Photon Number Resolving Detectors (PNR APDs)

• Photon Number Resolving Detectors (PNR APDs) are devices that convert standard APDs into systems capable of quantifying photon numbers using multiplexing and adaptive protocols.
• They employ architectures such as spatial segmentation, time-domain multiplexing, and adaptive storage-loop protocols to extract detailed photon counting information with high dynamic range.
• Key performance metrics, including detection efficiency, dynamic range, and multiphoton fidelity, are critically evaluated to address challenges like loss, cross-talk, and scaling.

Photon Number Resolving (PNR) Avalanche Photodiodes (APDs) constitute a central technology in quantum optics for quantifying discrete light levels beyond the binary response of conventional “click” detectors. Leveraging spatial or temporal multiplexing, segmentation, or adaptive storage-loop protocols, these devices transform intrinsically non-resolving APDs into effective photon-number-resolving detectors, critical for applications ranging from quantum metrology to scalable single-photon sources.

1. Physical Principles and Detector Architectures

PNR operation in APDs is generally achieved through multiplexed arrangements, where an incoming light pulse is distributed across multiple spatial or temporal APD elements, or via advanced adaptive feedback protocols controlling photon extraction. Key architectures include:

• Spatial Segmentation: Arrays of $M$ APD pixels, each acting as a binary on–off detector, receive photons split via waveguides or micro-optical elements. Simultaneous avalanches in $k$ pixels correspond to $k$ detected photons, providing a histogram of the input number (Jönsson et al., 2018).
• Time-domain Multiplexing: The optical pulse is successively routed and delayed through a tree of fibers or integrated delays, with each time bin probed by the same or multiple APDs. The number of distinct detection events is used to infer the incident photon number (Zhao et al., 8 Jul 2025).
• Adaptive Storage-Loop Protocols: A single APD is coupled to a low-loss optical storage loop with a controllable out-coupler. The loop sequentially extracts small fractions of the pulse, with feedback-adjusted out-coupling based on previous click history. The probabilistic click sequence encodes the photon number. Bayesian updates and information-theoretic feedback laws optimize extraction and inference, enabling sub-shot-noise resolution and dramatically extended dynamic range (Sullivan et al., 2023).
• SiPM/Multipixel Integration: Silicon photon multipliers (SiPMs) comprise densely packed APD microcells. The summed avalanche amplitude scales with the number of simultaneous photons detected, and analog signal processing allows direct photon-number readout, up to saturation of the array (Pomarico et al., 2010).

2. Theoretical Modeling and Key Performance Metrics

Performance metrics for PNR APDs are formalized using conditional probability matrices and figures of merit such as the PNR quality $Q_n$, detection efficiency $\eta$, and multiphoton fidelity. For multiplexed systems and adaptive protocols, key relations include:

• Conditional click probability:

$P_{k,m} = \Pr(S_O = k \mid S_I = m),$

with “desired” output $S_{O,\mathrm{desired}}(m) = m$ for $m \leq n$ and $S_{O,\mathrm{desired}}(m) = n$ for $m>n$ (Jönsson et al., 2018).

• Minimum efficiency for $M$-photon resolution:

$$\eta_{min}(M) = \left[ Q_{th} \frac{M^M}{M!} \right]^{1/M},$$

showing that for $M \gtrsim 5–10$, practical photon-number resolution requires $\eta \rightarrow 1$ (e.g., $\eta_{min}(10) \approx 0.99$) (Jönsson et al., 2018).

• Adaptive storage-loop Bayesian update:

$$\mathbf{P}(\vec{d}_k) = R_k R_{k-1} \cdots R_1 \mathbf{P}_0,$$

where $R_k$ encodes the probabilistic effect of the $k$-th extraction and $\mathbf{P}_0$ is the prior over $N_0$ (Sullivan et al., 2023).

• Estimation error scaling in $N$-plexes:

$$|\langle N_{est}^r \rangle - \langle (a^\dagger a)^r \rangle| = O(1/N),$$

i.e., increasing the number of APD elements $N$ reduces moment-estimation error inversely (Zhao et al., 8 Jul 2025).

Key metrics include dynamic range (maximum distinguishable photon number), mean-square error (MSE) relative to shot noise, detection efficiency, and dark count rates.

3. Multiplexing, Segmentation, and Scaling Laws

Segmentation and multiplexing enable non-resolving APDs to act as PNR detectors. A universal conclusion is the necessity of high efficiency and rapidly scaling segment count:

• Quadratic scaling: The number of required APDs $M$ scales as $M \sim n^2$ to resolve $n$ photons at moderate fidelity (e.g., $Q \geq 0.5$), imposing steep hardware requirements (Jönsson et al., 2018).
• Multiplexed architectures: Temporal or spatial multiplexed arrays distribute $n$ input photons across $M$ channels, reducing pileup and allowing discrete photon counting by summing “clicks.” Incident photon-number statistics are inferred by inverting the click distribution, often involving regularized matrix inversion for cross-talk and losses (Pomarico et al., 2010).
• Detector limitations: Losses in couplers (requiring $\eta_c > 99\%$), dark-count suppression ($p_d < 10^{-4}$), and pixel matching are critical to approach maximum photon resolvability (Jönsson et al., 2018). Even with ideal segmentation, practical limits are encountered for $n \gtrsim 10$.

4. Adaptive Storage-Loop PNR APDs

The adaptive storage-loop architecture (Sullivan et al., 2023) realizes a dramatic enhancement in photon-number resolution, tailored dynamic range, and sub-shot-noise estimation:

• Loop operation: The protocol extracts photon fractions iteratively, using Bayes’ theorem to update the posterior over $N_0$ after each measurement, and adaptively schedules the next fractional extraction $\epsilon_{k+1}$ to maximize information gain:

$$\epsilon_{k+1} = \arg\max_\epsilon \frac{\langle I_{G,k+1} \rangle - I_{G,k}}{\langle I_{A,k+1} \rangle - I_{A,k}},$$

where $I_G$ and $I_A$ denote gained and available information, respectively.

• Practical performance: For $\eta = 0.99$ and APD quantum efficiency $\gamma = 0.9$, Monte-Carlo simulations and analytic derivations demonstrate variance below the shot-noise line up to $N_0 \approx 100$ photons. This dynamic range and precision greatly surpass passive multiplexing or segmentation solutions (Sullivan et al., 2023).
• Implementation: Real-time Bayesian/entropy-driven feedback fits on a small FPGA, with loop delays $\tau \sim 10–100$ ns, yielding MHz-class repetition rates and requiring extremely low-loss tunable couplers.
• Comparison:
  • TES: True PNR to $\sim 20$ photons; $\sim$kHz rate; requires mK cooling.
  • SNSPD arrays: $\sim$4 photons at $\sim$100 MHz.
  • Storage-loop APD: Room temperature; dynamic range $\gg 10\times$ passive; high rate; minimal hardware overhead (Sullivan et al., 2023).

5. Detector Calibration, Cross-talk, and Noise

Calibration of segmented or multiplexed APDs addresses detection inefficiencies and inter-pixel cross-talk:

• Cross-talk modeling: Each avalanche has a branching probability $p_{ct}$ to trigger secondary avalanches. Observed click statistics obey a compound Poisson distribution with Fano factor $F = (1 + p_{ct})/(1 - p_{ct})$. Cross-talk correction matrices $C_{k,n}$ invert measured histograms to reconstruct the true input photon-number distribution (Pomarico et al., 2010).
• Noise sources: Parasitic up-conversion and dark counts set the noise floor, e.g., in up-conversion SiPM systems, a dark count probability per gate $\sim 2.3 \times 10^{-2}$ and cross-talk $p_{ct} \sim 30\%$ are typical, impacting resolution and dynamic range.
• Calibration via inversion: Finite cutoffs and regularized SVD are routinely employed to reconstruct photon statistics up to $k \sim 20$, enabling systematic error removal and accurate photon-number estimation (Pomarico et al., 2010).

6. Benchmarking Frameworks and Application Contexts

Benchmarking PNR APD performance utilizes conditional Fock-state preparation scenarios:

• Benchmark curves: A PNR APD surpasses an $M$-element ideal multiplex if its observed fidelity-success $(P_S, F)$ point lies above the corresponding $M$-detector boundary, meaning that no arrangement of $M$ ideal on–off detectors could have matched or exceeded its performance for state preparation tasks (Provazník et al., 2020).
• Efficiency thresholds: To outperform a $3$-detector multiplex preparing $|3\rangle$, a PNR APD requires $\eta_M \gtrsim 0.54$; for $|5\rangle$, $\eta_M \gtrsim 0.50$ suffices (Provazník et al., 2020).
• Multiplexed single-photon source optimization: Use of PNR detectors in heralded periodic single-photon sources enables higher efficiency and reduced system size. For binary-bulk time multiplexers, a single-photon probability $P_1 = 0.907$ is achieved at $\eta = 0.98$ with only $N=16$ units; threshold schemes require $N=128$ for $P_1 = 0.854$ (Bodog et al., 2020).

7. Limitations, Outlook, and Future Research Directions

Despite substantial progress, PNR APDs face persistent technical challenges:

• Extremely high per-channel efficiency is mandatory for moderate photon-number discrimination, with $\eta_{min}(M)$ approaching unity for $M \gtrsim 10$.
• Quadratic segment scaling ($M \sim n^2$) rapidly challenges miniaturization and integration.
• Loop-based adaptive architectures require ultralow loss and fast switching, with challenges in scaling and real-time computation. Truncation at maximum $N_{max}$ can bias photon-number estimates near this limit, necessitating careful design (Sullivan et al., 2023).
• Up-conversion SiPM approaches at room temperature are limited by modest overall efficiency ($\sim 4\%$) and substantial cross-talk, but provide fast gating and compatibility with telecom wavelengths (Pomarico et al., 2010).

Current research explores the extension of adaptive loops to multi-photon-resolving detectors (e.g., SNSPD/TES integration), optimization with known photon-number priors (Poissonian or thermal), and monolithic integration on photonic chips for scalable, low-loss PNR detection (Sullivan et al., 2023). Multiplexed ON–OFF detector arrays are predicted to enable MHz-rate quantum state engineering with success probabilities and fidelities previously inaccessible using cryogenic TES technology (Zhao et al., 8 Jul 2025).

The continuing evolution of high-efficiency, low-noise, and highly multiplexed APD architectures is expected to play a decisive role in the realization of scalable photonic quantum technologies, quantum-enhanced metrology, and large-scale quantum state preparation.