Silicon photomultiplier (SiPM)

A silicon photomultiplier is an array of small single-photon avalanche diodes, called microcells or pixels, connected in parallel to one output. Each microcell is biased a few volts above breakdown and has its own integrated quench resistor, so a photon that triggers an avalanche in one cell produces a standard charge pulse from that cell while the others stay ready. Because each cell gives the same pulse, the summed output is proportional to the number of cells that fired, and a SiPM can resolve 1, 2, 3 or more simultaneous photons as discrete peaks in a charge histogram. Typical devices are 1 × 1 mm² to 6 × 6 mm² with microcell pitches of about 10–75 µm; a 3 × 3 mm² device with 50 µm cells has 3600 of them. Breakdown voltages are commonly 25–55 V depending on the process, the gain is of order 10⁶, and the photon detection efficiency peaks at roughly 40–60% in the blue for current devices. Hamamatsu sells the technology as the MPPC (multi-pixel photon counter).

Gain and bias

When a microcell fires, its capacitance $C$ discharges from the bias $V$ to just above breakdown $V_{bd}$ and is recharged through the quench resistor. The charge per cell sets the gain:

$$G = \frac{C\,(V - V_{bd})}{q}$$

For a cell capacitance of 150 fF at 3 V overvoltage, $G$ = 2.8 × 10⁶, comparable with a photomultiplier tube and far above a linear-mode avalanche photodiode. The recharge time, the product of cell capacitance and quench resistance, is typically tens of nanoseconds. Breakdown voltage rises with temperature, commonly by a few tens of mV/K, so at fixed bias the overvoltage and hence gain and efficiency fall as the device warms; stable operation needs temperature control or bias compensation.

Photon detection efficiency

The photon detection efficiency (PDE) is the product of three factors: the quantum efficiency of the silicon, the geometric fill factor (the active fraction of each cell, reduced by the trenches, resistors and metal between cells), and the probability that a photocarrier triggers a sustained avalanche, which rises with overvoltage. Large cells have a higher fill factor and gain; small cells recover faster, saturate later and have a wider dynamic range. Silicon limits useful sensitivity to roughly 300–900 nm, with NIR-enhanced versions for lidar near 905 nm.

Noise sources

Thermally generated carriers fire cells at random, giving a dark count rate, typically tens to hundreds of kHz per mm² at room temperature, that falls steeply on cooling. This is the SiPM counterpart of dark current, and each dark count is indistinguishable from a single photon. Optical crosstalk occurs when photons emitted by the hot carriers in one avalanche trigger a neighbor, so a single photon produces a two- or three-cell pulse; afterpulsing occurs when trapped carriers are released during recharge. Both are typically a few percent to a few tens of percent, rising with overvoltage, and they add an excess noise factor to what would otherwise be Poisson photon statistics. Trenches between cells reduce crosstalk.

Saturation

Each cell counts at most one photon per recovery time, so the response becomes nonlinear once the number of photons per pulse approaches the number of cells. For short light pulses,

$$N_{\mathrm{fired}} = N_{\mathrm{cells}}\left[1 - e^{-N_{\mathrm{pe}}/N_{\mathrm{cells}}}\right],$$

where $N_{\mathrm{pe}}$ is the expected number of detected photons. A 3600-cell device receiving 1000 photoelectrons fires 873 cells, reading 13% low. Scintillation and calorimetry applications correct for this or choose smaller cells.

Where it is used

SiPMs have replaced photomultiplier tubes in much of positron emission tomography, where they read out scintillator crystals such as LYSO and work inside MRI magnets; they are also used in high-energy physics calorimeters, direct time-of-flight lidar, flow cytometry, and time-correlated single-photon counting. Advantages over the PMT are low bias voltage, insensitivity to magnetic fields, compactness and robustness to accidental illumination; disadvantages are a much higher dark count rate per unit area and temperature-dependent gain. Single-photon timing jitter is typically of order 100 ps for small devices and grows with device capacitance.

Common questions

What is the difference between a SiPM and a SPAD?

A SPAD is one Geiger-mode diode with a binary output. A SiPM is many SPADs summed in analog, so its output amplitude carries the photon number. Digital SiPMs and SPAD arrays read each cell individually instead.

Can a SiPM measure continuous light?

Yes, as a current or a count rate, as long as the rate per cell stays well below the inverse recovery time. At high flux the output saturates and bright light can heat the device.

How is the breakdown voltage found?

By extrapolating gain versus bias to zero gain, or from the knee of the reverse current-voltage curve in the dark.

References: B. E. A. Saleh, M. C. Teich, Fundamentals of Photonics, 3rd ed. (Wiley, 2019); D. Renker, E. Lorenz, "Advances in solid state photon detectors," J. Instrum. 4, P04004 (2009).