Silicon Photomultipliers: Principles & Applications

• Silicon Photomultipliers are solid-state, photon-number-resolving detectors composed of dense arrays of avalanche photodiode microcells operated in Geiger–Müller mode.
• They achieve high gain and efficient photon detection through microcell architectures with integrated quenching resistors and calibrated readout techniques.
• SiPMs deliver fast, linear responses and are widely applied in quantum optics, medical imaging, and particle physics, replacing traditional photomultiplier tubes in harsh environments.

Silicon photomultipliers (SiPMs) are solid-state, photon-number-resolving detectors consisting of dense arrays of avalanche photodiode (APD) microcells operated in Geiger–Müller mode. Each microcell acts as a binary Geiger-mode photon counter, and the aggregate response of typically hundreds to tens of thousands of cells enables single-photon sensitivity, high gain, and photon-number resolution. SiPMs have emerged as critical components across particle physics, nuclear instrumentation, medical imaging, quantum optics, and astroparticle experiments, steadily supplanting photomultiplier tubes in sectors where compactness, ruggedness, low voltage operation, and magnetic field tolerance are required (Garutti, 2011, Simon, 2018, Cassina et al., 2021). The following sections detail the core aspects of SiPM operation, device physics, acquisition strategies, performance metrics, calibration, and selected scientific applications.

1. Device Physics, Microcell Architecture, and Operating Principle

A SiPM comprises an array of micro-APDs (“microcells,” or “pixels”)—each a p–n junction reverse-biased a few volts above breakdown ($V_\mathrm{bias} > V_\mathrm{bd}$). Each microcell integrates a quenching resistor $R_q$ to arrest the Geiger discharge after a photoinduced avalanche (Section II A, (Cassina et al., 2021, Garutti, 2011)). The key physical components per microcell are:

• A reverse-biased p–n junction (sensitive to photon absorption).
• A polysilicon quenching resistor $R_q$ in series.
• Junction capacitance $C_j$ (setting stored charge and gain).
• Integration into planar or tileable architectures with parallel readout.

When a photon of sufficient energy is absorbed in the depletion region, it generates a photo-carrier, causing avalanche breakdown. The current pulse from the avalanche discharges $C_j$, and $R_q$ quenches the event by lowering the voltage below $V_\mathrm{bd}$, after which the cell resets over a recovery time $\tau_\mathrm{rec} \sim R_q C_j$ (typically tens–hundreds of ns; (Cassina et al., 2021)). Each cell outputs a nearly digital (“one-avalanche-per-photon”) signal, and the total current, summed across all cells, is proportional to the number of simultaneous avalanches ($k$), up to the limit set by the total number of cells ($N_\mathrm{cells}$).

The single-cell gain is

$R_q$0

where $R_q$1 is the elementary charge (Garutti, 2011), yielding typical gains of $R_q$2–$R_q$3.

The overall photon-detection efficiency (PDE) factorizes as

$R_q$4

where $R_q$5 is the breakdown probability at overvoltage $R_q$6, and the fill factor typically ranges from 20–75% (Otte et al., 2016, Biteau et al., 2015).

2. Dynamic Range and Photon-Number-Resolution Mechanisms

The dynamic range of a SiPM is fundamentally bounded by $R_q$7: in response to an incident photon-number distribution $R_q$8, the maximum number of “fired” microcells per pulse is $R_q$9 (Cassina et al., 2021). The conditional probability of $R_q$0 cells firing for $R_q$1 impinging photons with PDE $R_q$2 follows the binomial “dead-cell” model: $R_q$3 At low light levels, the mean number of fired cells is linear in incident flux: $R_q$4 for $R_q$5 ((Cassina et al., 2021), Eq. 2). For increasing photon flux, saturation occurs as all microcells are fired, leading to significant nonlinearity which must be accounted for in high-intensity environments (Garutti, 2011). This nonlinearity can be modeled as: $R_q$6 where $R_q$7 is the number of incident photons (Garutti, 2011).

Photon-number resolution exploits the high gain and the discrete output from each microcell; the output pulse-height spectrum exhibits resolved peaks corresponding to 0, 1, 2, ..., $R_q$8 fired cells, enabling enumeration of detected photons at the single-photon level (Section III, (Cassina et al., 2021)).

3. Signal Readout and Acquisition Strategies

SiPMs produce fast (<1 ns rise, tens–hundreds of ns fall) pulses which can be acquired by several methods (Cassina et al., 2021):

• Analog integration ("boxcar"): The output is amplified and electronically integrated over a fixed gate window ($R_q$9). Short gates (e.g., 10 ns) suppress afterpulsing and delayed crosstalk but are susceptible to timing jitter.
• Digital waveform sampling and offline integration: Waveforms are digitized at high rates (e.g., 5 GS/s DRS4) and integrated offline, allowing flexible gating and post-processing. This method enables shot-by-shot selection of the optimal analysis window, peak height, or area.
• Peak detection: After amplification and (optionally) shaping to slow the signal, the peak value is sampled, maximizing dynamic range and rejecting temporally dispersed noise.
• Time tagging/per-pulse counting: At very low flux, individual avalanches can be time-tagged directly, constructing single-photon time histograms.

Each approach provides different strengths in terms of dynamic range, temporal resolution, readout bandwidth, and rejection of spurious events. In mesoscopic or quantum optics applications, peak detection with fast sampling provides maximal photon-number linearity and robust discrimination against afterpulsing and delayed crosstalk ((Cassina et al., 2021), Figs. 3, 5–8).

4. Performance Metrics: Efficiency, Noise, and Linearity

Key SiPM performance metrics are:

• Photon Detection Efficiency (PDE): Modern SiPMs achieve PDE up to ~40–55% at peak (460 nm for Hamamatsu S13360-1350CS; blue-enhanced FBK NUV-HD reaches 56% at 395 nm; (Otte et al., 2016, Cassina et al., 2021)).
• Dark Count Rate (DCR): Intrinsic dark noise arises from thermally generated carriers. For typical devices (e.g., S13360-1350CS), DCR ≈ 90 kHz at room temperature, reducing steeply with temperature. DCR doubles every 8–10 K (Cassina et al., 2021, Otte et al., 2016).
• Crosstalk (ε): Optical photons generated in avalanches can trigger neighboring cells; prompt crosstalk can be as low as 3% (Hamamatsu LCT5 at 50 μm pitch, (Biteau et al., 2015)), with continued reduction in devices with trench isolation (Otte et al., 2016).
• Afterpulsing: Carrier trapping and delayed release yields additional avalanches, sub-1% in S13360 and similar devices (Cassina et al., 2021).
• Gain: Defined as the inter-peak spacing in the charge spectrum ($C_j$0). Measured via multi-Gaussian fitting or difference/variance methods (Cassina et al., 2021).
• Time resolution: Determined by rise-time, shaping, and electronics. Fast SiPMs achieve sub-100 ps FWHM for single photons (Feindt et al., 2024, Simon, 2018).

SiPMs also exhibit temperature-dependent breakdown voltage ($C_j$1 ≈ 20–60 mV/K, depending on device; the S14160 series achieves ∼30 mV/K, (Cattaneo et al., 2020)), necessitating compensated biasing for gain stability.

5. Calibration, Data Processing, and Nonlinearity Correction

Accurate photon-counting and linearity require systematic calibration and data processing (Cassina et al., 2021):

• Gain calibration (γ): Inter-peak spacing in pulse-height spectra is determined and typically set through multi-Gaussian fitting to dark or weakly-illuminated spectra.
• Assignment of integer photon counts ($C_j$2): Each event is binned by matching its amplitude to the calibrated gains.
• Cell recovery effects: For high rates or overlapping pulses, the single-cell impulse response is deconvolved to account for partial recharge and reduced gain.
• Saturation correction: The nonlinear response at high flux must be modeled with the "dead-cell" or similar binomial models. The inversion techniques include matrix deconvolution, maximum-likelihood, and Bayesian methods ((Cassina et al., 2021), Section III).
• Noise modeling: DCR and crosstalk effects are incorporated into the photon-number statistics, both in histogram fitting and in the calculation of higher-order moments (e.g., Fano factor analysis).

6. Scientific and Quantum Optics Applications

SiPMs are established as practical, low-cost, photon-number-resolving detectors for applications across quantum optics, mesoscopic photonics, and high-energy physics:

• Mesoscopic quantum optics: When calibrated and properly acquired, SiPMs allow shot-by-shot measurement of photon-number statistics for states with mean counts $C_j$3–20, verifying classical (Poisson, Bose–Einstein) and nonclassical photon statistics with sub-shot-noise sensitivity (Cassina et al., 2021).
• Quantum correlation quantification: SiPMs directly enable measurement of the noise-reduction factor,

$C_j$4

with $C_j$5 signifying quantum correlations (twin-beam states measured with $C_j$6 at $C_j$7 in (Cassina et al., 2021); classical beams yield $C_j$8).

• Homodyne-like measurements: Resolving photon numbers up to tens per pulse, SiPMs are well-suited for homodyne detection with mesoscopic local oscillators ($C_j$9 up to 50), and, when operated with fast or peak-detection readout protocols, faithfully capture the quantum statistics demanded by state reconstruction tasks (Cassina et al., 2021).
• Broader experimental physics: SiPMs have displaced PMTs in calorimetry, time-of-flight, tracking, and Cherenkov telescopes, particularly where size, voltage, or environmental robustness is required (Simon, 2018, Bretz et al., 2018, Biteau et al., 2015).

Methodological advances—such as peak-detection with GHz-class sampling and model-based corrections for cross-talk, afterpulsing, and nonlinearity—allow full exploitation of SiPMs' wide dynamic range and enable precision quantum-optics measurements.

---

Collectively, these features position SiPMs as the state-of-the-art solution for photon detection in domains demanding high photon-number resolution, broad dynamic range, and environmental ruggedness, as demonstrated across both foundational quantum-optics experiments and large-scale field deployments (Cassina et al., 2021, Bretz et al., 2018, Simon, 2018).