CERN’s detector upgrades show how radiation tolerance, precision timing and cooling must work together. The decisive engineering challenge is preserving a reliable measurement throughout the instrument’s life.
The measurement changes with the machine
A particle detector is usually introduced through its best performance: a spatial resolution, a timing precision or an ability to distinguish one particle from another. At a collider, those numbers have to survive an environment that gradually changes the instrument itself. Radiation modifies materials, temperature affects electronic response and successive collisions leave overlapping signals. The detector must remain understandable while all three evolve.
For Europe’s detector manufacturers and research laboratories, this makes longevity a measurement problem as much as a materials problem. A robust sensor, a stable cooling circuit and a reliable calibration procedure solve different parts of the same task. Improvements become scientifically useful when they can be carried through production, integration and operation without losing the connection between an electrical signal and the particle that generated it.
Timing has to survive radiation
The ATLAS High-Granularity Timing Detector is designed to distinguish tracks from collisions that are close together in space but separated in time. Its technical design sets an average timing target per track of 30 picoseconds at the beginning of its life, rising to 50 picoseconds after the planned High-Luminosity LHC exposure. Those are lifetime design targets, rather than an assertion that the complete installed detector has already achieved them.
One picosecond is a trillionth of a second. Achieving precision on this scale requires more than a fast semiconductor. The time recorded by the electronics must be related consistently to the particle’s arrival, and measurements from several detector elements must remain mutually aligned. A timing number is therefore meaningful only alongside its measurement conditions and whether it describes one hit or a reconstructed track.
Radiation tolerance also starts inside the sensor. A study of carbon-enriched low-gain avalanche detectors tested devices from different vendors after irradiation at two specified fluences. Carbon was added to the gain layer to reduce radiation-induced removal of electrically active acceptors, which affects internal signal amplification. The researchers measured collected charge, timing and hit efficiency in beam tests during 2021 and 2022. This was evidence about irradiated sensors under test, not a complete lifetime demonstration of an assembled experiment.
Cooling can change the ageing strategy
CMS’s barrel timing detector uses a different approach: scintillating crystals produce light that silicon photomultipliers detect. Radiation increases unwanted counts in those photomultipliers. A 2023 development study showed how small thermoelectric coolers could lower their operating temperature beyond that of the surrounding detector. The authors reported that cooling from about minus 35°C to minus 45°C approximately halved the dark count rate, with the extra cooler power largely offset by reduced leakage-current power. Thermoelectric cooling study
The same study explored reversing the coolers during technical stops to warm the photomultipliers and accelerate annealing, partly recovering performance. This makes the maintenance cycle part of the detector design. It does not remove radiation damage or establish the final instrument’s lifetime performance, but it shows why a component should be assessed across its operating and recovery states.
At infrastructure scale, CERN reported commissioning its first primary carbon dioxide cooling system in May 2026. The surface installation supplies the detector cooling units and extends CO₂ through the cooling chain; CERN reported eleven units operating at CMS and seven at ATLAS. The change addresses the use of refrigerants with much higher global warming potential as well as thermal engineering.
For suppliers, the implication is practical. Thermal stability, pressure management and service access belong in the detector’s acceptance requirements. A cooling plant reaching its operating point is an important milestone, but the eventual physics performance still depends on how the plant behaves with the final instrument, its changing heat loads and its control systems.
More channels create a readout problem
Fine segmentation helps researchers separate overlapping particle showers, but it also multiplies the information that must leave the detector. CMS’s High-Granularity Calorimeter combines silicon pads with scintillating tiles. A 2025 collaboration report described approximately six million silicon pads and 280,000 tiles, with data and trigger information transmitted through optical fibres numbering in the tens of thousands. The report focuses on progressively more complex beam tests and complete prototype readout chains. HGCAL system validation
This changes the significance of a successful beam test. Demonstrating that a sensor responds correctly answers one question. Demonstrating that its signal can be powered, digitised, transmitted and interpreted through the intended chain answers several more. Neither automatically establishes the reliability of every production unit, so the sequence of tests matters.
Production teams can make that distinction visible through separate records for electrical acceptance, mechanical assembly and performance after integration. If an assembly fails, those records help establish whether the underlying cause was a sensor defect, a connection or an operating condition. They also make substitution decisions more disciplined. A replacement component with an apparently equivalent headline specification should still be checked against the relevant interfaces and calibration assumptions. The measurement chain is the product being delivered, so its documentation should connect component characteristics with the conditions under which the assembled instrument was accepted.
ATLAS illustrates this distinction in its timing programme. CERN’s experimental physics department reported a 42-picosecond per-hit result from a pre-production sensor and readout-chip assembly. Separately, a 54-module demonstrator combined prototype sensors and chips with power, interlocks, cooling and the readout system. The second exercise addresses integration rather than simply repeating the first timing measurement at a larger scale.
Removing material introduces another manufacturing test
ALICE’s ITS3 upgrade shows why making a detector stronger is not always the objective. Material near the collision point can disturb the particles being measured. Its planned inner tracker therefore uses very thin silicon curved around the beam pipe, with air cooling and lightweight support. CERN’s account of the MOSAIX development describes a 26.6-centimetre-long prototype divided into 144 independently powered sensor tiles. Defective areas can be isolated rather than necessarily disabling the whole device.
That report records delivery of 36 wafers in February 2026 and demonstration of overall chip functionality, while yield and performance characterisation continued. The distinction is substantial: functional prototypes establish that an architecture can work; production characterisation establishes how consistently it can be made.
Such designs also change the economics of testing. When a large device contains many independently controlled regions, a simple pass-or-fail label may conceal useful information. Acceptance criteria need to specify which defects are tolerable, where they are located and how they affect the intended measurement. That is an engineering judgement to establish before production volumes increase, rather than an assumed consequence of using a larger wafer.
Calibration preserves the meaning of the signal
Ageing does not always mean immediate failure. CMS’s electromagnetic calorimeter provides an example in which radiation changes the transparency of lead tungstate crystals. A laser monitoring system tracks those optical changes. A 2023 paper released monitoring data from 2016–2018 and presented a machine-learning example for predicting crystal behaviour. The publication describes a dataset and prediction approach; it does not show that machine learning can dispense with measurements of the instrument.
Calibration therefore deserves the same design attention as the sensor and its electronics. If the response changes, researchers need a way to recognise the change and relate signals taken at different times. Forecasts may help plan interventions or study trends, while reference measurements establish what actually happened.
The strongest industrial opportunity in these programmes is consequently broader than supplying a component with a record specification. It includes reproducible assembly, documented tests, repairable services and calibration systems that allow researchers to distinguish changing hardware from changing physics. Contracts and technical reviews should make those interfaces explicit.
A useful final acceptance question is whether another team could reconstruct why a channel was accepted, how it was operated and which corrections were applied to its data. Retaining that history makes a complex detector auditable over its scientific lifetime. Precision then becomes a capability that can be maintained, rather than a number recorded once on a laboratory bench.
