TRISTAN Detector: KATRIN Upgrade

• TRISTAN detector is a silicon-drift-detector array that upgrades KATRIN for full-spectrum differential tritium-β spectroscopy aimed at keV-scale sterile neutrino searches.
• It uses 166-pixel SDD modules with advanced readout electronics and cooling, achieving high throughput (~10⁸ cps total) and energy resolution better than 300 eV FWHM.
• The system rigorously addresses detector response challenges—dead-layer losses, charge sharing, and ADC non-linearity—to ensure precise spectral measurements.

Searching arXiv for recent TRISTAN detector papers in the KATRIN context. The TRISTAN detector is the silicon-drift-detector focal-plane array and readout upgrade developed for the KATRIN experiment to convert it from an endpoint spectrometer for the active-neutrino mass into a high-statistics, differential tritium-$\beta$-spectroscopy instrument for a search for keV-scale sterile neutrinos. In this measurement mode, a sterile state of mass $m_4$ contributes a second branch to the spectrum,

$$\frac{d\Gamma}{dE_e} \;=\; (1-|U_{e4}|^2)\,\frac{d\Gamma}{dE_e}(m_\nu) + |U_{e4}|^2\,\frac{d\Gamma}{dE_e}(m_4),$$

and produces a kink-like distortion at $E_e = E_0 - m_4$, with $E_0 \simeq 18.6\,\mathrm{keV}$. TRISTAN is therefore designed for full-spectrum operation, very high throughput, and electron-response control at a level compatible with searches down to approximately $|U_{e4}|^2 \sim 10^{-6}$ (Houdy et al., 2020, Acharya et al., 24 Mar 2026).

1. Scientific role within KATRIN

In standard KATRIN operation, only the last $\sim 40$ eV below the tritium endpoint are analyzed, because that region is most sensitive to the active neutrino mass. TRISTAN changes the measurement concept: the spectrometer retarding potential is lowered, a broad energy interval is transmitted, and the detector measures the differential electron spectrum directly rather than serving only as a counter behind an integral MAC-E-filter scan (Houdy et al., 2020, Brunst et al., 2018).

The rate scale of this program is set by the tritium source. The KATRIN source activity is about $10^{11}\,\mathrm{Bq}$, and the full-spectrum TRISTAN mode was described as a case in which about $10^{10}$ electrons/s in a 20 cm beam tube must be handled. The detector system itself is therefore designed for operation at about $10^8\,\mathrm{cps}$ total, or about $m_4$0 per pixel, while maintaining an energy resolution better than $m_4$1 FWHM at $m_4$2–$m_4$3 and a threshold of about $m_4$4 (Houdy et al., 2020, Urban et al., 2021, Brunst et al., 2018).

The physics motivation is inseparable from systematics. A sterile-neutrino signal is a ppm- to sub-ppm-level shape distortion distributed over a wide energy range, so TRISTAN must control dead-layer response, backscattering, charge sharing, pile-up, readout nonlinearity, and long-term stability. This is why the detector upgrade is not merely a higher-rate focal plane, but a dedicated spectroscopic system optimized for electron response (Houdy et al., 2020, Mertens et al., 2020).

2. Detector architecture and evolution of the design

TRISTAN is based on multi-pixel silicon drift detectors (SDDs). The choice of SDD technology follows directly from the need to combine large pixel area with small anode capacitance: in an SDD, charge created in the silicon bulk is guided by drift rings to a very small anode, so capacitance remains in the few-$m_4$5\,fF range even for millimeter-scale pixels. This permits short shaping times and good energy resolution at high rate (Brunst et al., 2018, Urban et al., 2021).

The early full detector concept was a $m_4$6 focal plane segmented into 3486 hexagonal pixels of 3 mm diameter, divided into 21 identical modules. That concept resulted from simulations balancing several competing constraints: increasing the number of pixels to minimize pile-up, increasing pixel size to avoid charge sharing, maximizing transverse momentum to reduce backscattering, reducing the Larmor radius to avoid back-reflection into different pixels, and keeping overall complexity manageable (Houdy et al., 2020).

Later hardware development established a staged path of 7-pixel prototypes, an intermediate 47-pixel module, and the final 166-pixel module geometry. The 47-pixel device already followed the final module concept, with electronics arranged behind the detector chip so that closely packed tiling remains possible. In the subsequent KATRIN baseline, the focal plane consists of 9 detector modules of 166 pixels each, giving about 1500 pixels, while a 21-module configuration remains the larger-scale option (Urban et al., 2021, Siegmann et al., 2024, Gavin et al., 4 Feb 2026).

The 166-pixel module geometry is now the defining TRISTAN unit. Each monolithic chip has 166 hexagonal pixels, a chip size of about $m_4$7, a sensitive area of about $m_4$8, a thickness of about $m_4$9, and pixel dimensions corresponding to a circumscribed diameter of $$\frac{d\Gamma}{dE_e} \;=\; (1-|U_{e4}|^2)\,\frac{d\Gamma}{dE_e}(m_\nu) + |U_{e4}|^2\,\frac{d\Gamma}{dE_e}(m_4),$$0 and an equivalent circular diameter by area of $$\frac{d\Gamma}{dE_e} \;=\; (1-|U_{e4}|^2)\,\frac{d\Gamma}{dE_e}(m_\nu) + |U_{e4}|^2\,\frac{d\Gamma}{dE_e}(m_4),$$1. In the KATRIN detector section, the focal-plane diameter is roughly 20 cm (Houdy et al., 2020, Siegmann et al., 2024).

3. Module engineering, cooling, and readout chain

The first full TRISTAN module was conceived as one of the final detector sectors: a column about 16 cm high with a $$\frac{d\Gamma}{dE_e} \;=\; (1-|U_{e4}|^2)\,\frac{d\Gamma}{dE_e}(m_\nu) + |U_{e4}|^2\,\frac{d\Gamma}{dE_e}(m_4),$$2 cross-section, carrying a $$\frac{d\Gamma}{dE_e} \;=\; (1-|U_{e4}|^2)\,\frac{d\Gamma}{dE_e}(m_\nu) + |U_{e4}|^2\,\frac{d\Gamma}{dE_e}(m_4),$$3, 450 $$\frac{d\Gamma}{dE_e} \;=\; (1-|U_{e4}|^2)\,\frac{d\Gamma}{dE_e}(m_\nu) + |U_{e4}|^2\,\frac{d\Gamma}{dE_e}(m_4),$$4m thick SDD chip produced by the Max Planck Semiconductor Laboratory. The module had to operate inside KATRIN’s Monitor Spectrometer environment, about 1 m from the vacuum-to-air interface and inside a 6 T superconducting magnet, which imposed strong constraints on material choice, thermal design, and mechanical support (Houdy et al., 2020).

The detector architecture places amplification as close to the sensor as possible. TRISTAN pixels use an integrated JFET at the anode, followed by the custom ETTORE ASIC, a 12-channel low-noise charge-sensitive front end developed specifically for the project. The total anode capacitance quoted for the final-style pixels is $$\frac{d\Gamma}{dE_e} \;=\; (1-|U_{e4}|^2)\,\frac{d\Gamma}{dE_e}(m_\nu) + |U_{e4}|^2\,\frac{d\Gamma}{dE_e}(m_4),$$5. The readout electronics are mounted behind the detector plane, connected through wire bonds and flex circuits, because closely packed module tiling is required in the full focal-plane array (Houdy et al., 2020, Urban et al., 2021).

The mechanical implementation evolved together with the module design. Early module integration used a carbon-fiber-reinforced silicon-carbide structure attached to a copper block and cooled to $$\frac{d\Gamma}{dE_e} \;=\; (1-|U_{e4}|^2)\,\frac{d\Gamma}{dE_e}(m_\nu) + |U_{e4}|^2\,\frac{d\Gamma}{dE_e}(m_4),$$6 by a liquid-cooling circuit in vacuum. Later 166-pixel modules used a Cesic silicon-ceramic composite interposer, a copper cooling structure, and electronics boards mounted perpendicular to the detector plane. Cooling near $$\frac{d\Gamma}{dE_e} \;=\; (1-|U_{e4}|^2)\,\frac{d\Gamma}{dE_e}(m_\nu) + |U_{e4}|^2\,\frac{d\Gamma}{dE_e}(m_4),$$7 to $$\frac{d\Gamma}{dE_e} \;=\; (1-|U_{e4}|^2)\,\frac{d\Gamma}{dE_e}(m_\nu) + |U_{e4}|^2\,\frac{d\Gamma}{dE_e}(m_4),$$8 is used to suppress leakage current while limiting mechanical stress (Houdy et al., 2020, Siegmann et al., 2024).

The DAQ architecture matured into a dedicated remote analog-to-digital conversion system. One Tile Main Board serves one 166-pixel module. Each board uses 21 ADC cards, each with an 8-channel AD9257 ADC, providing 14-bit digitization at up to 65 MHz and a $$\frac{d\Gamma}{dE_e} \;=\; (1-|U_{e4}|^2)\,\frac{d\Gamma}{dE_e}(m_\nu) + |U_{e4}|^2\,\frac{d\Gamma}{dE_e}(m_4),$$9 input range. Data are transported optically, with a per-board streaming capacity of $E_e = E_0 - m_4$0 through 12 Samtec FireFly links. On the backend, Serenity-S1 FPGA cards process three modules each, so three cards cover the 9-module baseline. The firmware provides a trigger filter with $E_e = E_0 - m_4$1 between $E_e = E_0 - m_4$2 and $E_e = E_0 - m_4$3, a trapezoidal energy filter with shaping times between $E_e = E_0 - m_4$4 and $E_e = E_0 - m_4$5, and trigger efficiency better than 99% for $E_e = E_0 - m_4$6 events (Gavin et al., 4 Feb 2026).

4. Electron-response physics and calibration strategy

TRISTAN’s central detector problem is that electrons do not behave like x rays at the detector entrance window. The papers repeatedly identify four dominant non-Gaussian response terms: entrance-window or dead-layer losses, backscattering, charge sharing, and pile-up. These effects are benign for many x-ray applications but directly relevant for a sterile-neutrino search because they distort the measured $E_e = E_0 - m_4$7-spectrum shape (Brunst et al., 2018, Gugiatti et al., 2020).

The entrance window is especially important. For early prototype characterization, a simplified step-like dead-layer model was used with $E_e = E_0 - m_4$8 conversion electrons and detector tilting, yielding a dead layer in the order of $E_e = E_0 - m_4$9, with quoted values such as $E_0 \simeq 18.6\,\mathrm{keV}$0, $E_0 \simeq 18.6\,\mathrm{keV}$1, and $E_0 \simeq 18.6\,\mathrm{keV}$2 depending on the line used. A later SEM- and KESS-based study quoted $E_0 \simeq 18.6\,\mathrm{keV}$3 for an effective dead layer (Lebert et al., 2020, Mertens et al., 2020).

Subsequent Geant4-based work refined this into a surface model with an intentionally grown $E_0 \simeq 18.6\,\mathrm{keV}$4 $E_0 \simeq 18.6\,\mathrm{keV}$5 layer and a depth-dependent charge-collection efficiency,

$E_0 \simeq 18.6\,\mathrm{keV}$6

with fitted values $E_0 \simeq 18.6\,\mathrm{keV}$7, $E_0 \simeq 18.6\,\mathrm{keV}$8, and $E_0 \simeq 18.6\,\mathrm{keV}$9 in one dedicated backscattering study (Spreng et al., 2024).

Charge sharing was measured directly with a 7-pixel detector irradiated by $|U_{e4}|^2 \sim 10^{-6}$0. Using a $|U_{e4}|^2 \sim 10^{-6}$1 coincidence window, the observed fraction of charge-sharing events in the central pixel was $|U_{e4}|^2 \sim 10^{-6}$2, the effective threshold was $|U_{e4}|^2 \sim 10^{-6}$3–$|U_{e4}|^2 \sim 10^{-6}$4, and the inferred Gaussian charge-cloud size was $|U_{e4}|^2 \sim 10^{-6}$5. These events populate the low-energy tail in single-pixel spectra but can be reconstructed by summing coincident neighboring channels (Urban et al., 2021).

Backscattering is both a local detector effect and a beamline-coupled effect. In a tandem-detector study with TRISTAN hardware, the evaluated backscattering coefficient for silicon at $|U_{e4}|^2 \sim 10^{-6}$6 and normal incidence was about $|U_{e4}|^2 \sim 10^{-6}$7, and the coefficient more than doubled between $|U_{e4}|^2 \sim 10^{-6}$8 and $|U_{e4}|^2 \sim 10^{-6}$9 incidence. Because KATRIN’s electric and magnetic fields can return some backscattered electrons to the detector, these measurements are directly relevant for the final response model (Spreng et al., 2024).

To characterize full modules with electrons rather than x rays, the collaboration developed a thermionic electron gun delivering mono-energetic electrons up to $\sim 40$0 and rates in the order of $\sim 40$1 electrons per second. Under scanned full-module illumination, the residual pixel-to-pixel rate variation was about 12%, the total detected rate reached about $\sim 40$2, and a $\sim 40$3 electron peak with $\sim 40$4 FWHM was observed, close to the quoted silicon Fano limit of $\sim 40$5 FWHM (Urban et al., 2024).

5. Prototype campaigns and measured performance

The TRISTAN program advanced through several prototype generations. Early 7-pixel SDDs with hexagonal pixels of 0.5, 1, and 2 mm diameter were used for the first x-ray and electron campaigns. With a 2 mm pixel detector, the best room-temperature energy resolution was $\sim 40$6 FWHM at a peaking time of $\sim 40$7, while cooling to $\sim 40$8 shifted the optimum to $\sim 40$9 and improved the resolution to $10^{11}\,\mathrm{Bq}$0 FWHM (Brunst et al., 2018).

The same prototype line was operated in realistic neutrino-spectroscopy environments. At the Troitsk $10^{11}\,\mathrm{Bq}$1-mass experiment, a 7-pixel TRISTAN detector was run in both integral and differential mode, and mono-energetic electron measurements from $10^{11}\,\mathrm{Bq}$2 to $10^{11}\,\mathrm{Bq}$3 yielded energy resolutions between $10^{11}\,\mathrm{Bq}$4 and $10^{11}\,\mathrm{Bq}$5 FWHM at a peaking time of $10^{11}\,\mathrm{Bq}$6. That campaign produced the first sterile-neutrino constraints using TRISTAN technology and demonstrated compatibility with a MAC-E filter beamline (Brunst et al., 2019).

The intermediate 47-pixel module was then commissioned at the KATRIN Monitor Spectrometer. Using the $10^{11}\,\mathrm{Bq}$7 $10^{11}\,\mathrm{Bq}$8-32 conversion line at about $10^{11}\,\mathrm{Bq}$9 and operating at about $10^{10}$0, the measured energy resolution was homogeneously distributed in the range $10^{10}$1–$10^{10}$2 FWHM over the read-out pixels. In the reported setup, 43 of 47 pixels were read out, and per-pixel rates were $10^{10}$3–$10^{10}$4 (Urban et al., 2021).

The first 166-pixel module provided the decisive hardware demonstration. For detector S0-166-4 in the x-ray bench setup, the mean resolution at the Mn-$10^{10}$5 line was $10^{10}$6 FWHM, while the average at $10^{10}$7 x rays was $10^{10}$8 FWHM. For detector S0-166-6 in the electron-gun setup, the mean resolution at $10^{10}$9 electrons was $10^8\,\mathrm{cps}$0 FWHM, satisfying the TRISTAN requirement of $10^8\,\mathrm{cps}$1 FWHM at $10^8\,\mathrm{cps}$2 (Siegmann et al., 2024).

Additional module results quantify calibration quality and stability. The mean electronic noise at the operating point was $10^8\,\mathrm{cps}$3. The measured peak positions deviated by less than 0.5% from linearity over the tested calibration range, and per-pixel calibration slopes differed by up to 4%, so each pixel requires its own calibration. Over a 63 h stability run, count-rate fluctuations followed the expected $10^8\,\mathrm{cps}$4 decay law with residual fluctuations at most 0.08%, most pixel gain drifts remained below 5 eV with a maximum of 10 eV, and resolution variations stayed within 1 eV (Siegmann et al., 2024).

The shaping-time dependence makes the high-rate tradeoff explicit. The best x-ray resolution, $10^8\,\mathrm{cps}$5 FWHM at $10^8\,\mathrm{cps}$6, occurred at $10^8\,\mathrm{cps}$7, but at $10^8\,\mathrm{cps}$8 per pixel more than 60% of events would be discarded due to pile-up. At $10^8\,\mathrm{cps}$9, the resolution was $m_4$00 FWHM and more than 80% of events were retained, which is closer to the intended TRISTAN operating regime (Siegmann et al., 2024).

6. Systematics, sensitivity, and scope of the name

The mature TRISTAN program is now defined by detector-response modeling and DAQ-controlled systematics rather than by basic proof of principle. In the dedicated KATRIN sensitivity study, a four-month TRISTAN campaign with about 80% detector livetime and approximately $m_4$01 recorded events had the statistical power to probe $m_4$02 at the level of $m_4$03 for sterile-neutrino masses in the $m_4$04–$m_4$05 range. The same study found that the major experimental systematics investigated reduce the sensitivity by a factor of 10–50 over the same mass range (Acharya et al., 24 Mar 2026).

Among the DAQ-related effects, ADC non-linearity received a dedicated analysis. The measured differential non-linearity showed a stable comb-like structure with spike amplitudes of about $m_4$06. In the TRISTAN remote-ADC architecture, however, this effect can be reduced by operating at high count rate, by larger post-acceleration, by an analog Gatti slider offset, and by averaging over many uncorrelated channels. The resulting contribution to the sterile-neutrino sensitivity was shown to be reducible to a subdominant level (Gavin et al., 4 Feb 2026).

A common source of confusion is the name itself. In the KATRIN literature, TRISTAN denotes the multi-pixel SDD detector and readout upgrade for sterile-neutrino searches. The same name also appears in unrelated contexts, including an autonomous three-plane RPC cosmic-ray detector used in a 2018–2019 sea-level latitude survey and the proposed $m_4$07TRISTAN collider program; these are distinct from the KATRIN detector system (Saraiva et al., 2020, Hamada et al., 2022).

In the KATRIN context, the term has a precise meaning: TRISTAN is the detector technology that enables the experiment’s transition from endpoint counting to precision, high-rate, full-spectrum electron spectroscopy. Its defining features are the 166-pixel SDD module, integrated JFET and ETTORE-based low-capacitance readout, backend digital signal processing designed for about $m_4$08 counts per second per pixel, and a detector-response program built around dead-layer characterization, charge-sharing reconstruction, and backscattering modeling. Together, these elements establish TRISTAN as the central hardware platform for KATRIN’s planned keV sterile-neutrino search (Houdy et al., 2020, Siegmann et al., 2024).