Single-photon avalanche diode (SPAD)

A single-photon avalanche diode is a p-n junction reverse-biased a few volts beyond its breakdown voltage, the excess bias. In this Geiger mode the junction is unstable: one photogenerated carrier entering the high-field region can start an avalanche that sustains itself and grows to milliamperes, so a single photon produces a pulse large enough to drive a logic gate directly. That is the difference from a linear-mode avalanche photodiode, which is biased below breakdown and gives an output proportional to the light with a finite, noisy gain; the SPAD gives a digital click whose size carries no information and whose timing carries all of it.

An avalanche that sustains itself must be stopped. A quenching circuit, either a series resistor (passive quenching) or a fast comparator that pulls the bias below breakdown (active quenching), ends the avalanche and then restores the bias. During that interval the detector is blind, and this dead time, typically tens of nanoseconds, caps the count rate. For a detector that ignores photons arriving while it is dead, the measured rate is $R_m = R/(1+R\tau_d)$; with $\tau_d$ = 50 ns, a true rate of 10 million counts per second is recorded as 6.67 million, and the measured rate saturates at $1/\tau_d$ = 20 million.

Four figures characterize a SPAD. Photon detection efficiency is the product of quantum efficiency, the probability that a carrier triggers an avalanche (which rises with excess bias), and, for arrays, the fill factor. The dark count rate comes from thermal generation and trap-assisted tunneling in the multiplication region and plays the role that dark current plays in a linear detector; it falls steeply with cooling, which is why many modules run on a Peltier stage. Timing jitter, the spread in the delay from photon to output edge, is tens of picoseconds in thin silicon junctions and longer in thick ones. Afterpulsing is the release of carriers trapped during one avalanche, which triggers a false count after the bias returns; it is managed by a hold-off time that lengthens the dead time.

Material sets the spectral range. Silicon SPADs work from the blue to the near infrared and are made in standard CMOS processes, which allows arrays of thousands to millions of pixels with timing electronics beside each one; these are the sensors of direct time-of-flight lidar, including those in consumer phones, and of fluorescence-lifetime imaging. At 1550 nm, InGaAs/InP SPADs are used, and their high afterpulsing usually forces gated operation, with the bias raised only in windows where a photon is expected; their efficiency and dark counts at telecom wavelengths are well short of an SNSPD, which they compete against on cost and on operation without a cryostat. In lidar the SPAD array is the receiver for pulsed time-of-flight systems, the alternative to the coherent receiver of FMCW lidar.

References: S. Cova et al., Appl. Opt. 35, 1956 (1996); R. H. Hadfield, Nat. Photonics 3, 696 (2009); M. D. Eisaman et al., Rev. Sci. Instrum. 82, 071101 (2011).