Experimentally, careful amplitude discrimination (with thresholds or window discriminators) yields photon-number assignment errors under 10% for $F = \langle M^2 \rangle / \langle M \rangle^2$2 (Thomas et al., 2010).
• Time-multiplexed detection (fiber loop networks): Near-infrared Geiger-mode InGaAs APDs become effective pseudo-PNR detectors when a pulse is temporally split into $F = \langle M^2 \rangle / \langle M \rangle^2$3 bins, each interrogated by a binary APD. The distribution of click counts $F = \langle M^2 \rangle / \langle M \rangle^2$4 provides a calibrated mapping from input photon number $F = \langle M^2 \rangle / \langle M \rangle^2$5 to observed outcomes, achieving single-shot energy resolution at the few-attojoule level over dynamic ranges up to $F = \langle M^2 \rangle / \langle M \rangle^2$6 bins and MHz repetition rates (Eraerds et al., 2010).
• Multipixel and segmented SPAD arrays: Large arrays of binary detectors (e.g., $F = \langle M^2 \rangle / \langle M \rangle^2$7 APDs in parallel in an MPPC arrangement) can utilize the click statistics—including in regimes well above the few-photon-per-pixel limit—to infer photon-number-related information directly from click histograms. The binomial $F = \langle M^2 \rangle / \langle M \rangle^2$8 parameter ($F = \langle M^2 \rangle / \langle M \rangle^2$9) is used as a robust nonclassicality and state-characterization measure, avoiding artefacts induced by naive photon-counting statistics (Kröger et al., 2017).
• Integrated side-coupled waveguide SPADs: Distributing photons over $A$0 side-coupled SPADs with optimized graded splitting can approximate genuine photon-number resolution, as measured by POVM purity; approximately $A$1 segments are needed for POVM purity $A$290% for $A$3 photons, assuming low cross-talk and dark count probability ($A$4) (Nehra et al., 2017). The architecture minimizes losses due to nonunit per-segment quantum efficiency through photon recycling.

4. Measurement Methodologies and Data Analysis

Number-resolved avalanche detectors can operate either as single-shot discriminators (using amplitude or bin count for event-by-event assignment) or as statistical analyzers (extracting source photon-number distributions through repeated measurements and detector response modeling).

For amplitude-based discrimination, the measured histogram of avalanche voltages is modeled as a sum of Gaussians, with peak positions and widths parameterized according to detected photon number. The performance is then quantified by discrimination errors—computed from overlap integrals between adjacent peaks (Thomas et al., 2010, Dynes et al., 2011).

For multiplexed architectures and statistical tomography, the observable is the distribution $A$5 of click counts $A$6 for known input mean photon number $A$7. Bayesian inversion yields $A$8, allowing single-shot energy estimation with calibrated uncertainty (Eraerds et al., 2010). In complex architectures, full quantum click statistics—encoded by binomial laws, convolution models for crosstalk, and generalized POVMs—are used directly for source characterization (Kröger et al., 2017, Nehra et al., 2017).

5. Applications in Quantum Optics and Photon-State Characterization

Number-resolved avalanche detectors enable previously inaccessible metrology and state-discrimination tasks in photonic quantum information science:

• Higher-order correlation measurement: Self-differenced PNR-APDs in Hanbury Brown–Twiss geometry allow direct probing of higher-order field correlations, not just $A$9, by conditioning coincidence logic on avalanche amplitude thresholds. The realization of the generalized observable

$\mathfrak{F} \approx 1.07$0

provides sensitivity to photon bunching far in excess of standard $\mathfrak{F} \approx 1.07$1 techniques (Dynes et al., 2011).

• Nonclassicality and photon statistics: Click-based approaches, such as analysis with the $\mathfrak{F} \approx 1.07$2 parameter and high-fidelity reconstruction of source photon-number moments without full inversion, permit quantum-state discrimination over intensity ranges inaccessible to strict PNR detectors (Kröger et al., 2017).
• High-efficiency quantum photonic readout: For quantum information applications (Bell state measurements, quantum communication), room-temperature, high-speed, noncryogenic number-resolved APDs with discrimination up to a few photons per shot offer practical advantages over TES or VLPC detectors (Thomas et al., 2010).
• Resource-efficient quantum tomography: Statistical approaches using a single SPAD with known dead time, calibrated efficiency, and maximum-likelihood inversion enable number-state tomography on pulsed sources up to rates near one photon per dead time—providing low-cost, software-based PNR capabilities (Banner et al., 2023).

6. Fundamental and Practical Limitations

The principal constraints on the performance and utility of number-resolved avalanche detectors are set by:

• Avalanche multiplication noise: Even with minimized stochasticity, single-carrier-initiated avalanches are broad; separation of adjacent photon-number peaks becomes steadily more difficult as $\mathfrak{F} \approx 1.07$3 increases (Windischhofer et al., 2020, Yuan et al., 2010).
• Avalanche broadening and self-quenching: At high photon number, peaks overlap and amplitude scaling with photon number becomes sublinear due to series resistance and self-limiting effects (Thomas et al., 2010, Dynes et al., 2011).
• Saturation and finite multiplexing: For multiplexed systems, the number of modes or pixels $\mathfrak{F} \approx 1.07$4 fundamentally limits dynamic range; to sustain high POVM purity at $\mathfrak{F} \approx 1.07$5 photons requires $\mathfrak{F} \approx 1.07$6 (Nehra et al., 2017).
• Detector dead time, crosstalk, afterpulsing, and dark counts: These limit count rates, distort click statistics, and reduce purity. Careful characterization and modeling are required to maintain accuracy (Kröger et al., 2017, Eraerds et al., 2010, Nehra et al., 2017).
• Quenching and regime of operation: For Geiger-mode, above-breakdown operation, the avalanche saturates and all amplitude information is lost; only in the pre-saturation linear gain regime can intrinsic number resolution be realized within a single microcell (Windischhofer et al., 2020).

7. Prospects and Directions

Advances in material engineering, fast gating electronics, integrated photonics, and statistical signal processing continue to extend the accessible range and efficiency of number-resolved avalanche detectors. Future improvements focus on:

• Further reduction of amplification noise to approach the theoretical $\mathfrak{F} \approx 1.07$7 limit for higher primary charge numbers and larger spatial modes.
• Scalable integration of thousands of SPAD segments with low dark count rates and negligible crosstalk for high-purity PNR detection in photonic circuits (Nehra et al., 2017).
• Architectures which permit high-speed operation ($\mathfrak{F} \approx 1.07$8GHz) while retaining single-shot number discrimination over the largest possible photon-number range (Cheng et al., 2016).
• Statistical algorithms for number-state tomography compatible with time-varying sources and robust to moderate levels of dead time and afterpulsing (Banner et al., 2023).

Number-resolved avalanche detectors, including both analog and statistical-multiplexed forms, provide a practical and evolving technology for high-speed, high-fidelity photon-number sensing with clear relevance to both fundamental quantum optics and applied quantum engineering.