Silicon Photomultiplier (SiPM)
Take ten thousand Geiger counters, each the size of a red blood cell, and wire them together in parallel. Now they can count single photons of light — and tell you how many landed.
Detects light pulses ranging from a single photon to tens of thousands of photons per nanosecond with rise times under 1 ns. Unlike a vacuum photomultiplier tube, a SiPM is a few-mm² silicon chip that operates from a single 25–55 V bias, weighs almost nothing, is unaffected by magnetic fields, cannot be damaged by ambient light, and costs a fraction of a comparable PMT. SiPMs now form the photon-counting front-end of medical PET scanners, automotive LiDAR receivers, neutrino observatories, dark-matter detectors, and the high-luminosity LHC's CMS hadronic calorimeter.
The chip is etched into roughly 10,000 microcells per square millimetre, each microcell ~10–50 µm across. Every microcell is an avalanche photodiode reverse-biased a few volts above its breakdown voltage — so-called Geiger mode. In this regime a single photon-induced electron-hole pair triggers a self-sustaining avalanche that briefly draws 100 µA through the cell, producing a binary 'fired' signal. Each microcell is wired to a passive quenching resistor (typically 100 kΩ–1 MΩ polysilicon) in series; when the avalanche current develops, the resistor drops enough voltage to quench the avalanche in 1–10 ns, and the cell recharges to its operating point in 20–100 ns, ready for the next photon. All microcell outputs are connected in parallel on the chip, so the total output current is the sum of however many microcells fired in the same pulse — giving a quasi-analog 'photon counting' waveform whose pulse height resolves individual photon-number peaks (1, 2, 3, ... photoelectrons) up to roughly 100 photons per ~30 ns window before pile-up saturates the device.
In plain terms
Imagine a rain detector made of ten thousand tiny buckets — but each bucket only knows two things: 'a raindrop landed in me' or 'no raindrop yet'. By counting how many of the ten thousand buckets got hit in the same second, you can measure how heavy the rain is. A SiPM does the same trick with light: each tiny bucket is a 'micro-Geiger counter' that flashes when struck by a single photon, and adding them up tells you how bright the light pulse was.
Also called: SiPM, MPPC, Geiger-mode APD array, solid-state photomultiplier.
Why it matters
- SiPMs replace the photomultiplier tube in nearly every new medical PET scanner since ~2014 (e.g. GE Signa PET/MR, Siemens Biograph Vision, Philips Vereos). Their immunity to magnetic fields makes simultaneous PET/MRI possible, which a PMT cannot survive.
- Time-of-flight LiDAR rangefinders in autonomous vehicles use SiPM arrays to time photon arrival to <50 ps, giving few-cm distance resolution at 100+ m range from a sensor smaller than a coin.
- The Hyper-Kamiokande neutrino observatory in Japan (commissioning 2027) will instrument 40,000 large-area photodetectors, with R&D-stage SiPM-based alternatives planned for future upgrades — a generational shift from the vacuum-tube PMTs of the original Super-K.
- The CMS Hadronic Endcap Calorimeter at the LHC was upgraded in 2018 from PMTs to Hamamatsu MPPC SiPMs (4-mm² area each) to survive the high-luminosity LHC's 3000 fb⁻¹ radiation environment — a step PMTs could not have made.
Where it lives
- Hamamatsu MPPC S13360 / S14160 series — 1×1 mm² to 6×6 mm² SiPMs used in PET, LiDAR, and HEP detectors worldwide
- Onsemi (SensL) C-Series / J-Series SiPMs — widely used in time-of-flight LiDAR and consumer time-of-flight cameras
- Broadcom AFBR-S4N series — 4-mm² SiPM arrays for automotive LiDAR with integrated quenching networks
- CMS Hadronic Endcap Calorimeter Upgrade — 15,840 Hamamatsu S12572 SiPMs replacing the PMTs after Long Shutdown 2 (2018–2022)
Key numbers
- Microcell pitch: 10–75 µm (Hamamatsu MPPC family) (Smaller pitch = more cells = higher dynamic range but lower fill factor)
- Photon detection efficiency (peak): ~50% at 420 nm (Hamamatsu S14160) (Comparable to a bialkali PMT)
- Dark count rate at 25 °C: 100 kHz/mm² to 1 MHz/mm² (Falls ~2× per 10 °C cooling)
- Gain: 10⁵ – 10⁶ (Set by operating overvoltage; comparable to vacuum PMT)
- Single-photon timing resolution: ~50–100 ps RMS (Limited by avalanche transit-time spread)
- Operating bias voltage: 25–55 V (vs. >1000 V for PMT) (Single low-voltage supply; easy electronics)
- Breakdown temperature coefficient: ~22 mV/°C (Hamamatsu S14160 datasheet)
Surprising facts
- A typical 1 × 1 mm² SiPM contains 10,000 individual single-photon detectors on a chip thinner than a human hair. Each detector individually says only 'yes I saw a photon' or 'no'; the sum produces a photon-number-resolving waveform.
- The breakdown voltage of a SiPM has a temperature coefficient of roughly 22 mV/°C. To keep gain stable across an outdoor LiDAR sensor over a 100 °C operating range, the bias supply must track temperature to better than ±0.1 °C-equivalent — typically done with a per-channel Peltier or a closed-loop calibration LED.
- Sergio Cova's group at Politecnico di Milano introduced the active quenching circuit for single-photon avalanche diodes in 1981 (Cova et al., Rev. Sci. Instrum. 52, 408), making practical photon counting possible. Hamamatsu's commercial MPPC line followed in the mid-2000s. The transition from laboratory curiosity to dominant photon-counting technology took about a generation.
- SiPMs survive ambient daylight unharmed — they merely saturate. A vacuum photomultiplier tube exposed to room lighting at operating voltage will damage its photocathode within seconds, an asymmetry that revolutionised the practical operation of large detector arrays.
Open questions
- Optical crosstalk between microcells: When a microcell fires, the avalanche generates roughly 3 secondary photons per 10⁵ carriers crossing the depletion region (hot-carrier electroluminescence). Some of these secondary photons travel inside the silicon and trigger neighbouring microcells, producing spurious 'extra' counts that scale with overvoltage. The exact crosstalk probability — typically a few percent — depends on microscopic details of the depletion region geometry and is hard to predict from first principles. SiPM datasheets quote it as a measured parameter, not a calculated one.
A short history
Geiger-mode avalanche behaviour in silicon p-n junctions was first studied by Haitz at Shockley Laboratory in the early 1960s. Sergio Cova's group at Politecnico di Milano established practical single-photon avalanche diodes through the 1980s, building on McIntyre's 1966 theory of avalanche multiplication in silicon and introducing the active quenching circuit in 1981. Through the 1990s, Russian groups (Sadygov, Golovin, Buzhan) showed that arrays of SPAD microcells coupled by quenching resistors and read out on a common bus produced a quasi-analog photon-counting device — the silicon photomultiplier. Hamamatsu Photonics commercialised the multi-pixel photon counter (MPPC) in the mid-2000s. By the mid-2010s, SiPMs had reached parity with PMTs in single-photon timing and detection efficiency at far lower cost, weight, and operating voltage, opening the door to mass deployment in medical imaging, LiDAR, and high-energy physics. The transition was largely complete by 2020, with PMTs retaining a niche only in very-large-area (m²-class) applications where SiPM tiling remains cost-prohibitive.
Sources
- Cova et al. — Rev. Sci. Instrum. 67, 3440 (1996): Avalanche photodiodes and quenching circuits for single-photon detection (https://doi.org/10.1063/1.1147132)
- Hamamatsu MPPC Technical Information Handbook (2024 edition) (https://www.hamamatsu.com/content/dam/hamamatsu-photonics/sites/documents/99_SALES_LIBRARY/ssd/mppc_kapd9008e.pdf)
- Buzhan et al. — Nucl. Instr. Methods A 504, 48 (2003): Silicon photomultiplier and its possible applications (https://doi.org/10.1016/S0168-9002(03)00749-6)
- CMS Collaboration — HCAL Endcap SiPM upgrade TDR (CMS-TDR-010, CERN 2012) (https://cds.cern.ch/record/1481837)
- Acerbi & Gundacker — Nucl. Instr. Methods A 926, 16 (2019): Understanding and simulating SiPMs (https://doi.org/10.1016/j.nima.2018.11.118)