Ask ten particle physicists whether SiPM vs PMT is even a real debate anymore, and eight will tell you silicon won years ago. I think that’s lazy. I spent years reading out photon signals on CERN beamlines, and the honest answer is that silicon photomultipliers have won almost everywhere new detectors get designed — but the old photomultiplier tube is still quietly beating them on the one metric nobody likes to put in a marketing slide: raw collection area per dollar.
That’s the thesis of this piece, and it’s why most “SiPM vs PMT” explainers online are useless: they read like a spec sheet, not a verdict. I’ll give you the numbers, but I’ll also tell you where I’d still specify a vacuum tube in 2026, and why the same silicon economics reshaping particle detectors are, weirdly, the same ones reshaping the optical links inside data centers.
Why Physicists Are Asking SiPM vs PMT At All
A photomultiplier tube is a vacuum-tube relic: a photocathode knocks loose an electron when a photon hits it, and a chain of dynodes multiplies that single electron into a measurable pulse. It’s been the workhorse of particle and nuclear physics since the 1930s, and it still holds the record for sheer collecting area — a single 20-inch PMT, the kind used in neutrino experiments like JUNO, stares at more liquid scintillator than an entire wall of silicon sensors could cover for the same price.
A silicon photomultiplier flips the architecture entirely. It’s an array of thousands of tiny avalanche photodiodes, each one a single-photon avalanche diode (SPAD) running in Geiger mode, wired together on one chip. Instead of a fragile glass tube full of vacuum, you get a solid-state sensor the size of a fingernail that survives being dropped, works next to a magnet, and runs on a bias voltage a 9-volt battery could nearly supply.
Source: Hamamatsu S14161 MPPC datasheet comparison; JUNO Central Detector PMT specifications (HPK/NNVT average)
The Numbers Behind the SiPM vs PMT Debate
On photon detection efficiency, silicon simply wins. A modern SiPM like Hamamatsu’s S14161 hits roughly 50% peak efficiency at 450 nm, while the 20-inch PMTs in JUNO — some of the best tubes ever built — average around 30.1%. That’s not a marginal gap; it means a SiPM-based detector needs meaningfully less scintillator or exposure time to reach the same statistical precision.
The voltage story is even more lopsided. PMTs need kilovolt-level high-voltage supplies to accelerate electrons down the dynode chain — bulky, power-hungry, and a genuine safety consideration in a detector with thousands of channels. SiPMs operate below 50 volts. I’ve built readout boards for both, and the difference isn’t cosmetic: a kilovolt HV distribution system is a subsystem in itself, with its own cabling, shielding, and failure modes. A SiPM array just needs a PCB trace.
Then there’s the magnetic field problem, which is where PMTs really fall apart. Near an accelerator beamline or an MRI magnet, PMT gain can drop by 50% or more in fields above a few hundred gauss, because the electron cascade itself gets bent off course. SiPMs, being solid-state, don’t care about ambient magnetic fields at all — which is exactly why every new detector designed to sit close to a magnet coil defaults to silicon now.
Source: AP Technologies Ltd, SiPM vs PMT technical comparison; Berkeley Nucleonics SiPM/PMT readout notes
Where the Photomultiplier Tube Still Wins
Here’s the part the SiPM cheerleaders skip. A 20-inch PMT has roughly 2,000 times the photocathode area of a single large SiPM tile. To match that collecting area with silicon, you tile hundreds of SiPMs together — and every added tile brings its own dark counts, its own readout channel, and its own cost. For a huge, sparse detector volume like a kiloton-scale neutrino tank, a wall of cheap glass tubes still beats a wall of silicon on cost-per-square-meter, full stop.
Dark counts used to be silicon’s other weakness: early SiPMs ran around 1 MHz per mm² of spurious noise pulses. That’s improved enormously — modern devices are down below 100 kHz per mm², and the rate roughly halves for every 8°C you cool the sensor. But it hasn’t gone to zero, and for ultra-low-background searches like dark matter detection, that noise floor still matters in a way it simply doesn’t for a PMT sitting at room temperature.
The Data Center Parallel Nobody’s Drawing
Here’s my contrarian bit. The exact same avalanche-photodiode physics inside a SiPM — a semiconductor junction multiplying a single absorbed photon into a detectable current — is what sits behind the receiver end of high-speed fiber-optic links, including the ones now proliferating inside AI data centers as operators chase lower-latency, higher-bandwidth interconnects. It’s the same underlying transition happening in both worlds: bulky, high-voltage, single-purpose hardware giving way to cheap, low-voltage silicon that scales by adding more identical units rather than building one bigger tube.
I don’t think that’s a coincidence — it’s the same manufacturing curve. Once a photon-counting technology moves onto a CMOS-compatible silicon process, it inherits the entire semiconductor industry’s cost trajectory, the same way thin silicon tracking detectors did. Vacuum tubes never get that ride down the cost curve, no matter how good the glassblowing gets. That’s the deeper reason SiPM keeps winning new designs even in places where PMTs still out-perform them on raw numbers today.
It also explains why the SiPM roadmap looks nothing like the PMT roadmap looked. PMT improvements came in increments measured in decades — a better photocathode alloy here, a faster dynode geometry there. SiPM performance instead tracks silicon fab nodes: photon detection efficiency, dark count rate, and cell density have all improved on something closer to a Moore’s-law curve than a vacuum-tube one, according to the manufacturer comparisons from AP Technologies and the readout benchmarking from Berkeley Nucleonics. That’s the pattern I’d bet on continuing: not a sudden PMT extinction, but a shrinking set of niches where a 90-year-old vacuum tube remains the economically correct answer.
⚡ PHOTON’S TAKE
SiPM vs PMT isn’t a contest silicon has finished winning — it’s a contest silicon wins by default every time a detector gets redesigned from scratch. PMTs survive where you need brute collecting area on a budget, which is why JUNO is still full of glass tubes today. But every new experiment I’ve seen specified in the last five years defaults to silicon unless someone actively argues against it. That’s the real signal: the burden of proof flipped.
What Comes Next for Photon Detection
I’d bet the next decade doesn’t produce a clean winner — it produces hybrids. Large-volume neutrino and dark matter experiments will keep using PMTs where area-per-dollar dominates, while every compact, magnet-adjacent, or timing-critical application keeps migrating to SiPM. The interesting frontier is single-photon avalanche diode arrays that go fully digital, timestamping individual photons on-chip instead of just counting current — early versions already exist in lab prototypes, and I expect that’s where the real disruption lands by the early 2030s.
What I’d watch for next: whether that same digital-SPAD approach jumps from physics labs into commercial optical-networking silicon the way avalanche photodiodes already have. If it does, the CERN-to-data-center pipeline I’ve been tracking on this site — from my own path out of particle detector electronics — will have run in both directions at once. That’s the pattern worth watching, not the tired “silicon eats everything” headline every other outlet already ran.
