How this instrument works
A silicon photomultiplier is not one detector but thousands of them: an array of tiny avalanche photodiodes called microcells, each biased above its breakdown voltage and wired to fire independently the instant a photon triggers it. Photon detection efficiency describes the odds that a single incoming photon actually produces one of those firings, and that odds is not one number but a chain of three: PDE = QE × FF × Pt. Each factor is a probability for a different physical step, and because a photon must clear all three in sequence, the probabilities multiply rather than average.
Quantum efficiency, QE, is the chance a photon that reaches the silicon actually gets absorbed and produces an electron-hole pair; it depends on wavelength and the depth of the depletion region, which is why a datasheet plots QE as a curve rather than a single figure. Fill factor, FF, is pure geometry — the photosensitive fraction of each microcell's footprint, after subtracting the quenching resistor, the readout trace, and the trench that isolates one cell optically and electrically from its neighbours. Avalanche trigger probability, Pt, is the chance a carrier born inside the high-field region starts a self-sustaining avalanche instead of drifting away or recombining, and it climbs with overvoltage — the bias held above the breakdown point.
Treat PDE as an average of the three percentages and the number looks far better than it is: (80 + 75 + 60) ⁄ 3 comes to 71.7%, nearly double the true figure. The multiplication is not a modelling choice; it reflects that absorption, geometric acceptance, and avalanche initiation are three separate gates a photon must pass in order, so the real yield is their product. The edge case worth remembering is fill factor's ceiling: shrinking microcells for higher dynamic range keeps the trench width roughly fixed, so the dead fraction of each cell grows even as the sensor gets smaller on paper.
- Enter the sensor's Quantum efficiency, % at your working wavelength — read it off the datasheet's QE(λ) curve at the peak or the wavelength you actually plan to detect.
- Enter the microcell's Fill factor, % from the datasheet's cell-geometry specification; smaller-pitch devices built for wide dynamic range usually quote a lower figure.
- Enter the Avalanche trigger probability, % for your chosen bias point — datasheets plot Pt, or PDE directly, against overvoltage, the volts held above breakdown.
- Read Photon detection efficiency, % — the product of all three, the figure that predicts how many incident photons the sensor will actually register as pulses.
Worked example — an 80/75/60 SiPM at rated bias
Take a blue-sensitive SiPM run at its datasheet operating point: Quantum efficiency, % = 80, Fill factor, % = 75, and Avalanche trigger probability, % = 60. PDE = 80 × 75 × 60 ⁄ 10,000 = 36.0%. Out of every 100 photons landing on the sensor's active window, only 36 actually produce a countable pulse — the other 64 are lost at one of the three gates along the way.
The multiplication matters for anyone speccing a photon-counting system, from a PET-scanner designer estimating coincidence timing resolution to a LIDAR engineer budgeting return-signal margin. Averaging the same three figures instead of multiplying them would report 71.7%, a number nearly double the sensor's real yield — enough to make a design that looks comfortably specced actually fall short of its required signal-to-noise ratio once it is built.
Questions
Why does PDE multiply the three factors instead of averaging them?
Because QE, FF, and Pt each describe a separate gate a photon must clear in sequence — being absorbed, landing in the active area, and triggering an avalanche — and probabilities for sequential independent events multiply. Averaging 80%, 75%, and 60% gives 71.7%, nearly double the true 36% yield, which is why datasheet-average shortcuts systematically overstate a sensor's real performance.
Is quantum efficiency the same thing as photon detection efficiency?
No. QE is only the first of three multiplied factors — the probability an absorbed photon creates a carrier pair. A sensor can advertise 90% QE and still deliver a much lower PDE once fill factor and avalanche trigger probability are folded in, so quoting QE alone systematically overstates how many photons a SiPM will actually register.
What limits a SiPM's fill factor?
Every microcell needs a quenching resistor, a readout trace, and a trench that isolates it optically and electrically from its neighbours, and none of that area collects light. Fill factor falls further as microcell pitch shrinks, because the trench width stays roughly fixed while the cell's total area drops — the tradeoff behind choosing small cells for dynamic range over large cells for raw PDE.
Does raising the bias voltage always improve PDE?
It raises trigger probability, Pt, which climbs with overvoltage — the bias held above the breakdown point — but the gain has a ceiling and a cost: dark count rate, afterpulsing, and optical crosstalk all rise alongside it. Manufacturers publish a recommended operating overvoltage that balances detection efficiency against that added noise, rather than pushing Pt toward its maximum.
Can PDE ever reach 100 percent?
Only on paper. Each of QE, FF, and Pt is a probability capped at 100%, so their product is too, but no physical microcell reaches 100% fill factor — isolation trenches and quenching resistors always take some area. Real high-end SiPMs peak in the 40-60% range at their optimum wavelength and bias, well short of that theoretical ceiling.
What happens to PDE at wavelengths away from the sensor's peak?
It falls, because quantum efficiency is wavelength-dependent while fill factor and trigger probability stay essentially fixed for a given bias. A SiPM optimized for blue light near 420 nm can lose much of its QE — and therefore its PDE — toward the red end of the visible spectrum, which is why datasheets plot QE or PDE as a curve across wavelength rather than a single number.