FBK VUV-HD3 SiPM for LXe Detectors

• Fondazione Bruno Kessler’s VUV-HD3 is a vacuum-ultraviolet-sensitive SiPM featuring a novel triple-doping process that reduces afterpulsing for enhanced performance in LXe detectors.
• The device achieves a measured 24.3% PDE at 175 nm, low dark count rates, and high single-photoelectron gain, meeting stringent nEXO operational requirements.
• Its design incorporates a thick SiO₂ surface layer that induces distinctive optical interference patterns, affecting performance in both vacuum and in-situ liquid xenon environments.

Fondazione Bruno Kessler’s VUV-HD3, also written VUVHD3 in some nEXO publications, is a vacuum-ultraviolet-sensitive silicon photomultiplier developed for cryogenic noble-liquid detectors, especially liquid-xenon systems operating near the xenon scintillation wavelength of $175\ \mathrm{nm}$. It is the third generation of FBK’s VUV-HD line and was engineered for nEXO with a triple-doping process intended to reduce afterpulsing while preserving high VUV response. Across a sequence of measurements in vacuum, in liquid xenon, and in dedicated optical microscopy and spectroscopy setups, the device has been characterized at the levels of PDE, reflectance, correlated noise, secondary photon emission, and detector-level in-situ efficiency (Gallina et al., 2022).

1. Device definition, lineage, and reported physical form

The VUV-HD3 is a P-on-N SiPM intended for LXe scintillation readout at cryogenic temperature. In the nEXO characterization campaign it is described as the newest VUV-sensitive FBK SiPM generation, developed specifically for operation around $163\ \mathrm{K}$ and around $175\ \mathrm{nm}$, with the key process change being a novel triple-doping technology to suppress afterpulses and permit operation at higher over-voltage (Gallina et al., 2022). All characterized FBK VUVHD3 samples in that study came from the same engineering wafer produced for nEXO.

Different studies report the device in slightly different but compatible geometrical conventions. The nEXO performance study reports a photosensitive area of $5.96\times 5.56\ \mathrm{mm^2}$, a microcell pitch of $35\times 35\ \mu\mathrm{m^2}$, and bare-die packaging (Gallina et al., 2022). A dark-emission spectroscopy study reports the FBK VUV-HD3 as a nominal $6\times 6\ \mathrm{mm^2}$ SiPM with $80\%$ fill factor and the same $35\times 35\ \mu\mathrm{m^2}$ SPAD pitch (McLaughlin et al., 2021). An external-cross-talk study likewise lists it as a $6\times 6\ \mathrm{mm^2}$ bare device, with a breakdown voltage of $29.3\ \mathrm{V}$ at $163\ \mathrm{K}$0 (Guan et al., 2023).

At $163\ \mathrm{K}$1, the weighted-average breakdown voltage reported for nEXO samples is $163\ \mathrm{K}$2, with temperature coefficient $163\ \mathrm{K}$3 and microcell capacitance $163\ \mathrm{K}$4 (Gallina et al., 2022). The recovery time constant is $163\ \mathrm{K}$5, attributed to a polysilicon quench path, and is substantially longer than the corresponding HPK VUV4 value reported in the same study (Gallina et al., 2022). A room-temperature dark-emission study reports $163\ \mathrm{K}$6 breakdown at $163\ \mathrm{K}$7, consistent with the expected thermal shift (McLaughlin et al., 2021).

2. Surface stack, thin-film optics, and optical parameterization

A defining feature of the VUV-HD3 is its thick silicon-dioxide surface layer. The nEXO study states that VUVHD3 shares the same surface coating stack as VUVHD1, including a $163\ \mathrm{K}$8 $163\ \mathrm{K}$9 cover layer, and explicitly links the observed interference oscillations in vacuum PDE to that layer (Gallina et al., 2022). A later analytic PDE fit constrained the oxide thickness more tightly to $175\ \mathrm{nm}$0, with campaign-to-campaign consistency around $175\ \mathrm{nm}$1–$175\ \mathrm{nm}$2 in angular-interference scans (Croix et al., 21 Aug 2025). In LoLX-2 the FBK installation is windowless, so the optical response is governed directly by the SiO$175\ \mathrm{nm}$3-on-Si stack rather than by an added quartz window (Li et al., 17 Oct 2025).

The analytic PDE model used for VUV-HD3 factorizes the response as

$175\ \mathrm{nm}$4

with

$175\ \mathrm{nm}$5

so that geometrical fill factor, thin-film transmission, and internal avalanche/collection efficiency are separated explicitly (Croix et al., 21 Aug 2025). For the preferred global parameterized fit, the reported FBK VUV-HD3 parameters are:

Parameter	Best-fit value
$175\ \mathrm{nm}$6	$175\ \mathrm{nm}$7
$175\ \mathrm{nm}$8	$175\ \mathrm{nm}$9
$5.96\times 5.56\ \mathrm{mm^2}$0	$5.96\times 5.56\ \mathrm{mm^2}$1
$5.96\times 5.56\ \mathrm{mm^2}$2	$5.96\times 5.56\ \mathrm{mm^2}$3
$5.96\times 5.56\ \mathrm{mm^2}$4	$5.96\times 5.56\ \mathrm{mm^2}$5
$5.96\times 5.56\ \mathrm{mm^2}$6	$5.96\times 5.56\ \mathrm{mm^2}$7

These values place the effective front collection boundary essentially at the surface and the junction center at about $5.96\times 5.56\ \mathrm{mm^2}$8, which is consistent with strong electron-dominated response in the UV/VUV regime. The same study reports no observable “shadowing” effect in FBK angular scans, unlike HPK devices, suggesting smaller or asymmetric surface structures rather than tall occluding resistors (Croix et al., 21 Aug 2025).

The thin-film consequences of this stack were already established in a reflectance study of FBK VUV-HD1 devices and wafer mates fabricated with the same $5.96\times 5.56\ \mathrm{mm^2}$9 film and technology. That work extracted oxide thicknesses $35\times 35\ \mu\mathrm{m^2}$0 and $35\times 35\ \mu\mathrm{m^2}$1, observed strong interference oscillations in $35\times 35\ \mu\mathrm{m^2}$2–$35\times 35\ \mu\mathrm{m^2}$3 reflectance, and measured diffuse reflectance at $35\times 35\ \mu\mathrm{m^2}$4 of $35\times 35\ \mu\mathrm{m^2}$5 and $35\times 35\ \mu\mathrm{m^2}$6 for two FBK VUV-HD1 variants (Lv et al., 2019). Because VUV-HD3 is reported to share the same surface coating stack as VUVHD1, these results provide a direct optical basis for VUV-HD3 modeling, although the 2019 reflectance paper did not itself measure VUV-HD3 (Lv et al., 2019, Gallina et al., 2022). The nEXO energy-resolution study also uses a normal-incidence vacuum reflectivity at $35\times 35\ \mu\mathrm{m^2}$7 of $35\times 35\ \mu\mathrm{m^2}$8 for FBK (Gallina et al., 2022).

3. Cryogenic operating performance for nEXO

At LXe-relevant temperature, the VUV-HD3 satisfies the principal nEXO SiPM requirements while exhibiting a characteristic trade-off between high PDE and correlated noise. The nEXO study reports that the single-photoelectron gain exceeds the required $35\times 35\ \mu\mathrm{m^2}$9 from about $6\times 6\ \mathrm{mm^2}$0 over-voltage upward, that the dark count rate remains comfortably below the nEXO limit of $6\times 6\ \mathrm{mm^2}$1 across the scanned range, and that the $6\times 6\ \mathrm{mm^2}$2 PDE at $6\times 6\ \mathrm{mm^2}$3 over-voltage reaches $6\times 6\ \mathrm{mm^2}$4, well above the $6\times 6\ \mathrm{mm^2}$5 design threshold (Gallina et al., 2022).

At $6\times 6\ \mathrm{mm^2}$6 and $6\times 6\ \mathrm{mm^2}$7, the reported FBK VUV-HD3 metrics are:

Metric	Value
DCR	$6\times 6\ \mathrm{mm^2}$8
$6\times 6\ \mathrm{mm^2}$9	$80\%$0
$80\%$1	$80\%$2
$80\%$3	$80\%$4
$80\%$5	$80\%$6
$80\%$7	$80\%$8
$80\%$9	$35\times 35\ \mu\mathrm{m^2}$0
Projected $35\times 35\ \mu\mathrm{m^2}$1 at $35\times 35\ \mu\mathrm{m^2}$2	$35\times 35\ \mu\mathrm{m^2}$3

Here

$35\times 35\ \mu\mathrm{m^2}$4

and this quantity is the nEXO correlated-noise figure of merit over a $35\times 35\ \mu\mathrm{m^2}$5 window (Gallina et al., 2022). The reported value, $35\times 35\ \mu\mathrm{m^2}$6, lies at the requirement boundary of $35\times 35\ \mu\mathrm{m^2}$7 but is described as acceptable at the $35\times 35\ \mu\mathrm{m^2}$8 operating point within uncertainties (Gallina et al., 2022).

The same study shows that the main VUV-HD3 improvement over earlier FBK generations is in delayed and afterpulse components rather than in prompt optical crosstalk. Direct prompt crosstalk is not substantially reduced relative to earlier FBK devices, and at $35\times 35\ \mu\mathrm{m^2}$9 the mean additional prompt avalanches are an order of magnitude larger than in HPK VUV4 devices measured in the same campaign (Gallina et al., 2022). Conversely, FBK offers higher PDE and lower DCR than HPK at the same over-voltage, so the comparison is not monotonic in a single metric.

At detector level, these properties produce a projected nEXO energy resolution of $6\times 6\ \mathrm{mm^2}$0 at $6\times 6\ \mathrm{mm^2}$1, compared with $6\times 6\ \mathrm{mm^2}$2 for HPK VUV4 under the same model assumptions (Gallina et al., 2022). The reported optimum for FBK lies at low-to-moderate over-voltage, around $6\times 6\ \mathrm{mm^2}$3–$6\times 6\ \mathrm{mm^2}$4, because PDE saturates above about $6\times 6\ \mathrm{mm^2}$5 while CAF worsens as $6\times 6\ \mathrm{mm^2}$6 grows faster than $6\times 6\ \mathrm{mm^2}$7 (Gallina et al., 2022).

4. Wavelength dependence, angular response, and dense-media extrapolation

Beyond the single point at $6\times 6\ \mathrm{mm^2}$8, VUV-HD3 has been characterized over a much broader spectral range. Absolute PDE measurements from $6\times 6\ \mathrm{mm^2}$9 to $29.3\ \mathrm{V}$0 at $29.3\ \mathrm{V}$1 show a voltage-dependent rise, a broad maximum in the visible near $29.3\ \mathrm{V}$2–$29.3\ \mathrm{V}$3, and a decline toward the NIR; the same analytic model reproduces the curvature near $29.3\ \mathrm{V}$4–$29.3\ \mathrm{V}$5 and the NIR roll-off (Croix et al., 21 Aug 2025). Relative angular scans from $29.3\ \mathrm{V}$6 to $29.3\ \mathrm{V}$7 show strong wavelength- and angle-dependent oscillations, again characteristic of thin-film interference in the $29.3\ \mathrm{V}$8 $29.3\ \mathrm{V}$9 layer (Croix et al., 21 Aug 2025).

In the VUV, the nEXO study measured PDE versus wavelength from $163\ \mathrm{K}$00 to $163\ \mathrm{K}$01 at about $163\ \mathrm{K}$02 and $163\ \mathrm{K}$03 over-voltage and found clear interference oscillations whose maxima and minima align between TRIUMF measurements at $163\ \mathrm{K}$04 and IHEP measurements at $163\ \mathrm{K}$05 (Gallina et al., 2022). The same study notes that these oscillations are expected to be damped in liquid xenon because of refractive-index matching, consistent with the reflectance analysis of the shared oxide stack (Gallina et al., 2022).

For extrapolation to dense media, the 2025 analytic model argues that at LXe scintillation wavelength the surface transmission, rather than the internal avalanche probability, is the dominant limitation. With $163\ \mathrm{K}$06, the model gives $163\ \mathrm{K}$07 at $163\ \mathrm{K}$08 in LXe, and since absorption is extremely shallow and $163\ \mathrm{K}$09 is a good approximation in deep VUV, the predicted PDE per unit fill factor is also about $163\ \mathrm{K}$10 (Croix et al., 21 Aug 2025). The paper explicitly states that the actual PDE is $163\ \mathrm{K}$11 and does not report the fill factor for FBK in that particular model fit (Croix et al., 21 Aug 2025). For LAr wavelengths near $163\ \mathrm{K}$12–$163\ \mathrm{K}$13, the same work concludes that the thick oxide is not optimized because $163\ \mathrm{K}$14 absorption becomes the dominant loss mechanism (Croix et al., 21 Aug 2025).

A plausible implication is that VUV-HD3’s present surface design is well matched to LXe operation but not to direct LAr scintillation readout. The same modeling also indicates that reducing oxide thickness would materially improve transmission at $163\ \mathrm{K}$15, so the thick oxide should be interpreted as a device-specific compromise between VUV response, fabrication constraints, and other performance requirements rather than as a generally optimal VUV solution (Croix et al., 21 Aug 2025).

5. Secondary photon emission, spatial non-uniformity, and cross-talk statistics

Secondary photon emission from VUV-HD3 has been measured in two complementary modes: dark-noise-induced avalanches and laser-stimulated single-SPAD avalanches. In the dark-emission study, the quantity of interest is the secondary photon yield

$163\ \mathrm{K}$16

integrated over $163\ \mathrm{K}$17–$163\ \mathrm{K}$18 (McLaughlin et al., 2021). For the FBK VUV-HD3, the measured values are $163\ \mathrm{K}$19 at $163\ \mathrm{K}$20 over-voltage, $163\ \mathrm{K}$21 at $163\ \mathrm{K}$22, and $163\ \mathrm{K}$23 at $163\ \mathrm{K}$24, showing an approximately linear increase in the explored range (McLaughlin et al., 2021). The spectrum rises from a low level at $163\ \mathrm{K}$25–$163\ \mathrm{K}$26 into the red/NIR and exhibits a pronounced modulation consistent with thin-film interference in the SiO$163\ \mathrm{K}$27 coating (McLaughlin et al., 2021).

The same work used emission microscopy images to map the spatial distribution of emitted light. For FBK, the images show a small number of highly localized hotspots, randomly distributed over the SiPM area and generally confined within individual SPADs, with no pronounced corner clustering (McLaughlin et al., 2021). When normalized to the same current per unit area, the HPK device’s RMS spatial light intensity is reported to be $163\ \mathrm{K}$28 times higher than FBK’s, indicating substantially greater spatial non-uniformity in HPK (McLaughlin et al., 2021).

The laser-stimulated MIEL study isolates the emission from a repeatedly triggered central SPAD and then reconstructs the source spectrum by correcting for throughput and source-to-objective transmission. At $163\ \mathrm{K}$29 over-voltage, the FBK VUV-HD3 yields $163\ \mathrm{K}$30 photons per avalanche at the source, integrated from $163\ \mathrm{K}$31 to $163\ \mathrm{K}$32, corresponding to $163\ \mathrm{K}$33 photons per charge carrier (Raymond et al., 2024). The same study reports $163\ \mathrm{K}$34 photons per avalanche escaping into air and $163\ \mathrm{K}$35 photons per avalanche escaping into LXe when integrated over all angles, with no significant temperature dependence observed within uncertainties over $163\ \mathrm{K}$36–$163\ \mathrm{K}$37 (Raymond et al., 2024). The source-corrected spectrum peaks near $163\ \mathrm{K}$38 and retains a small shoulder near $163\ \mathrm{K}$39, while the objective spectrum carries the oxide-induced interference pattern (Raymond et al., 2024).

These emission studies bear directly on cross-talk. The external-cross-talk paper identifies FBK VUV-HD3 explicitly, models its internal OCT in LAB without reflective film with a combined Geometric+Borel distribution, and models the external OCT generated by reflective photon return with a pure Borel component (Guan et al., 2023). For VUV devices, including FBK VUV-HD3, external OCT is negligible in air without film but becomes significant in LAB with reflective film and increases with over-voltage (Guan et al., 2023). The same paper states that across devices the external component can reach up to $163\ \mathrm{K}$40 depending on applied bias voltage in reflective configurations (Guan et al., 2023).

A recurring misconception is that larger externally observed photon emission must imply larger internal prompt crosstalk. The dark-emission study shows the opposite comparison between manufacturers: HPK VUV4 has a higher external secondary photon yield in the measured $163\ \mathrm{K}$41–$163\ \mathrm{K}$42 band, yet much lower internal direct crosstalk than FBK, indicating that trench and cell design suppress internal crosstalk largely independently of externally measurable photon emission (McLaughlin et al., 2021). The laser-stimulated study reinforces this interpretation by attributing FBK’s higher internal optical crosstalk in part to dielectric-filled trenches that are ineffective at absorbing photons, unlike the tungsten-filled absorptive trenches of VUV4 (Raymond et al., 2024).

6. In-situ liquid-xenon response and detector-level implications

The most direct system-level comparison of VUV-HD3 was performed in the LoLX-2 liquid xenon detector, where $163\ \mathrm{K}$43 FBK VUV-HD3 channels and $163\ \mathrm{K}$44 HPK VUV4 channels were operated simultaneously on mixed faces of a $163\ \mathrm{K}$45 cubic active LXe volume at about $163\ \mathrm{K}$46 and $163\ \mathrm{K}$47 (Li et al., 17 Oct 2025). At $163\ \mathrm{K}$48, orthogonal-distance-regression fits to detected charge per unit sensor area give HPK/FBK ratios of $163\ \mathrm{K}$49 for $163\ \mathrm{K}$50 and $163\ \mathrm{K}$51 for $163\ \mathrm{K}$52, implying that under these in-situ conditions HPK is $163\ \mathrm{K}$53–$163\ \mathrm{K}$54 less efficient than FBK (Li et al., 17 Oct 2025). The same paper cites a normal-incidence vacuum benchmark of $163\ \mathrm{K}$55 for the HPK/FBK PDE ratio, so the in-situ difference is substantially larger than would be inferred from vacuum measurements alone (Li et al., 17 Oct 2025).

To explain that discrepancy, the LoLX-2 analysis models the SiPM response as

$163\ \mathrm{K}$56

where $163\ \mathrm{K}$57 is an empirical surface-shadowing term (Li et al., 17 Oct 2025). In that framework, FBK’s reported geometric advantages are decisive: it is windowless in this installation, whereas HPK includes a $163\ \mathrm{K}$58 quartz window, and the fill factors quoted for the comparison are approximately $163\ \mathrm{K}$59 for FBK and $163\ \mathrm{K}$60 for HPK (Li et al., 17 Oct 2025). The simulation predicts HPK/FBK ratios of $163\ \mathrm{K}$61 and $163\ \mathrm{K}$62 without shadowing, and $163\ \mathrm{K}$63 and $163\ \mathrm{K}$64 once HPK shadowing is included, in good agreement with the measured values (Li et al., 17 Oct 2025).

The same modeling also clarifies the role of the thick FBK oxide under LXe incidence distributions. At $163\ \mathrm{K}$65, the thicker FBK $163\ \mathrm{K}$66 layer shifts interference so that transmission approaches zero near $163\ \mathrm{K}$67 incidence, even though near-normal transmission is similar to HPK before shadowing is applied (Li et al., 17 Oct 2025). This means that the VUV-HD3 does not simply “win” because its oxide is optically benign; rather, its higher fill factor, windowless configuration, and absence of HPK-like high-angle surface shadowing dominate once the full angular distribution of scintillation photons is included (Li et al., 17 Oct 2025).

The detector-level lesson is that the relevant figure of merit is the angle-averaged in-situ PDE, not the vacuum PDE at normal incidence. This conclusion is reinforced by the secondary-emission studies: external crosstalk depends on red/NIR photon emission, geometry, reflections, and neighboring-device PDE in that band, while VUV scintillation detection depends on a separate thin-film transmission problem. For tightly packed arrays, FBK’s lower dark-avalanche SPY in $163\ \mathrm{K}$68–$163\ \mathrm{K}$69, lower hotspot density, and more uniform emission are favorable for suppressing external crosstalk systematics (McLaughlin et al., 2021). At the same time, its pronounced thin-film interference makes the VUV-HD3 more wavelength- and angle-structured than devices with thinner surface layers, so optical transport simulations must retain that structure rather than reducing the device to a single scalar PDE (Li et al., 17 Oct 2025).