Hamamatsu VUV4 SiPMs in Cryogenic Applications

• Hamamatsu VUV4 is a family of vacuum-ultraviolet-sensitive silicon photomultipliers engineered for direct detection of noble-gas scintillation in liquid xenon (~175 nm) and liquid argon (~127–128 nm).
• These devices demonstrate stable cryogenic electrical behavior with reproducible breakdown voltages, low dark noise, and effective bias tracking critical for rare-event detector applications.
• Photon detection performance varies with model, package design, and measurement conditions, making precise evaluation of PDE, surface optics, and angular response essential for system-level integration.

Hamamatsu VUV4 is the fourth-generation family of Hamamatsu vacuum-ultraviolet-sensitive Multi-Pixel Photon Counters (MPPCs), i.e. silicon photomultipliers engineered for direct detection of noble-gas scintillation, especially liquid xenon near $175$ nm and liquid argon near $127\text{--}128$ nm. In the literature, “VUV4” denotes a family rather than a single part number: representative devices include the S13370-6050CN, S13370-6075CN, S13371-6050CQ, and S13370-6152, which were studied for nEXO, liquid-argon detectors, OLAF, SBC-LAr10, and related cryogenic systems (Lv et al., 2019, Gallina et al., 2022).

1. Family definition and nomenclature

The VUV4 designation is explicitly used for Hamamatsu’s fourth-generation VUV-sensitive SiPMs, but the published literature attaches that family name to several distinct devices and package formats. This is a recurring source of ambiguity. In particular, the same family label covers single-die devices, quad devices, and studies that emphasize different package or window configurations rather than a single canonical sensor (Lv et al., 2019, Pershing et al., 2022).

Device or designation	Explicitly stated characteristics	Representative context
S13370-6050CN	$6\times 6~\mathrm{mm^2}$, $50~\mu\mathrm{m}$, 60% fill factor	LAr detector characterization; vacuum reflectance; nEXO VUV4-50
S13370-6075CN	$6\times 6~\mathrm{mm^2}$, $75~\mu\mathrm{m}$, 70% fill factor, windowless	Cryogenic PDE study; OLAF reference form
S13371-6050CQ(-02)	Quad package with four cells; $50~\mu\mathrm{m}$	nEXO VUV4-Q-50; direct-LAr 128 nm study
S13370-6152	$6\times 6~\mathrm{mm^2}$, $50~\mu\mathrm{m}$	2019 nEXO characterization

The literature also distinguishes between family-level references and explicitly identified installed hardware. For example, the OLAF facility explicitly discusses “VUV4-series windowless Silicon Photomultipliers” and cites the Hamamatsu product form S13370-6075CN, but it does not explicitly state that the installed OLAF sensor is S13370-6075CN (Shi et al., 15 Jan 2026). A closely related point is that the “-Q-” device studied by nEXO is a quad sensor on one ceramic package rather than a separate VUV generation (Gallina et al., 2022).

This family-level usage matters because comparisons across papers are often comparisons across different VUV4 subtypes, different batches, or different package realizations rather than identical sensors. A plausible implication is that “Hamamatsu VUV4” should be treated as a platform name whose reported performance depends materially on model, surface structure, window configuration, and measurement convention.

2. Developmental lineage and pre-VUV4 background

The VUV4 family emerged from a broader Hamamatsu program on short-wavelength-sensitive MPPCs. Earlier work on UV-enhanced Hamamatsu MPPC prototypes already established two themes that remain central in VUV4 studies: the importance of windowless or reduced-window designs, and the tension between direct short-wavelength sensitivity and environmental robustness. In the 2012 NEXT-related study, Hamamatsu provided UV-enhanced MPPC prototypes on the S10362-33-050C platform; the windowless UV-enhanced device reached a maximum PDE of $(43.9\pm1.4)\%$ at $127\text{--}128$0 nm and $127\text{--}128$1 at $127\text{--}128$2 nm, but that paper neither used the term “VUV4” nor measured below $127\text{--}128$3 nm, so it is indirect precursor evidence rather than direct VUV4 characterization (Yahlali et al., 2012).

The next step was the pre-VUV4 VUV2/VUV3 era. Hamamatsu’s early VUV-sensitive MPPCs labeled VUV2 and VUV3 demonstrated direct liquid-argon scintillation sensitivity and cryogenic operation, thereby establishing the basic feasibility of direct $127\text{--}128$4 nm detection in silicon without a wavelength shifter. Those devices were explicitly not VUV4, but they are part of the technological lineage: the paper presents them as VUV-sensitive MPPC prototypes with cryogenic gain, breakdown-voltage, dark-count, and direct-LAr measurements (Igarashi et al., 2015).

A further transitional result appears in the 2017 cryogenic readout study of the S13370-3050CN, identified there as “VUV4 generation.” That work was not primarily a PDE paper; its main contribution was to show that an array of 16 VUV4 sensors could be read out as a single detector at liquid-xenon temperature while preserving single-photoelectron separation, using a low-power AD8011-based summing front-end designed for matrices up to $127\text{--}128$5 sensors (Arneodo et al., 2017). This suggests that by the time of the commercial VUV4 studies, Hamamatsu’s VUV-sensitive MPPC program had already become tightly coupled to system-level readout design for large LXe and LAr instruments.

3. Cryogenic electrical behavior and noise

Across the VUV4 literature, the most stable electrical result is the temperature dependence of breakdown voltage. Multiple studies converge on a coefficient near $127\text{--}128$6. For nEXO’s VUV4-50 and VUV4-Q-50 at $127\text{--}128$7 K, the measured breakdown voltages are $127\text{--}128$8 V and $127\text{--}128$9 V, with $6\times 6~\mathrm{mm^2}$0 and $6\times 6~\mathrm{mm^2}$1, respectively (Gallina et al., 2022). In batch characterization of 32 VUV4 Quads for SBC-LAr10, the temperature dependence of $6\times 6~\mathrm{mm^2}$2 was found to be $6\times 6~\mathrm{mm^2}$3 (Hawley-Herrera et al., 2024). In a liquid-argon-detector study of the S13370-6050CN, the breakdown voltage decreases linearly with temperature above $6\times 6~\mathrm{mm^2}$4 K with slope about $6\times 6~\mathrm{mm^2}$5, while below $6\times 6~\mathrm{mm^2}$6 K the decrease becomes slower (Wang et al., 2021).

Cryogenic dark noise is typically very low. At $6\times 6~\mathrm{mm^2}$7 K and $6\times 6~\mathrm{mm^2}$8, the S13370-6050CN shows a dark count rate of about $6\times 6~\mathrm{mm^2}$9, and at $50~\mu\mathrm{m}$0 K and $50~\mu\mathrm{m}$1 the total correlated signal probability is approximately $50~\mu\mathrm{m}$2, or “less than 10%” in the paper’s summary (Wang et al., 2021). For nEXO’s newer VUV4 samples at $50~\mu\mathrm{m}$3 K and $50~\mu\mathrm{m}$4 V overvoltage, the reported dark count rate is $50~\mu\mathrm{m}$5, with mean extra correlated-avalanche charge $50~\mu\mathrm{m}$6, RMS fluctuation $50~\mu\mathrm{m}$7, and correlated avalanche fluctuation $50~\mu\mathrm{m}$8 (Gallina et al., 2022). Earlier nEXO measurements on the S13370-6152 at $50~\mu\mathrm{m}$9 K and $6\times 6~\mathrm{mm^2}$0 V overvoltage already showed strong cryogenic noise performance, with dark-noise rate $6\times 6~\mathrm{mm^2}$1 and number of correlated avalanches $6\times 6~\mathrm{mm^2}$2 in the $6\times 6~\mathrm{mm^2}$3 post-trigger window (Gallina et al., 2019).

Batch studies add the manufacturing and operating-point perspective. In SBC-LAr10’s 32-Quad campaign, the average gain slope was $6\times 6~\mathrm{mm^2}$4; the average DCR temperature coefficient was $6\times 6~\mathrm{mm^2}$5, corresponding to a reduction factor of 7 for every $6\times 6~\mathrm{mm^2}$6 K drop in temperature; and the average temperature dependence of $6\times 6~\mathrm{mm^2}$7 was $6\times 6~\mathrm{mm^2}$8 (Hawley-Herrera et al., 2024). The paper further argues that $6\times 6~\mathrm{mm^2}$9 estimated from the average across all SiPMs is a better estimator than $75~\mu\mathrm{m}$0 from individual SiPMs, whereas the opposite is true for $75~\mu\mathrm{m}$1, gain, $75~\mu\mathrm{m}$2, and DCR. This suggests that device-to-device spread is straightforward to resolve for breakdown and gain, but more analysis-limited for correlated-avalanche probability.

Taken together, these measurements establish the VUV4 family as cryogenically stable in the electrical sense: breakdown voltage is strongly temperature dependent but reproducible, dark noise is low enough for rare-event detectors, and correlated noise can remain within stringent specifications at moderate overvoltage. The practical consequence is that bias tracking with temperature is not optional; it is the central control variable for stable cryogenic operation.

4. VUV photon-detection performance in liquid xenon and liquid argon

Published VUV4 PDE values are not interchangeable, because they were measured at different wavelengths, temperatures, media, angles, and with different definitions of what counts as detected light. This is the central interpretive issue in the VUV4 literature.

For liquid xenon applications, the earliest dedicated nEXO study of the S13370-6152 measured PDE at $75~\mu\mathrm{m}$3 nm and $75~\mu\mathrm{m}$4 K, not at $75~\mu\mathrm{m}$5 nm in LXe. The reported values were $75~\mu\mathrm{m}$6 at $75~\mu\mathrm{m}$7 V and $75~\mu\mathrm{m}$8 at $75~\mu\mathrm{m}$9 V, corresponding to saturation PDEs of $50~\mu\mathrm{m}$0 and $50~\mu\mathrm{m}$1, both well below the $50~\mu\mathrm{m}$2 saturation PDE advertised by Hamamatsu (Gallina et al., 2019). That study concluded that VUV4 noise performance was excellent but PDE was marginal or inadequate for nEXO in the tested samples.

Newer nEXO measurements on new Hamamatsu samples changed that picture materially. For HPK VUV4 MPPCs at $50~\mu\mathrm{m}$3 K and $50~\mu\mathrm{m}$4 V overvoltage, the average PDE at $50~\mu\mathrm{m}$5 nm is reported as $50~\mu\mathrm{m}$6, with projected nEXO energy resolution $50~\mu\mathrm{m}$7, comfortably within the experiment’s requirement (Gallina et al., 2022). The same paper concludes that Hamamatsu VUV4 and FBK VUVHD3 are both viable for nEXO, with different tradeoffs: Hamamatsu is lower in PDE than FBK, but better in correlated-noise behavior.

For liquid argon, direct $50~\mu\mathrm{m}$8 nm performance has been measured more than once. In a direct-LAr study at $50~\mu\mathrm{m}$9 K and $6\times 6~\mathrm{mm^2}$0 bar, the S13370-6075CN yielded a photon detection efficiency of $6\times 6~\mathrm{mm^2}$1 at $6\times 6~\mathrm{mm^2}$2 nm for operation at $6\times 6~\mathrm{mm^2}$3 V of overvoltage, while the S13371-6050CQ-02 with its quartz window removed yielded $6\times 6~\mathrm{mm^2}$4 (Pershing et al., 2022). A later cryogenic PDE study of the S13370-6075CN measured $6\times 6~\mathrm{mm^2}$5 and showed that the PDE decreases substantially at cryogenic temperature across $6\times 6~\mathrm{mm^2}$6 nm, with the strongest degradation below $6\times 6~\mathrm{mm^2}$7 nm (Álvarez-Garrote et al., 2024). That paper explicitly notes that its $6\times 6~\mathrm{mm^2}$8 nm result is compatible with the earlier $6\times 6~\mathrm{mm^2}$9 measurement once one accounts for different definitions, including whether reflected photons were subtracted.

The broader spectral study is important because it shows that VUV4 is not simply a “127 nm detector.” At $50~\mu\mathrm{m}$0 K, the S13370-6075CN still had $50~\mu\mathrm{m}$1 PDE at $50~\mu\mathrm{m}$2 nm and $50~\mu\mathrm{m}$3 at $50~\mu\mathrm{m}$4 nm, but only $50~\mu\mathrm{m}$5 at $50~\mu\mathrm{m}$6 nm and $50~\mu\mathrm{m}$7 at $50~\mu\mathrm{m}$8 nm (Álvarez-Garrote et al., 2024). This confirms that cryogenic VUV4 operation retains broad spectral sensitivity while penalizing the short-wavelength end most strongly.

A common misunderstanding is to treat the 2021 liquid-argon-detector study of S13370-6050CN as an absolute PDE result at LAr scintillation wavelength. It is not. That paper measures relative quantum efficiency at $50~\mu\mathrm{m}$9 nm, normalized to room temperature, and explicitly does not provide an absolute $(43.9\pm1.4)\%$0 nm PDE (Wang et al., 2021). The distinction is essential because relative blue-light response in a TPB-like readout chain is not the same quantity as direct VUV PDE.

5. Surface optics: reflectivity, angular response, and self-emission

VUV4 devices are not only photodetectors; they are also optical surfaces. Their reflectivity is substantial, angle dependent, and in LXe can materially modify detector light transport.

In liquid xenon, three S13370-series VUV4 samples with $(43.9\pm1.4)\%$1 cells were measured to have specular reflectivities at $(43.9\pm1.4)\%$2 incidence of $(43.9\pm1.4)\%$3, $(43.9\pm1.4)\%$4, and $(43.9\pm1.4)\%$5 (Nakarmi et al., 2019). The same study found device-to-device relative PDE differences of about $(43.9\pm1.4)\%$6 at normal incidence, and for one sample showed that at $(43.9\pm1.4)\%$7 the PDE had fallen to roughly 60% of its normal-incidence value. Both PDE and specular reflectivity decreased with angle in LXe (Nakarmi et al., 2019).

A second LXe reflectivity study reinforced the point that VUV4 behaves differently from FBK VUV-HD and smooth silicon surfaces. For a Hamamatsu MPPC S13370 VUV4 sample, the main specular reflectivity decreases with angle of incidence, whereas the other measured samples increased with angle (Wagenpfeil et al., 2021). The same paper attributes two distinct secondary peaks in the reflection scans to bias lines inclined to the microcell surface with a predominant slope of $(43.9\pm1.4)\%$8, and states that the total reflectivity of the VUV4 is up to $(43.9\pm1.4)\%$9 relative larger than the pure specular reflectivity (Wagenpfeil et al., 2021). This is a particularly important result for optical Monte Carlo: a one-lobe specular model is incomplete.

Vacuum reflectance measurements add wavelength and microstructure information. Two explicitly identified Hamamatsu-VUV4 devices, S13370-6050CN and S13370-6075CN, show decreasing specular reflectance with increasing angle of incidence and no oscillatory interference structure, unlike FBK devices. At $127\text{--}128$00 nm, their diffuse reflectances are $127\text{--}128$01 for the $127\text{--}128$02 device and $127\text{--}128$03 for the $127\text{--}128$04 device (Lv et al., 2019). The larger-pixel device is more specular and less diffuse, which the paper links to larger fill factor and less exposed microstructure.

The VUV4 literature also contains two complementary views of avalanche-induced self-emission. In dark-condition spectroscopy, the HPK VUV4 MPPC emits $127\text{--}128$05 over $127\text{--}128$06 nm at $127\text{--}128$07 V overvoltage, approximately a factor of two above the FBK VUV-HD3 in the same measured band, with no interference pattern and with more hotspots, especially in one corner of the device (McLaughlin et al., 2021). In laser-stimulated measurements, the VUV4 yields $127\text{--}128$08 photons per avalanche into the microscope objective at $127\text{--}128$09 K and a modeled $127\text{--}128$10 photons emitted per avalanche in LXe at $127\text{--}128$11 V overvoltage, with no significant temperature dependence observed within uncertainties (Raymond et al., 2024). The latter work interprets VUV4’s lower internal crosstalk relative to FBK as a structural effect, likely linked to absorbing tungsten trenches, not simply lower source-photon production.

These optical results collectively imply that VUV4 performance cannot be reduced to a scalar PDE alone. Reflectivity, off-specular structure, diffuse scattering, and secondary photon emission all contribute to the effective behavior of the device in a tiled cryogenic detector.

6. Detector integration, comparative performance, and interpretive caveats

The family has now been used or proposed in several detector contexts. In OLAF, Hamamatsu vacuum-ultraviolet SiPMs are immersed directly in liquid argon to eliminate wavelength-shifter and light-guide systematics; the paper explicitly discusses VUV4-series windowless SiPMs and cites the S13370-6075CN form, but stops short of identifying the exact installed sensor by full model number (Shi et al., 15 Jan 2026). In SBC-LAr10, 32 VUV4 Quads, totaling 128 individual SiPM channels, were characterized for use as a scintillation veto in a xenon-doped liquid-argon bubble chamber (Hawley-Herrera et al., 2024). In earlier LXe readout work, arrays of S13370-3050CN VUV4-generation MPPCs were summed at $127\text{--}128$12 K while retaining single-photoelectron sensitivity, with electronics designed for matrices up to $127\text{--}128$13 sensors (Arneodo et al., 2017).

The strongest caution against overgeneralization comes from in-situ comparison studies. In the LoLX liquid-xenon detector at $127\text{--}128$14 K and $127\text{--}128$15, packaged HPK VUV4 SiPMs produced only $127\text{--}128$16 as much scintillation signal per nominal unit area as FBK VUV-HD3 for $127\text{--}128$17, and $127\text{--}128$18 for $127\text{--}128$19, i.e. HPK was 33–38% less efficient under those operating conditions (Li et al., 17 Oct 2025). That discrepancy relative to simpler PDE expectations was resolved in simulation by an angular- and wavelength-dependent model incorporating surface shadowing and the lower HPK fill factor $127\text{--}128$20 versus $127\text{--}128$21 for FBK (Li et al., 17 Oct 2025). This result does not negate the benchmark PDE measurements; it shows that normal-incidence or bench-top PDE does not automatically predict in-situ LXe response.

Several recurrent misconceptions follow directly from the literature. First, VUV4 is not one part number. Second, not every paper that measures a Hamamatsu VUV-sensitive SiPM is measuring VUV4; the 2012 NEXT UV-enhanced MPPC work is indirect precursor data and explicitly does not establish behavior at $127\text{--}128$22 or $127\text{--}128$23 nm (Yahlali et al., 2012). Third, not every “PDE” is defined identically: some studies report relative PDE, some absolute PDE, some photocurrent-based effective PDE, some zero-counting PDE, and some explicitly include or do not subtract reflected photons (Gallina et al., 2019, Álvarez-Garrote et al., 2024). Fourth, room-temperature or normal-incidence values should not be transplanted uncritically into LXe or LAr detector models, because VUV4 response depends on wavelength, temperature, angle of incidence, package geometry, and optical environment (Nakarmi et al., 2019, Li et al., 17 Oct 2025).

The cumulative picture is therefore specific rather than generic. Hamamatsu VUV4 is a mature VUV-sensitive SiPM family with demonstrable direct sensitivity to LXe and LAr scintillation, low cryogenic dark noise, and strong relevance for large rare-event detectors. At the same time, its reported performance is irreducibly conditional: family member, batch, medium, and measurement convention all matter. The literature supports treating VUV4 not as a single number on a datasheet, but as a class of cryogenic VUV photodetectors whose device physics and system behavior must be modeled together.