Radiation Detectors and Transducers

Photomultiplier tubes remain the detector of choice wherever photon flux is genuinely scarce: single-molecule fluorescence microscopy, flow cytometry, scintillation counting in clinical and nuclear laboratories, and high-sensitivity atomic emission and fluorescence spectrometry (ICP-OES instruments still frequently specify PMTs for trace-element channels). Photon-counting PMTs and their solid-state cousins, single-photon avalanche diodes (SPADs), underlie modern quantum optics experiments and LIDAR ranging systems. Silicon photodiodes and photodiode arrays dominate routine analytical instrumentation: essentially every benchtop UV-Vis spectrophotometer sold today for quality-control and teaching laboratories uses a diode-array detector, because the ability to acquire a full spectrum in milliseconds — critical for HPLC and flow-injection detection, where the sample is present only transiently — outweighs the modest sensitivity penalty relative to a scanning PMT instrument. CCD detectors, an evolution of the same photodiode-array principle with on-chip charge-transfer readout, are standard in Raman spectrometers, portable and benchtop ICP-OES instruments (as CCD or CMOS array detectors replacing photomultiplier-based scanning monochromators), and in astronomical spectroscopy. Thermal detectors persist wherever the wavelength range defeats semiconductor photon detectors: FTIR spectrometers routinely use deuterated triglycine sulfate (DTGS, a pyroelectric detector) for mid-infrared work, or liquid-nitrogen-cooled mercury cadmium telluride (MCT, a photon detector engineered with a very small bandgap) when higher speed and sensitivity are required. Bolometers cooled to cryogenic temperatures remain essential in far-infrared and terahertz astronomy and in some far-IR analytical benches. Beyond the classical categories, engineered hybrid and next-generation detectors continue to blur the lines described here: avalanche photodiodes (APDs) add internal gain to the photovoltaic mechanism (bridging toward PMT-like sensitivity in a solid-state package), and CMOS active-pixel sensors have largely displaced traditional CCDs in consumer and increasingly in scientific array instruments due to lower cost and on-chip signal processing, while still obeying exactly the photovoltaic transduction principles derived here. Understanding detectors this way — as instances of a small number of transduction mechanisms with predictable noise floors — is the throughline connecting this material to full instrument design in UV-Vis Spectrophotometer Instrumentation and to the specialized detector choices used in Detectors for Atomic Spectroscopy.

Ask the tutor: How does the choice between a scanning-PMT and a diode-array design actually change the noise and speed tradeoffs inside a real UV-Vis spectrophotometer?

A radiation detector is a transducer: a device converting one form of energy (radiant power, $P$, in watts) into another (electrical current or voltage) that a data system can digitize. The transduction can proceed by three physically distinct routes.

Photoemissive detectors (phototubes, photomultiplier tubes) rely on the external photoelectric effect: a photon with energy $h\nu$ exceeding the photocathode's work function $\phi$ ejects an electron into vacuum. That single photoelectron is then accelerated through a chain of dynodes held at successively higher potentials, each impact knocking loose several secondary electrons, so that one photon can ultimately deposit $10^6$–$10^8$ electrons on the anode. This internal amplification, occurring before any external circuit noise is added, is what gives PMTs their extraordinary sensitivity — they can count single photons.

Photovoltaic/photoconductive detectors (silicon photodiodes, PDAs, CCDs) rely on the internal photoelectric effect: a photon absorbed within a semiconductor promotes an electron from the valence band to the conduction band, creating an electron-hole pair. In a photodiode, a built-in or reverse-bias electric field across a p-n junction sweeps these carriers apart before they recombine, producing a photocurrent proportional to photon flux. There is no internal multiplication (barring avalanche photodiodes, an engineered exception), so every photoelectron must be measured against the read noise of whatever amplifier follows it.

Thermal detectors (thermocouples, thermopiles, bolometers, pyroelectric detectors) do not care about individual photon energies at all. Radiation is absorbed by a blackened element, raising its temperature by an amount proportional to the total absorbed power, and that temperature change is transduced into a voltage (thermocouple, via the Seebeck effect), a resistance change (bolometer), or a transient polarization current (pyroelectric detector, which requires a chopped or modulated beam to generate any signal at all, since it only responds to $dT/dt$).

The common thread is responsivity: every detector has a characteristic responsivity $\mathcal{R}$ (amps per watt, or volts per watt) linking incident power to output signal. What differs categorically is where noise enters. In a PMT, the photon-to-electron conversion at the cathode is amplified before any resistor or amplifier adds noise, so PMTs approach the fundamental limit set by the randomness of photon arrival itself (shot noise). In a photodiode, the tiny photocurrent must be converted to a voltage by a transimpedance amplifier whose own noise — Johnson noise in the feedback resistor, amplifier input noise — often dominates unless the light level is high. In a thermal detector, the fundamental limit is not photon shot noise but the random thermal fluctuation of the sensing element itself, which sets a floor essentially independent of wavelength.

Notice the crucial architectural inversion in a diode-array instrument: the sample is placed before the monochromator (or the light is dispersed after passing through the sample), and the grating spreads the spectrum across the fixed array rather than being scanned. This is what allows an entire spectrum to be captured in one integration period — milliseconds — rather than the seconds to minutes needed to mechanically scan a grating past a single PMT.

Photomultiplier gain. A PMT with $n$ dynode stages and secondary emission factor $\delta$ per stage produces an overall current gain

where:

• $\delta$ — mean number of secondary electrons ejected per incident electron at one dynode (typically 3–6, and voltage-dependent)
• $n$ — number of dynode stages (typically 9–12)
• $G$ — overall multiplication factor, typically $10^6$–$10^8$

Reading: because $\delta$ depends on the interdynode voltage (roughly $\delta \propto V_d^{0.7}$), the overall gain is extremely sensitive to the total high-voltage supply — a common practical fact: PMT gain (and hence sensitivity) is tuned by adjusting the applied voltage, and a small voltage drift produces a large signal drift.

Anode current and signal. The anode current from a PMT illuminated with radiant power $P$ (W) at wavelength $\lambda$ is

More practically, using cathode responsivity $\mathcal{R}_k$ (A/W, itself proportional to quantum efficiency $\eta$ and $\lambda$):

where:

• $\mathcal{R}_k(\lambda) = \eta(\lambda) e/(h\nu)$ — cathode responsivity, amps of photoelectron current per watt of incident light
• $P$ — incident radiant power (W)
• $G$ — dynode-chain current gain (dimensionless)
• $i_a$ — anode current (A)

Reading: the anode current scales linearly with incident power over many decades, until dynode saturation sets in at high photon flux — the operational definition of the PMT's linear range.

Photodiode responsivity. For a silicon photodiode with quantum efficiency $\eta(\lambda)$,

where:

• $\eta(\lambda)$ — quantum efficiency, fraction of incident photons producing a collected electron-hole pair (typically 0.6–0.9 for Si in the visible)
• $e$ — elementary charge, $1.602\times10^{-19}\ \text{C}$
• $h$ — Planck constant, $6.626\times10^{-34}\ \text{J s}$
• $c$ — speed of light, $2.998\times10^8\ \text{m s}^{-1}$
• $\lambda$ — wavelength (m)

Reading: responsivity rises with $\lambda$ (more photons per watt at longer wavelength) until $\eta$ collapses as photon energy approaches the semiconductor bandgap — this produces the characteristic photodiode responsivity curve that peaks near 900 nm for silicon and then falls sharply past $\sim$1100 nm.

Noise-equivalent power (universal figure of merit).

where:

• $\overline{i_n^2}$ — mean-square noise current in the measurement bandwidth (A²)
• $\mathcal{R}$ — detector responsivity (A/W)
• NEP — radiant power that produces a signal equal to the rms noise (i.e., signal-to-noise ratio of 1)

Reading: smaller NEP is better; NEP lets you compare a PMT, a photodiode, and a thermocouple on one scale regardless of their totally different transduction physics.

Detectivity (normalized, area-independent figure of merit, mainly used for IR thermal/photon detectors):

where:

• $A$ — detector active area (cm²)
• $\Delta f$ — noise bandwidth (Hz)
• $D^{*}$ — specific detectivity, units cm·Hz$^{1/2}$/W

Reading: $D^*$ lets you compare detectors of different sizes fairly; larger $D^*$ means a more sensitive detector per unit area and bandwidth.

Example 1 (foundational): PMT anode current and required gain. A photomultiplier tube has a cathode quantum efficiency $\eta = 0.22$ at $\lambda = 450$ nm and is illuminated with a very weak beam of radiant power $P = 2.5\times10^{-13}$ W (a typical low-light spectrofluorometric signal). The dynode chain has $n = 9$ stages with secondary emission factor $\delta = 4.8$ per stage. Step 1 — cathode responsivity.

Step 2 — photoelectron current at the cathode (before multiplication).

This corresponds to a photoelectron rate of $i_k/e = 1.25\times10^{5}$ electrons/s — a real but minuscule current, far below what any op-amp could measure directly. Step 3 — dynode gain.

Computing stepwise: $4.8^2 = 23.04$, $4.8^4 = 23.04^2 = 530.8$, $4.8^8 = 530.8^2 = 2.817\times10^5$, so

Step 4 — anode current.

A 27 nA current is easily measured by a conventional current-to-voltage amplifier — this is exactly why the internal gain matters: it converts an unmeasurable femtoampere signal into a comfortably instrument-grade nanoampere signal, entirely inside the vacuum tube, before external circuit noise is added.

Limiting-case check: if the applied high voltage were increased so that $\delta$ rose from 4.8 to 5.5 (a modest ~15% change, physically reasonable for a few-hundred-volt increase), gain becomes $5.5^9$. Since $5.5^2=30.25$, $5.5^4=915.06$, $5.5^8=837,\!240$, $5.5^9 = 4.60\times10^6$ — a 3.4-fold increase in gain from a 15% voltage-driven change in $\delta$, confirming the strong nonlinear voltage sensitivity predicted in the derivation.

Example 2 (applied): Choosing between a photodiode and a PMT for a fixed-wavelength absorbance measurement. A UV-Vis instrument monitors absorbance at 254 nm using a deuterium lamp. The beam reaching the detector after passing through a moderately absorbing sample carries $P = 4.0\times10^{-9}$ W. Two detector options are available:

• Silicon photodiode: $\eta(254\ \text{nm}) = 0.35$ (Si response is weak in the deep UV due to strong surface absorption), no internal gain, feedback resistor $R_f = 10^{9}\ \Omega$ in a transimpedance amplifier with equivalent input noise current $i_{n,\text{amp}} = 2.0\times10^{-15}\ \text{A}/\sqrt{\text{Hz}}$, plus Johnson noise from $R_f$.
• PMT: cathode $\eta(254\ \text{nm}) = 0.20$ (bialkali photocathode), gain $G = 1.0\times10^{6}$, anode dark current equivalent noise negligible at this signal level (well above single-photon regime).

Step 1 — photodiode signal. Responsivity:

Numerator: $(0.35)(1.602\times10^{-19})(254\times10^{-9}) = 1.424\times10^{-26}$. Denominator: $1.986\times10^{-25}$.

Step 2 — photodiode noise. Johnson noise current in $R_f$ (bandwidth $\Delta f = 1$ Hz, $T = 298$ K):

Total noise (amplifier and Johnson, added in quadrature since independent):

Step 3 — photodiode signal-to-noise ratio.

Step 4 — PMT signal and shot-noise-limited S/N. Cathode responsivity:

At the cathode, before gain, the relevant noise is photon shot noise on the photoelectron current:

(Because both signal and shot noise are multiplied by the same gain $G$ before reaching the anode, $G$ cancels in the ratio, so the S/N calculated at the cathode equals the S/N measured at the anode.)

Interpretation. At this comparatively bright signal level ($P = 4$ nW, well above single-photon-counting territory), the photodiode's higher deep-UV quantum efficiency advantage is outweighed by the fact that shot noise scales as $\sqrt{i_k}$ while Johnson noise depends only on $R_f$ and $T$ — but the numbers above actually show the photodiode outperforming the PMT here (S/N $6.3\times10^4$ vs. $2.3\times10^4$), because at nanowatt-level UV signals with a very large feedback resistor, Johnson noise is small enough that the photodiode's simplicity and adequate quantum efficiency win. This is exactly why modern diode-array UV-Vis instruments are entirely practical for routine absorbance work despite lacking PMT-style gain — but it also shows why for genuinely faint signals (fluorescence, Raman, trace atomic emission), where $P$ drops toward the femtowatt range, the PMT's shot-noise-limited advantage (no added amplifier or Johnson noise before amplification) reasserts itself, as Example 1 demonstrated.

Example 3 (exam/competition level): Detectivity comparison and NEP-limited detection limit for a thermal detector in the infrared. A thermopile infrared detector has an active area $A = 4.0\ \text{mm}^2 = 0.040\ \text{cm}^2$, is operated with a noise bandwidth $\Delta f = 12.5$ Hz (typical for a chopped, lock-in-detected IR bench at 25 Hz chopping frequency with a lock-in time constant giving this equivalent noise bandwidth), and has a manufacturer-specified detectivity $D^{*} = 5.0\times10^{8}\ \text{cm}\,\text{Hz}^{1/2}/\text{W}$ at 10 μm. A student wants to know: what is the smallest modulated IR power this detector can distinguish from noise (S/N = 1), and how does that compare to the power actually delivered by a sample cell with 0.15% transmittance change on a 2.0 mW incident IR beam? Step 1 — NEP from $D^*$. Rearranging the definition of detectivity:

This is the radiant power that would give S/N = 1 in this detector under these exact bandwidth and area conditions. Step 2 — power actually delivered by the sample's absorbance change. A transmittance change of 0.15% on a 2.0 mW beam corresponds to a power change

Step 3 — resulting signal-to-noise ratio.

Step 4 — interpretation and limiting check. A signal-to-noise ratio of about 2100 means this transmittance change is very comfortably measurable — the noise floor (1.4 nW) is more than three orders of magnitude below the signal (3.0 μW). Turning the question around: the smallest transmittance change resolvable at S/N = 3 (a common detection-limit convention) is

or about 2 parts per million transmittance change — illustrating why thermopile/bolometer-based IR benches, though far noisier per photon than a PMT, remain entirely adequate for routine mid-IR spectrometry because IR sources deliver comparatively large powers per resolution element and absorbance changes of interest (often several percent) are enormously larger than this floor.

Limiting case built into the check: as $\Delta f \to 0$ (infinitely long integration, a single very slow scan), $\text{NEP} \to 0$ and detection sensitivity improves without bound — consistent with the general principle that averaging longer always reduces noise-limited detection limits, up to the point where $1/f$ drift noise (not modeled here) takes over.

The photomultiplier tube's practical development is inseparable from the RCA engineering laboratories of the 1930s and 40s, but the conceptual foundation — the photoelectric effect itself — traces to Albert Einstein, whose 1905 explanation of the photoelectric effect (for which he received the 1921 Nobel Prize in Physics) established that light ejects electrons from a metal surface only above a threshold frequency, with excess energy appearing as electron kinetic energy: $E_k = h\nu - \phi$. This single equation is the physical basis of every photocathode in every PMT built since. The engineering path from "the photoelectric effect exists" to "a usable single-photon-counting detector" required decades of vacuum-tube and secondary-emission physics, much of it driven by wartime radar and later by nuclear and particle physics, where scintillation counting (pairing a scintillator crystal with a PMT) became indispensable. Analytical chemists inherited a mature, off-the-shelf technology by the 1950s and 60s, when commercial spectrophotometers (Beckman, Cary) began routinely incorporating PMTs, displacing the barrier-layer (selenium) photocells used in earlier colorimeters.