Detector Selection & Comparison — Comprehensive Guide

1.1The Detection Chain

Every optical measurement terminates at a detector — the device that converts photons into an electrical signal suitable for recording, processing, or display. The detection chain begins with incident photons, proceeds through absorption and carrier generation inside the detector element, and ends with an electrical output (current or voltage) that represents the optical signal. The fidelity of this conversion determines the ultimate sensitivity, speed, and accuracy of any optical instrument, from a simple power meter to a time-resolved spectroscopy system [1, 2].

The quality of the detection process is not set by the detector alone. The complete detection system includes the detector element, any bias or power supply, the transimpedance or voltage amplifier, and the signal processing electronics. Noise enters at every stage, and the choice of detector determines which noise sources dominate and what signal levels are recoverable. Selecting the right detector for a given measurement is therefore one of the most consequential decisions in optical system design [1, 4].

1.2Photon Detectors vs. Thermal Detectors

Optical detectors divide into two fundamental classes based on their physical mechanism of operation [2, 3].

Photon detectors (also called quantum detectors) rely on the direct interaction of individual photons with the detector material. An absorbed photon transfers its energy to an electron, either liberating it from a surface (external photoelectric effect) or promoting it across a semiconductor bandgap (internal photoelectric effect). Because absorption depends on photon energy, photon detectors are inherently wavelength-selective — they respond only to photons above a threshold energy set by the material's work function or bandgap. Photon detectors are fast (nanosecond to picosecond response) and can achieve high quantum efficiency, but their spectral range is limited by material choice [1, 3].

Thermal detectors absorb incident radiation as heat, producing a temperature change that alters a measurable electrical property — resistance (bolometer), voltage (thermocouple/thermopile), or spontaneous polarization (pyroelectric). Because the detection mechanism depends on absorbed energy rather than photon energy, thermal detectors respond to all wavelengths with nominally flat spectral responsivity. The tradeoff is speed: thermal time constants are microseconds to milliseconds, orders of magnitude slower than photon detectors. Thermal detectors also have lower detectivity than cooled photon detectors in most spectral bands [2, 3].

1.3Point Detectors vs. Array Detectors

A second classification axis distinguishes detectors by spatial capability [1, 6].

Point detectors (single-element detectors) produce a single output signal proportional to the total optical power incident on their active area. Photodiodes, photomultiplier tubes (PMTs), avalanche photodiodes (APDs), and pyroelectric detectors are all point detectors. They are used wherever the measurement requires integrating all incident light — power measurement, single-channel spectroscopy, photon counting, and laser beam monitoring.

Array detectors arrange many detector elements in a linear (1D) or area (2D) format, enabling simultaneous capture of spatial or spectral information. Charge-coupled devices (CCDs), complementary metal-oxide-semiconductor (CMOS) sensors, and photodiode arrays (PDAs) are the principal array technologies. Array detectors dominate in imaging and multichannel spectroscopy, where they eliminate the need for mechanical scanning.

This topic covers the full landscape of detector types with emphasis on selection criteria and performance comparison. Subsequent topics in the Measurement & Detection category treat individual detector families — photodiodes, photomultiplier tubes, imaging sensors, and scientific cameras — in full depth.

2.1Photoemissive Detectors

Photoemissive detectors operate on the external photoelectric effect: an incident photon ejects an electron from a photosensitive surface (the photocathode) into vacuum. The liberated photoelectron is then collected or amplified before reaching the anode [5, 10].

The photomultiplier tube (PMT) is the preeminent photoemissive detector. A PMT consists of a photocathode, a series of secondary-emission electrodes called dynodes, and a collection anode, all housed in an evacuated glass or metal envelope. Each dynode multiplies the electron current by a factor δ (typically 3–10), so an n-stage PMT achieves a total gain G = δⁿ — routinely 10⁵ to 10⁷. This enormous internal gain allows PMTs to detect individual photons with timing resolution of 1–2 nanoseconds. PMT photocathodes are available in many compositions (bialkali, multialkali, GaAs, GaAsP), each optimized for a different spectral range within roughly 110 nm to 1700 nm. The principal limitations of PMTs are moderate quantum efficiency (typically 10–40%, though GaAsP cathodes reach 45%), large physical size, high-voltage power supply requirements (1000–2000 V), and sensitivity to magnetic fields [5, 6, 10].

Microchannel plates (MCPs) extend the photoemissive concept with a compact array of miniature electron multiplier channels. MCPs provide spatial resolution along with gain of 10³ to 10⁷ and sub-nanosecond timing, finding application in image intensifiers and time-of-flight detectors [5].

2.2Photovoltaic Detectors (Photodiodes)

Photovoltaic detectors exploit the internal photoelectric effect in a semiconductor p-n or p-i-n junction. An absorbed photon with energy exceeding the bandgap creates an electron-hole pair in or near the depletion region. The built-in electric field separates the carriers, producing a photocurrent proportional to incident optical power [1, 4, 8].

PIN photodiodes are the most common photovoltaic detector. The intrinsic (i) layer between the p and n regions widens the depletion zone, improving absorption efficiency and reducing junction capacitance. PIN photodiodes have no internal gain (G = 1), quantum efficiencies of 60–90%, and bandwidths from DC to tens of gigahertz depending on active area and bias. Available materials span the ultraviolet through mid-infrared: silicon (190–1100 nm), germanium (800–1800 nm), InGaAs (900–1700 nm), and extended InGaAs (900–2600 nm) [1, 4].

Avalanche photodiodes (APDs) add internal gain by operating near the reverse-breakdown voltage, where photogenerated carriers undergo impact ionization and avalanche multiplication. APD gains range from 10 to a few hundred, with the tradeoff of added excess noise and the need for stable, high bias voltage (50–500 V depending on material) [1, 8].

Silicon photomultipliers (SiPMs) consist of an array of microcell APDs operated in Geiger mode (above breakdown), where each cell fires a standardized pulse upon detecting a single photon. The summed output of many microcells produces an analog signal proportional to photon count. SiPMs achieve gains of 10⁵–10⁶ with photon detection efficiencies of 15–50%, compact size, low voltage (25–75 V), and magnetic immunity — positioning them between PMTs and APDs in the detector landscape [6].

2.3Photoconductive Detectors

Photoconductive detectors are semiconductors whose electrical resistance changes when photons generate excess charge carriers. An applied bias drives the photogenerated carriers through the material, producing a measurable current change. Unlike photodiodes, photoconductive detectors do not require a junction; the entire bulk of the material serves as the active region [2, 3].

Photoconductive materials cover spectral ranges inaccessible to silicon and InGaAs: lead sulfide (PbS, 1–3.2 μm), lead selenide (PbSe, 1–5 μm), indium antimonide (InSb, 1–5.5 μm), and mercury cadmium telluride (HgCdTe or MCT, 1–16 μm depending on composition). The extended infrared response comes at the cost of higher noise — many photoconductive detectors exhibit strong 1/f noise — and most require cryogenic cooling to achieve their full detectivity [2, 3].

2.4Thermal Detectors

Thermal detectors convert absorbed radiation into a temperature change and measure the resulting change in an electrical property [2, 3].

Pyroelectric detectors use ferroelectric crystals (LiTaO₃, DLATGS, PZT) whose spontaneous polarization changes with temperature. They respond only to changes in incident power, requiring a chopped or modulated input. Pyroelectric detectors operate at room temperature with flat spectral response from ultraviolet through far-infrared, making them the standard reference detector for FTIR spectrometers. Typical responsivities are 10³–10⁴ V/W with bandwidths of Hz to kHz [3, 9].

Bolometers measure resistance change with temperature. Room-temperature bolometers (microbolometer arrays) are widely used in uncooled thermal imaging. Cooled bolometers (superconducting transition-edge sensors) achieve the highest sensitivity of any detector and are used in sub-millimeter and terahertz astronomy [3].

Thermopiles are series-connected thermocouples that produce a voltage proportional to temperature difference between the active and reference junctions. They provide true DC response (no chopping required) and are the standard sensor in broadband radiometers and laser power meters designed for high average powers [9].

2.5Array and Imaging Detectors

Array detectors integrate many detector elements onto a single substrate for spatially resolved detection [1, 5].

Charge-coupled devices (CCDs) are silicon-based arrays where photogenerated charge accumulates in potential wells during exposure and is then shifted out sequentially for readout. CCDs offer low read noise (2–10 electrons), high quantum efficiency (up to 95% for back-illuminated devices), and excellent charge transfer efficiency. Their spectral range follows silicon: approximately 200–1100 nm. Standard CCDs have read rates of 1–10 megapixels/second; scientific CCDs trade speed for noise performance [1, 5].

Electron-multiplying CCDs (EMCCDs) add an on-chip gain register that amplifies signal charge before the readout amplifier, effectively eliminating read noise. EMCCDs can detect single photons in imaging mode and are used in low-light microscopy and astronomy [5].

CMOS image sensors integrate an amplifier and digitizer at each pixel, enabling parallel readout, higher frame rates, and lower power consumption than CCDs. Scientific CMOS (sCMOS) sensors achieve read noise below 1.5 electrons with frame rates exceeding 100 fps at megapixel resolution, making them competitive with CCDs for all but the most photon-starved applications [5].

3.1Quantum Efficiency

The quantum efficiency η of a photodetector is the probability that an incident photon generates a detectable charge carrier. It is defined as the ratio of collected electron-hole pairs (or photoelectrons, for photoemissive detectors) to incident photons [1, 4]:

Where: I_ph = photocurrent (A), e = elementary charge (1.602 × 10⁻¹&sup9; C), P = incident optical power (W), h = Planck constant (6.626 × 10⁻³⁴ J·s), ν = optical frequency (Hz).

External quantum efficiency (EQE) accounts for all losses including surface reflection, incomplete absorption, and carrier recombination. Internal quantum efficiency (IQE) considers only absorbed photons and is always greater than or equal to EQE. For most practical selection purposes, specifications refer to external quantum efficiency [1].

Quantum efficiency is wavelength-dependent. At short wavelengths, surface absorption and recombination reduce η. At long wavelengths approaching the bandgap cutoff, absorption drops because photon energy is insufficient to excite carriers. Between these limits, well-designed detectors achieve QE of 80–95% (back-illuminated silicon CCDs) or 60–90% (PIN photodiodes) [1, 5].

3.2Responsivity

Responsivity R is the more practical engineering figure of merit — it quantifies the electrical output per unit of incident optical power [1, 4, 8]:

Where: λ = wavelength (nm for the numerical approximation), h = Planck constant, c = speed of light (3 × 10⁸ m/s).

This relationship reveals an important feature: for constant quantum efficiency, responsivity increases linearly with wavelength. Longer-wavelength photons carry less energy, so more photons arrive per watt, generating more photoelectrons. This linear rise continues until the wavelength approaches the material's bandgap cutoff, where absorption collapses and responsivity drops abruptly to zero [1, 4].

The ideal responsivity (η = 1) defines an upper limit at each wavelength:

At 500 nm, the ideal responsivity is 0.403 A/W. At 1550 nm, it is 1.250 A/W. Real detectors fall below this line by a factor equal to their quantum efficiency [4, 8].

3.3Spectral Responsivity of Common Materials

Each detector material has a characteristic responsivity curve shaped by its bandgap and surface properties [1, 3, 4]:

Silicon (E_g = 1.12 eV, λ_c ≈ 1100 nm): Responsivity rises from the UV through the visible, peaks near 900–960 nm at approximately 0.5–0.6 A/W (η ≈ 70–85%), and drops sharply beyond 1000 nm. UV-enhanced silicon extends useful response to 190 nm.

Germanium (E_g = 0.67 eV, λ_c ≈ 1850 nm): Covers the near-infrared with peak responsivity near 1500 nm (0.7–0.9 A/W). Higher dark current than InGaAs at comparable wavelengths limits its sensitivity.

InGaAs (E_g ≈ 0.73 eV for standard composition, λ_c ≈ 1700 nm): Peak responsivity of 0.9–1.1 A/W near 1550 nm with QE of 70–85%. Extended InGaAs compositions push the cutoff to 2.0–2.6 μm at the expense of increased dark current.

HgCdTe (E_g tunable by Hg:Cd ratio): Cutoff wavelength adjustable from approximately 1 μm to beyond 16 μm. Typical QE of 60–80% with anti-reflection coating. Requires cryogenic cooling (77 K for MWIR/LWIR compositions).

3.4Converting Between QE and Responsivity

The relationship between quantum efficiency and responsivity is frequently needed when comparing detector specifications across manufacturers, since some quote QE and others quote responsivity [1, 4]:

This conversion depends on wavelength, which means that a detector with higher responsivity at a longer wavelength does not necessarily have higher quantum efficiency. Comparing detectors on a QE basis removes the wavelength dependence and provides a more fundamental measure of how well the detector converts photons to electrons [1].

4.1The Importance of Noise

A detector's sensitivity is not determined by its responsivity alone — it is the ratio of signal to noise that governs the smallest detectable optical power. Every photodetector generates noise from multiple physical mechanisms, and understanding these sources is essential for predicting performance and selecting the right detector for a given signal level [1, 2, 4].

The total noise from independent sources adds in quadrature (root-sum-of-squares), not linearly. This means that in most operating regimes, one or two noise sources dominate and the others can be neglected. Identifying the dominant noise source for a specific measurement condition is the key to practical detector selection [1, 8].

4.2Shot Noise

Shot noise arises from the discrete nature of electric charge. Photocurrent consists of individual electrons arriving at random intervals governed by Poisson statistics. The root-mean-square (rms) shot noise current is [1, 4, 8]:

Where: e = elementary charge (1.602 × 10⁻¹&sup9; C), I_ph = average photocurrent (A), Δf = measurement bandwidth (Hz).

Shot noise is fundamental — it cannot be eliminated by detector design or cooling. It sets the theoretical limit on signal-to-noise ratio for any photocurrent measurement and increases with the square root of both current and bandwidth [1].

4.3Dark Current Noise

Dark current I_d flows through a photodetector even in the absence of light, generated by thermally excited carriers in the semiconductor or thermionic emission from a photocathode. Dark current contributes its own shot noise [1, 4]:

Dark current is strongly temperature-dependent. For silicon photodiodes, I_d roughly doubles for every 8–10°C increase in temperature. Cooling a detector by 20°C can reduce dark current by an order of magnitude, which directly improves low-light sensitivity. InGaAs, Ge, and HgCdTe detectors have progressively higher dark currents due to their smaller bandgaps, making cooling essential for sensitive measurements at longer wavelengths [1, 3, 4].

4.4Johnson-Nyquist (Thermal) Noise

Johnson noise is generated by the random thermal motion of charge carriers in any resistive element. In the detection circuit, it arises primarily from the load resistor R_L and the detector's shunt resistance R_sh [1, 4, 8]:

Where: k_B = Boltzmann constant (1.381 × 10⁻²³ J/K), T = temperature (K), R = resistance (Ω), Δf = bandwidth (Hz).

Johnson noise is independent of current and depends only on temperature, resistance, and bandwidth. Increasing the load resistance reduces Johnson noise but also reduces bandwidth (through the RC time constant), creating a fundamental speed-sensitivity tradeoff for photodiode receivers [1, 8].

4.51/f (Flicker) Noise

Flicker noise has a power spectral density that increases inversely with frequency, making it the dominant noise source at low frequencies (below roughly 1 kHz for many detectors). The physical origins vary by detector type — trapping/detrapping of carriers at defect sites in semiconductors, conductance fluctuations in resistive films — but the 1/f spectral shape is universal [2, 3, 8].

Photoconductive detectors (PbS, PbSe, HgCdTe in PC mode) exhibit particularly strong 1/f noise, which is why these detectors are almost always used with chopped (AC) illumination and lock-in detection at frequencies above the 1/f corner frequency. Photovoltaic detectors (photodiodes) have much lower 1/f noise because the detection mechanism does not depend on bias current through the bulk material [2, 3].

4.6Excess Noise Factor

Detectors with internal gain (APDs, PMTs, SiPMs) amplify the signal, but the gain process is inherently stochastic and introduces additional noise beyond the shot noise of the multiplied current. The excess noise factor F quantifies this penalty [1, 8]:

Where: M = avalanche gain, k = ionization ratio (ratio of hole to electron ionization coefficients; k ≈ 0.02 for Si, k ≈ 0.4–0.7 for InGaAs).

For PMTs, the excess noise factor is approximately F = 1 + 1/δ, where δ is the secondary emission ratio per dynode. With δ = 5–10, F is typically 1.1–1.2 — remarkably low, which is one reason PMTs excel at low-light detection. For APDs, F increases with gain and is significantly worse (F = 2–10 at typical operating gains), limiting the useful gain before noise degrades SNR [1, 5, 8].

4.7Total Noise and Signal-to-Noise Ratio

The total rms noise current in a detector system combines all independent sources in quadrature [1, 4]:

Where: M = internal gain (M = 1 for photodiodes without gain), F = excess noise factor (F = 1 for detectors without gain).

The signal-to-noise ratio (SNR) for a detector with gain M is [1]:

Where: R = responsivity (A/W), P = incident optical power (W).

In the shot-noise-limited regime (high signal, low amplifier noise), SNR simplifies to √(ηP/(2hνΔf)), depending only on quantum efficiency, signal power, photon energy, and bandwidth — independent of detector gain [1].

5.1Noise Equivalent Power (NEP)

The noise equivalent power (NEP) is the single most useful figure of merit for comparing detector sensitivity. NEP is defined as the incident optical power required to produce a signal equal to the rms noise — equivalently, the optical power that yields an SNR of unity in a 1 Hz bandwidth [1, 2, 9]:

Where: i_n = rms noise current spectral density (A/√Hz), R = responsivity (A/W), v_n = rms noise voltage spectral density (V/√Hz), R_V = voltage responsivity (V/W).

A smaller NEP corresponds to a more sensitive detector. NEP is wavelength-dependent because responsivity varies with wavelength. Manufacturers typically quote the minimum NEP, which occurs at the wavelength of peak responsivity [7, 9].

5.2Minimum Detectable Power

The practical minimum detectable power P_min for a measurement with bandwidth Δf is [7, 9]:

This equation shows the profound impact of bandwidth on detection limits. Reducing the measurement bandwidth by a factor of 100 (e.g., from 10 MHz to 100 kHz) reduces P_min by a factor of 10. This is why lock-in amplifiers, with their extremely narrow effective bandwidths (down to mHz), can detect signals far below the noise floor of wideband measurements [1, 7].

5.3Specific Detectivity (D*)

Comparing NEP values across detectors with different active areas and bandwidths is misleading, because larger detectors collect more noise. The specific detectivity D* (pronounced “D-star”) normalizes NEP by detector area and bandwidth, providing an intrinsic material/design figure of merit [2, 3, 9]:

Where: A_d = detector active area (cm²), Δf = bandwidth (Hz), NEP = noise equivalent power (W/√Hz).

The unit cm·√Hz/W is called the Jones, in honor of R. Clark Jones who introduced the concept. Higher D* means better intrinsic detection capability. D* values range from approximately 10⁸ Jones for uncooled thermal detectors to 10¹³ Jones or higher for cooled HgCdTe and InSb in the infrared [2, 3].

5.4D* vs. Wavelength and the BLIP Limit

The specific detectivity of every photon detector peaks at a wavelength just shorter than its bandgap cutoff and drops at shorter wavelengths where the quantum efficiency decreases. D* also has a fundamental upper limit set by the background radiation that every detector receives. When the noise is dominated by photon fluctuations from the thermal background (at 300 K for room-temperature environments), the detector is said to be at the background-limited infrared photodetector (BLIP) limit [2, 3]:

Where: Φ_b = background photon flux density (photons/s/cm²), η = quantum efficiency.

The BLIP limit decreases with increasing wavelength because longer-wavelength detectors are sensitive to more of the room-temperature background. This is why mid- and long-wave infrared detectors must be cryogenically cooled — cooling reduces both the detector's dark current and the background flux incident upon it [2, 3].

6.1Bandwidth and Rise Time

The temporal response of a photodetector determines how rapidly it can follow changes in incident optical power. Two related parameters characterize speed: the 3-dB bandwidth f_3dB (the frequency at which the output signal amplitude drops to 1/√2 of its low-frequency value) and the 10–90% rise time t_r [1, 4, 8]:

A detector with a 1 GHz bandwidth has a rise time of approximately 350 ps. This relationship assumes a single-pole frequency response, which is a reasonable approximation for most photodiode circuits [1, 8].

6.2RC-Limited Bandwidth

For photodiodes operated in a transimpedance or simple load-resistor circuit, the bandwidth is often limited by the RC time constant of the junction capacitance C_j and the total circuit resistance R [1, 4]:

Where: R_total = parallel combination of load resistance, shunt resistance, and amplifier input impedance; C_j = junction capacitance (F).

Junction capacitance scales with active area and decreases with increasing reverse bias (which widens the depletion region). This creates a fundamental tradeoff: larger active areas collect more light but have higher capacitance and lower bandwidth. High-speed detectors minimize area and operate at reverse bias to reduce C_j [1, 4, 8].

6.3Transit-Time-Limited Bandwidth

Carriers generated in the depletion region must traverse it to reach the contacts. The transit time t_tr sets a second bandwidth limit [1, 4]:

Where: v_s = carrier saturation velocity (≈ 10⁷ cm/s for Si, ≈ 6 × 10⁶ cm/s for InGaAs), w = depletion width (cm).

The overall bandwidth is determined by the slower of the two limits: f_3dB ≈ 1/√(1/f²_RC + 1/f²_tr). Optimized high-speed photodiodes balance depletion width to simultaneously minimize transit time and capacitance [1, 4].

6.4Gain-Bandwidth Product

APDs exhibit a gain-bandwidth product (GBP) that is approximately constant for a given device. Increasing the gain beyond a certain point causes the bandwidth to decrease proportionally [1, 8]:

Typical GBPs are 100–300 GHz for Si APDs and 50–150 GHz for InGaAs APDs. A Si APD with GBP = 200 GHz can operate at M = 100 with 2 GHz bandwidth, or M = 10 with 20 GHz bandwidth. This tradeoff is critical for telecommunications and lidar applications where both sensitivity and speed are required [1, 8].

6.5Temporal Response by Detector Type

Detector speeds span many orders of magnitude. PIN photodiodes achieve 10 ps to 10 ns rise times (bandwidth 35 MHz to 35 GHz), with the fastest performance from small-area, reverse-biased configurations. APDs are somewhat slower at 100 ps to 10 ns due to avalanche buildup time. PMTs achieve 0.5–10 ns rise times, with the fastest mesh-dynode and MCP-PMTs reaching 150 ps for GHz-class photon counting. SiPMs produce 0.3–1 ns rise times but require 10–100 ns recovery per microcell [1, 5, 6, 8].

At the other end of the speed spectrum, photoconductive detectors (PbS, PbSe) are limited to 1–100 μs response times by carrier recombination dynamics. Pyroelectric detectors have thermal time constants of 0.1–10 ms. Thermopiles are the slowest class at 10–100 ms. Array detectors (CCDs, sCMOS) are fundamentally limited by frame rate and readout architecture rather than photoresponse speed [2, 3].

The choice of detector speed must match the temporal characteristics of the measurement. Continuous-wave power measurement requires only DC response. Modulated-signal detection (chopped beams, lock-in amplifiers) requires bandwidth exceeding the modulation frequency. Pulsed laser characterization demands nanosecond or sub-nanosecond response. Time-correlated single-photon counting (TCSPC) pushes timing resolution below 100 ps [1, 6].

7.1Bandgap Energy and Cutoff Wavelength

The spectral range of a photon detector is fundamentally determined by the bandgap energy E_g of its semiconductor material. A photon must have energy hν ≥ E_g to excite an electron across the gap and generate a detectable carrier. This sets a long-wavelength cutoff [1, 3]:

Beyond the cutoff wavelength, the detector becomes transparent and unresponsive. Matching the detector material to the signal wavelength is therefore the first and most constraining step in detector selection [3].

7.2Silicon (Si)

Silicon photodetectors cover the ultraviolet through near-infrared: approximately 190–1100 nm (E_g = 1.12 eV at 300 K). Silicon is the most mature and widely available detector material. Standard silicon photodiodes peak in responsivity near 900–960 nm. UV-enhanced variants use thinner surface layers and anti-reflection coatings to extend useful response down to 190 nm at the cost of reduced NIR sensitivity [1, 4].

Silicon detectors are available as PIN photodiodes, APDs, CCDs, CMOS sensors, and SiPMs. The combination of low dark current, high shunt resistance, mature fabrication technology, and low cost makes silicon the default choice for any application within its spectral range [1, 4].

7.3Germanium (Ge)

Germanium extends coverage to 1800 nm (E_g = 0.67 eV) and was historically the primary near-infrared detector material. Germanium photodiodes have higher dark current than InGaAs at comparable wavelengths — approximately 100× higher at room temperature — and consequently higher NEP. Germanium has largely been superseded by InGaAs for demanding NIR applications but remains in use for broadband power measurement where its wider spectral range (overlapping silicon in the visible) provides convenience [1, 4].

7.4Indium Gallium Arsenide (InGaAs)

Standard InGaAs (In₀.₅₃Ga₀.₄₇As lattice-matched to InP) covers 900–1700 nm (E_g ≈ 0.73 eV) and is the preferred detector for the telecommunications wavelength bands at 1310 nm and 1550 nm. InGaAs photodiodes offer quantum efficiency of 70–85%, low dark current (nanoamperes for mm-scale areas), and bandwidths to tens of GHz — making them the highest-performance room-temperature NIR detectors available [1, 4].

Extended-composition InGaAs pushes the cutoff to 2.0–2.6 μm by increasing the indium fraction, which reduces the bandgap. The penalty is substantially increased dark current (due to the smaller bandgap), requiring thermoelectric cooling to −20°C or below for sensitive operation. Extended InGaAs fills the gap between standard InGaAs and the mid-infrared materials [3, 4].

7.5Lead Salt Detectors (PbS, PbSe)

Lead sulfide (PbS) covers 1–3.2 μm and lead selenide (PbSe) covers 1–5 μm at room temperature, with the spectral range extending to longer wavelengths when cooled. These photoconductive detectors are inexpensive and require only moderate cooling (thermoelectric) but exhibit strong 1/f noise and slow response (microseconds to milliseconds). They are widely used in gas analysis, flame detection, and NDIR spectroscopy where their spectral range and low cost outweigh their noise and speed limitations [2, 3].

7.6Indium Antimonide (InSb)

InSb (E_g = 0.17 eV at 77 K, λ_c ≈ 5.5 μm) is the standard detector for the 3–5 μm midwave infrared (MWIR) atmospheric window. InSb detectors achieve D* values of 10¹¹ Jones when cooled to 77 K but are essentially unusable at room temperature due to enormous dark currents. InSb is available in both single-element and focal plane array formats [2, 3].

7.7Mercury Cadmium Telluride (HgCdTe / MCT)

HgCdTe is unique among detector materials because its bandgap is continuously tunable from approximately 0 to 1.5 eV by adjusting the Hg:Cd ratio (Hg₁₋ₓCdₓTe). This allows a single material system to span the entire 1–16 μm range. Short-wave (SWIR) compositions cover 1–2.5 μm, midwave (MWIR) compositions cover 3–5 μm, and long-wave (LWIR) compositions cover 8–14 μm. HgCdTe detectors achieve D* of 10¹⁰–10¹¹ Jones (LWIR) to 10¹³ Jones (SWIR, 77 K). The main drawbacks are the requirement for cryogenic cooling (typically 77 K for MWIR/LWIR) and high cost driven by difficult epitaxial growth [2, 3].

7.8Thermal Detector Materials

Thermal detectors — pyroelectric crystals (LiTaO₃, DLATGS), bolometer films (VOₓ, amorphous Si), and thermopile junctions (Bi-Sb, Bi-Te) — respond to absorbed heat rather than to individual photons. Their spectral response is essentially flat from ultraviolet through far-infrared (limited only by the window material), making them invaluable as broadband reference detectors and for spectral regions beyond the reach of photon detectors. The cost of wavelength universality is detectivity 3–5 orders of magnitude below cooled photon detectors [2, 3, 9].

8.1Why Gain Matters

In a detector without internal gain, the photocurrent must compete with the noise of the subsequent electronic amplifier. If the signal photocurrent is smaller than the amplifier's equivalent input noise current, the measurement is amplifier-noise-limited rather than detector-noise-limited, and detector performance is wasted. Internal gain amplifies the photocurrent before it reaches the amplifier, reducing the effective impact of amplifier noise by the gain factor [1, 6, 8].

The benefit of gain is not unlimited. Every gain mechanism introduces excess noise — the random fluctuation of the gain process itself. The optimal gain is the value that minimizes the total noise contribution, balancing the reduction of amplifier noise against the increase in excess noise [1, 8].

8.2PMT Gain: Dynode Multiplication

In a photomultiplier tube, a single photoelectron is accelerated toward the first dynode, where it liberates δ secondary electrons upon impact. Each secondary electron is accelerated to the next dynode, multiplied by δ again, and so on through n dynode stages [5, 10]:

Where: δ = secondary emission ratio per dynode (typically 3–10, voltage-dependent), n = number of dynode stages (typically 8–12).

With δ = 5 and n = 10, G = 5¹⁰ ≈ 10⁷. The gain is adjusted by changing the high-voltage supply, which changes δ. PMT gain is extraordinarily high — far exceeding any solid-state detector — and the excess noise is remarkably low (F ≈ 1.1–1.2) because the multiplication process at each dynode is well-regulated by the large inter-dynode electric field [5, 10].

The combination of high gain and low excess noise makes PMTs the dominant choice for single-photon detection in the UV and visible. However, the photocathode quantum efficiency (10–45%) is substantially lower than silicon photodiodes (60–90%), meaning that fewer incident photons produce detectable events [5, 6].

8.3APD Gain: Avalanche Multiplication

Avalanche photodiodes operate in reverse bias near the breakdown voltage, where the electric field in the multiplication region is high enough to cause impact ionization. A primary photocarrier gains sufficient energy between collisions to generate a secondary electron-hole pair, and the process cascades [1, 8]:

Where: M = avalanche multiplication factor (typically 10–200 for Si, 10–40 for InGaAs).

Unlike PMT gain, APD gain carries a significant excess noise penalty quantified by F(M). The McIntyre model (Section 4.6) shows that F increases with M and depends strongly on the ionization ratio k. Silicon APDs (k ≈ 0.02) have much lower excess noise than InGaAs APDs (k ≈ 0.4–0.7) at the same gain, which is why Si APDs can operate usefully at gains of 100+ while InGaAs APDs are typically limited to M < 30 [1, 8].

APDs also exhibit a gain-bandwidth product constraint (Section 6.4): increasing gain reduces bandwidth. The practical advantage of APDs over PIN photodiodes is greatest at moderate signal levels where amplifier noise dominates — the APD's internal gain overcomes the amplifier noise contribution that would limit a PIN receiver [1, 8].

8.4SiPM Gain: Geiger Mode

Silicon photomultipliers operate each microcell above the breakdown voltage (Geiger mode). In this regime, a single photon triggers a self-sustaining avalanche that produces a large, standardized current pulse — typically 10⁵–10⁶ electrons per triggered cell. The avalanche is quenched by an integrated resistor that drops the voltage below breakdown, resetting the cell for the next detection event [6].

The photon detection efficiency (PDE) of a SiPM combines the fill factor (fraction of active area covered by microcells), the quantum efficiency of each microcell, and the avalanche triggering probability. Typical PDE values are 15–50%, peaking in the blue-green (420–550 nm) for standard devices. SiPMs offer several advantages over PMTs: compact size, low operating voltage (25–75 V vs. 1000+ V), magnetic field immunity, and ruggedness. The main disadvantages are optical crosstalk between microcells, afterpulsing, and limited dynamic range at high photon rates (microcell recovery time limits counting linearity) [6].

8.5Signal-to-Noise with Gain: When Gain Helps

Internal gain improves SNR only when amplifier noise is significant. In the shot-noise-limited regime, gain provides no benefit because it amplifies signal and shot noise equally (with the additional penalty of excess noise). The crossover point where gain begins to help is when the amplifier noise exceeds the shot noise of the unamplified photocurrent [1, 8].

For very low light levels (picowatts or below), the amplifier-noise regime dominates for all ungained detectors, and PMTs or SiPMs with their high gain become essential. At moderate light levels (microwatts), APDs offer a modest SNR improvement over PIN photodiodes. At high light levels (milliwatts), PIN photodiodes with no gain are optimal because they avoid excess noise entirely and are not at risk of saturation [1, 6, 8].

9.1Cooling Requirements

The need for cooling depends on the detector material and the required sensitivity [2, 3, 4]:

No cooling required: Si photodiodes, Si APDs, PMTs, SiPMs, pyroelectric detectors, thermopiles.

Thermoelectric (TE) cooling recommended: InGaAs (−20 to −40°C for best NEP), Ge (reduces dark current significantly), PbS, PbSe. TE coolers are compact and require no consumables.

Cryogenic cooling required: InSb (77 K), HgCdTe for MWIR/LWIR (77 K), cooled bolometers. Liquid nitrogen dewars or Stirling-cycle coolers add cost, bulk, and maintenance.

CCD/CMOS sensors: Scientific-grade CCDs are typically TE-cooled to −20 to −80°C; deep-depletion variants may use LN₂ cooling.

Cooling reduces dark current exponentially but adds system complexity. For field-portable instruments, the limitation to uncooled or TE-cooled detectors often drives the material choice as much as spectral range does [3, 4].

9.2Operating Mode: Photovoltaic vs. Photoconductive

Photodiodes can operate in two modes [1, 4]:

Photovoltaic mode (zero bias): The photodiode generates a voltage/current from absorbed photons without external bias. This mode minimizes dark current and 1/f noise, providing the best sensitivity for DC and low-frequency measurements. However, the slower response (higher junction capacitance without bias) limits bandwidth.

Photoconductive mode (reverse bias): Applying reverse bias widens the depletion region, reducing capacitance and increasing speed while maintaining linear response over a wider dynamic range. Dark current increases, so photoconductive mode is preferred for high-speed applications and photovoltaic mode for low-noise, low-frequency applications [1, 4].

9.3Dynamic Range and Saturation

Dynamic range is the ratio of the maximum measurable signal to the minimum detectable signal (NEP × √Δf). Photodiodes offer the widest dynamic range — up to 8–10 decades of linear response for well-designed circuits. PMTs saturate at modest anode currents (typically 1–100 μA) and can be damaged by excess light. APDs must avoid exceeding the maximum rated gain to prevent device damage from runaway avalanche. Array detectors have pixel-level full-well capacity that defines their linear range [1, 4, 5, 8].

9.4Magnetic Immunity and Environmental Robustness

PMTs are sensitive to magnetic fields — even the Earth's field can deflect electron trajectories enough to alter gain. Magnetic shielding (mu-metal) is required for stable operation in variable magnetic environments. All solid-state detectors (photodiodes, APDs, SiPMs, CCDs, CMOS) are immune to magnetic fields, making them preferred for applications near MRI systems, particle accelerators, or other sources of strong fields [5, 6].

Solid-state detectors are also more mechanically robust than PMTs. A PMT's vacuum envelope is fragile; photodiodes survive vibration and shock that would destroy a PMT. For industrial, field, and aerospace applications, solid-state detectors win on ruggedness [5, 6].

9.5Size, Power, and Cost

Size spans four orders of magnitude: a bare photodiode chip occupies < 1 cm³, while a PMT module with high-voltage supply may require 10–100 cm³. SiPMs match photodiodes in compactness while providing PMT-like gain. CCD and CMOS sensors integrate millions of elements on a single chip but require substantial electronics for readout [5, 6].

Power consumption ranges from microwatts (photodiode in photovoltaic mode) to several watts (PMT high-voltage supply, TE-cooled detector module, scientific camera electronics). Cryogenically cooled systems consume 10–100 W for the cooler alone [3].

Cost ranges from under $10 for a basic silicon photodiode to $50–500 for an InGaAs photodiode, $200–2000 for a PMT, $500–5000 for an APD module, and $5,000–50,000+ for a scientific CCD or CMOS camera system. For large-volume applications, the unit cost of silicon devices drops dramatically [4, 5, 6].

10.1A Systematic Approach

Selecting a detector is a constrained optimization problem. The following six-step workflow narrows the candidate pool efficiently by applying the most constraining requirements first [4, 6, 8]:

Step 1 — Wavelength Range → Material. Identify the spectral range of the signal. This immediately limits the candidate materials (Section 7). If the signal falls within the silicon range (190–1100 nm), the widest selection of detector types and the lowest costs are available. NIR signals (900–1700 nm) point to InGaAs. SWIR and beyond require specialized materials with increasing cost and cooling demands.

Step 2 — Signal Level → Gain Requirement. Estimate the optical power at the detector. If the signal is nanowatts or above, a PIN photodiode with a good amplifier is usually sufficient. If the signal is picowatts or below, internal gain (APD, PMT, SiPM) becomes necessary to overcome amplifier noise. If single photons must be detected, only PMTs, SiPMs, and SPADs (single-photon avalanche diodes) qualify.

Step 3 — Speed Requirement → Bandwidth. Determine the temporal characteristics of the measurement. CW or slowly varying signals require only DC to kHz bandwidth. Modulated signals need bandwidth exceeding the modulation frequency. Pulsed signals demand nanosecond response. Photon timing (TCSPC) requires sub-nanosecond precision. This eliminates slow detectors (thermal, PbS/PbSe) from fast applications.

Step 4 — Spatial Requirement → Point vs. Array. If the measurement integrates all incident light, use a single-element (point) detector. If spatial or spectral imaging is required, use an array detector (CCD, CMOS, FPA). Multichannel spectroscopy can use either a scanning monochromator + point detector or a spectrograph + array detector.

Step 5 — SNR Calculation. With a candidate detector identified, calculate the expected SNR using the equations from Sections 4–5. Verify that the detection limit (P_min = NEP × √Δf) is below the expected signal level with adequate margin (typically ≥ 10 dB, preferably ≥ 20 dB).

Step 6 — Practical Constraints → Final Selection. Apply the practical filters from Section 9: cooling requirements, size, power budget, cost, magnetic environment, vibration, and lifetime. If multiple candidates survive, choose the simplest and least expensive option that meets the SNR requirement with margin.

10.2Common Pitfalls

Several common errors lead to poor detector selection:

Specifying gain without considering noise. A high-gain detector is not inherently more sensitive than a low-gain detector. If the measurement is shot-noise-limited, gain adds excess noise without improving SNR.

Ignoring bandwidth. A wideband measurement includes noise from the full bandwidth. Reducing bandwidth (by filtering or lock-in detection) is often more effective than upgrading the detector.

Choosing based on peak QE alone. QE at the signal wavelength, not peak QE, determines performance. A detector with 90% peak QE at 500 nm but only 5% QE at the signal wavelength of 350 nm is inferior to one with 60% peak QE and 40% QE at 350 nm.

Overlooking saturation. A detector optimized for sensitivity may saturate under normal operating conditions. Always verify the maximum power handling of the detector.

Neglecting amplifier noise. The detector is only half the system. A low-NEP detector paired with a noisy amplifier wastes the detector's performance. System-level NEP, including amplifier contributions, is the meaningful metric.

10.3Selection Workflow in Practice