Wiki / Detectors / Photomultiplier (PMT)

Photomultiplier (PMT)

Amplifies a faint signal — fluorescence, microscopy, single-point detection — into a readable one, and reports whether it clears the tube’s own dark floor.

Open in the canvas →

A live trace, not a picture of one — but this preview is not interactive. Open it in the canvas to move things, change parameters, and save or export your own version.

In the real world

A photodiode turns one absorbed photon into one electron. That is a vanishingly small amount of charge, and once the signal is weak enough, the amplifier reading it contributes more electrical noise than the light contributes current — the measurement stops being about the light at all. A photomultiplier tube solves this by amplifying the photoelectron before any electronics touch it.

Light lands on a photocathode, a thin film in an evacuated glass envelope, and ejects a photoelectron. A strong electric field accelerates it onto a dynode — an electrode held at a few hundred volts more positive — hard enough that the impact knocks loose several secondary electrons. Those are accelerated onto the next dynode, and so on down a chain of typically 8 to 12 stages before the whole shower is collected at the anode. If each stage yields δ secondary electrons per incident one, the total gain over n stages compounds:

G = \delta^{\,n}, \qquad \delta \propto V_{\text{stage}}^{\,k}

A modest per-stage yield compounds into an enormous total: δ ≈ 4 over 10 dynodes is a gain near 10⁶. Because δ depends on the accelerating voltage, gain is set by the supply voltage — and is steeply sensitive to it, which is why PMT gain is always plotted on a log axis against voltage.

A single photoelectron therefore arrives at the anode as a pulse of ~10⁶ electrons — far above the noise of any reasonable amplifier. This is what makes a PMT able to register individual photons, and it is the entire reason the instrument exists.

Gain is not sensitivity

The most common misconception about PMTs is that turning up the gain makes the instrument more sensitive. It does not. Gain multiplies everything arriving at the first dynode — the signal and the tube's own noise alike — so the ratio between them is fixed before any amplification happens.

That noise has a specific source. The photocathode is warm, so electrons occasionally escape it by thermal energy alone, with no photon involved. Each one is amplified into a full-size output pulse indistinguishable from a real detection. This is dark current, and its rate is what a datasheet quotes as dark counts per second. Cooling the tube reduces it — which is why photon-counting instruments often run their PMTs cooled — but no amount of gain will, because gain amplifies the dark electrons by exactly the same factor.

So the useful figure of merit is the ratio of signal to dark, and the only ways to improve it are to collect more light or to lower the dark rate. Where the gain genuinely matters is in getting the signal clear of the downstream electronics' noise floor — which it does spectacularly well.

What it costs

A PMT is not simply a better photodiode. Its photocathode quantum efficiency — the fraction of arriving photons that eject a photoelectron at all — is typically only 20–40% at its peak, and falls off sharply outside the band the cathode material was chosen for^[1]. A silicon photodiode reaches 80–90% over a much broader range. The PMT wins not by converting more photons, but by amplifying the few it does convert before anything can bury them.

Photocathode material sets the accessible band much as semiconductor bandgap does for a photodiode: bialkali cathodes peak in the blue and are effectively blind past ~650 nm, while extended-red and multialkali types reach into the near infrared^[1]. Beyond roughly 900 nm there is no practical photocathode at all, which is why near-infrared work returns to semiconductor detectors.

Two practical constraints matter on a real bench. Output is linear only up to a maximum anode current; beyond it, space charge in the last dynode stages compresses the response and a brighter input stops reading brighter. And a PMT exposed to room light while powered can be permanently damaged — which is why they live in light-tight housings, are interlocked to the room lights in some labs, and are always powered down before anything is opened.

In OpticalSetup

The PMT reads the same relative ray weight every detector in this palette does, then applies an electron gain to it. Gain is set as a power of ten, from ×1 to ×10⁷, matching how a real tube's gain-versus-voltage curve is specified. This is the element to reach for when a signal is genuinely faint: specimen fluorescence collected through an objective typically arrives carrying somewhere around 10⁻⁴ to 10⁻³ of relative weight, which a plain photodetector reports as a number too small to compare against anything. A gain of 10⁵ lifts exactly that signal into a readable range.

Speed, and the transit-time limit

Pulsed light gives the PMT the same scope trace the photodetector draws, with the same Time interval, Time offset and Sync controls — and the same convolution with its own Response time, which here defaults to 2 ns. That default is not arbitrary: a photomultiplier's speed is limited by the spread in how long electrons take to cross the tube, which for a standard electrode design can put the rise time above 10 ns, while optimized designs reach well below 1 ns^[2]. A slow tube on a fast train does not draw a blurred train — it draws a flat level, because it never resolved the pulses at all.

The dark floor, and why gain cannot beat it

The equivalent dark input is the tube's own dark current expressed as the light level that would produce the same output — referred to the photocathode, so it sits alongside the signal and is amplified by the same gain. The panel reports both the amplified dark floor and the signal / dark ratio, and that ratio is deliberately independent of gain: sweep the gain across every decade it offers and the ratio does not move at all. Lower the dark floor, or collect more light, and it does. That is the single most useful thing this model has to say.

The state line answers the questions in the order they matter. Saturated comes first, because once the output clips at the configured maximum the number is no longer trustworthy at all — a brighter input reads the same as a dimmer one. Otherwise it reports whether the signal clears the dark floor: below dark floor when the tube's own noise is larger than the signal, marginal when it is less than three times larger, and linear range when it is comfortably measurable.

\text{output} = \min(\text{max}, \; \Sigma w \cdot G), \qquad \frac{S}{D} = \frac{\Sigma w}{d}

Amplified output clips at the configured maximum. The signal-to-dark ratio divides the summed ray weight by the equivalent dark input — G cancels, which is exactly the point.

Simplified vs. reality

Gain here is a plain multiplier on relative ray weight, not a dynode cascade: there is no supply voltage, no stage count, no δ, and no gain drift with voltage or temperature. The dark floor is a fixed threshold you set, not a rate — the tracer is deterministic, so nothing fluctuates, there are no dark counts to integrate, and no shot noise on the signal itself. That means the reported ratio is a clean comparison of two configured levels, not a predicted measurement SNR, and it will never reproduce the √N behaviour that governs how long a real experiment must integrate.

Nothing about the photocathode is modelled: no quantum efficiency, no spectral response, and no blindness past the red cutoff — so a PMT here reads 900 nm light exactly as readily as 400 nm, which no real bialkali tube would. Saturation is a hard clip rather than the gradual space-charge compression of a real tube, there is no afterpulsing, no dead time, no dynode fatigue, and no damage from overexposure. Compare readings between configurations, never as an absolute count rate.