Scintillator-Integrated MCP-PMTs (SCI-IMPs)

• SCI-IMPs are integrated scintillator–MCP photomultiplier tubes that use the scintillator as the vacuum window, reducing optical interfaces and enhancing prompt photon detection.
• They eliminate multiple coupling layers by depositing the photocathode directly on the scintillator, which minimizes reflection losses and improves sensitivity to deep-UV and Cherenkov photons.
• Prototype implementations with BaF₂ and BGO demonstrate ultrafast coincidence timing (~46–55 ps at 511 keV and down to 16.8 ps with cosmic muons) while retaining energy measurement capability.

Searching arXiv for the cited SCI-IMP and related MCP-PMT scintillator papers. Scintillator-integrated microchannel plate photomultiplier tubes (SCI-IMPs) are vacuum photodetectors in which the scintillator itself serves as the window faceplate (WFP) of the MCP-PMT, with the photocathode deposited directly on the vacuum side of that scintillator faceplate. In this geometry, ionizing radiation deposits energy directly in the scintillator WFP, the scintillator emits optical photons, the photocathode converts some of these photons to photoelectrons, and the MCP stack amplifies the resulting signal at ultrafast timescales. The defining objective is to combine ultrafast timing, ideally in the few-tens-of-picoseconds regime, with energy measurement over a wide range from 511 keV annihilation photons to GeV-scale cosmic-ray muons (Ota et al., 3 Oct 2025).

1. Conceptual definition and detector class

SCI-IMPs differ from both a conventional MCP-PMT and a standard scintillator-plus-photodetector assembly. In a conventional MCP-PMT, the window is a transparent optical window and radiation sensitivity normally comes from photons generated elsewhere. In a standard scintillator-plus-photodetector assembly, scintillation photons must cross one or more optical boundaries before reaching the photocathode or silicon sensor. By contrast, an SCI-IMP removes external optical coupling layers because the scintillator is built directly into the tube as the entrance window (Ota et al., 3 Oct 2025).

The immediate consequence is architectural rather than merely incremental. Removal of optical interfaces reduces reflection losses, which are especially severe in the ultraviolet, and makes direct use of deep-UV scintillation or prompt Cherenkov-related photons more practical. This is central to the two material realizations reported so far: BaF$_2$-based and BGO-based SCI-IMPs. The detector class was introduced to address a longstanding tradeoff between high-precision timing and energy measurement, a tradeoff that earlier Cherenkov-radiator-integrated MCP-PMTs had approached from the timing side only. Previous integrated-radiator work by the same group had shown about 30 ps FWHM CTR for 511 keV annihilation photons, but that detector emphasized Cherenkov detection and effectively sacrificed energy information (Ota et al., 3 Oct 2025).

SCI-IMPs therefore occupy a specific position within fast radiation instrumentation: they are not merely MCP-PMTs used near scintillators, but directly integrated scintillator–MCP photodetectors intended to retain calorimetric information while operating in a timing regime usually associated with prompt-photon devices.

2. Device architecture and signal formation

Two SCI-IMP prototypes were fabricated: BaF$_2$-IMP and BGO-IMP. In both, the standard window faceplate of the MCP-PMT was replaced by a scintillator disk of approximately 22 mm diameter $\times$ 3.2 mm thickness, while the photocathode was deposited only over the central 11 mm diameter region. The devices function effectively as single-channel MCP-PMTs (Ota et al., 3 Oct 2025).

The fabrication sequence combines scintillator integration with photocathode protection. A thin Al$_2$O$_3$ intermediate layer of a few nanometers is inserted between the integrated material and the photocathode to avoid harmful chemical interaction during deposition. The scintillator plate then serves as the front vacuum boundary, the photocathode is formed on its vacuum side, and the MCP stack and anode are completed within the tube. The divider configuration was optimized following prior work; the bias voltages were set approximately to about 200 V between the photocathode and MCP input and about 1100 V between MCP output and anode (Ota et al., 3 Oct 2025).

The signal chain is direct:

1. Ionizing radiation deposits energy in the scintillator WFP.
2. The scintillator emits optical photons.
3. Photons propagate directly to the photocathode without external coupling layers.
4. The photocathode generates photoelectrons.
5. The MCP multiplies the charge.
6. The anode collects a fast electrical pulse.

This geometry is particularly consequential for materials whose useful prompt photons lie in the UV. A plausible implication is that the SCI-IMP concept is best understood as an optical-boundary-minimization strategy implemented inside the vacuum detector itself, rather than as a mere packaging variation.

3. Scintillator selection, photocathode matching, and material-specific behavior

The two demonstrated implementations use barium fluoride (BaF$_2$) and bismuth germanate (BGO), with different photocathodes selected to match their emission properties (Ota et al., 3 Oct 2025).

Material	Reported properties	Photocathode
BaF$_2$	Density 4.89 g/cm$^3$; light yield 8.5 photons/keV; fast component near 220 nm; slow component near 300 nm	CsTe
BGO	Density 7.13 g/cm$^3$; light yield 10.7 photons/keV; broad emission with cutoff around 300 nm	multialkali

For BaF$_2$, the attraction is its extremely fast UV emission. The reported decay constants are 0.078 ns with 0.99%, 0.747 ns with 5.35%, and 689 ns with 93.66%. The fast component is favorable for timing, but the slow component has higher light yield and degrades timing while causing pile-up and baseline instability. The CsTe photocathode was therefore chosen because it is sensitive mainly below 300 nm, helping preferentially detect the fast UV emission and suppress the problematic slow component. The reported quantum efficiency was around ~17% QE at 220 nm in the text, while Table I lists max QE = 19.6% at 240 nm (Ota et al., 3 Oct 2025).

For BGO, the relevant point is not fast scintillation in the usual sense, but hybrid operation through both scintillation and Cherenkov-related prompt photons. Table I gives decay constants of 45.8 ns with 8.2% and 365 ns with 91.8%, while the Appendix reports time-correlated single-photon-counting values of $_2$0, $_2$1, $_2$2, and fast component fraction 6.62%. The multialkali photocathode was chosen to provide broad sensitivity to both Cherenkov and scintillation photons, with Table I giving max QE = 20.9% at 400 nm (Ota et al., 3 Oct 2025).

These choices produce different timing regimes. In BaF$_2$3, timing follows the usual scintillation-statistics argument: the paper explicitly notes that CTR is inversely proportional to the square root of light yield or deposited energy. In BGO, the decisive variable is whether the first detected photons are prompt Cherenkov photons or delayed scintillation photons. This suggests that material selection in SCI-IMPs is not reducible to decay time alone; prompt-photon transport, photocathode spectral selectivity, and first-photon statistics are co-equal determinants of timestamp formation.

4. Experimental methods and reported performance

The 511 keV measurements used a $_2$4Na point source of approximately 1.18 MBq in coincidence geometry, with tungsten collimators of aperture $_2$5 mm$_2$6 and thickness 50 mm in front of each detector. Waveforms were acquired with a Tektronix DPO71254C oscilloscope at 100 GS/s; for both detector types, 100,000 data points per waveform corresponding to 1000 ns were recorded. The BaF$_2$7-IMP pair used 5 GHz oscilloscope bandwidth and one layer of PTFE tape on the entrance surface to reflect more photons back toward the photocathode. The BGO-IMP pair used 3 GHz bandwidth and black tape on the entrance surface to suppress temporal fluctuations caused by reflected Cherenkov photons in the BGO plate (Ota et al., 3 Oct 2025).

Energy was reconstructed from waveform charge integration: 500 ns from the trigger time for BaF$_2$8-IMP and 1000 ns for BGO-IMP. The photopeak and Compton contamination were modeled with Gaussian + exponential, and energy resolution was defined as the FWHM of the Gaussian photopeak component using

$_2$9

Timing pickoff used ROOT TSpline3 interpolation followed by a CFD-like threshold scan from 0.2% to 12.8% in 0.2% increments relative to each pulse height. CTR was defined as the FWHM of the time-difference distribution between the two detectors (Ota et al., 3 Oct 2025).

The main reported results are summarized below.

Detector configuration	Energy resolution at 511 keV	CTR
BaF$\times$0-IMP pair	35.1 ± 0.6%, 35.2 ± 0.5%	52.3 ± 1.2 ps, 54.8 ± 1.3 ps, 51.5 ± 1.4 ps FWHM
BGO-IMP pair	37.5 ± 0.2% for BGO-IMP B; 51.2 ± 0.3% for BGO-IMP A	46.4 ± 5.5 ps, 48.3 ± 5.1 ps, 50.1 ± 4.2 ps FWHM
BaF$\times$1-IMP pair, cosmic muons	keV-equivalent threshold >1500 keV	25.1 ± 2.8 ps FWHM
BaF$\times$2-IMP + BGO-IMP, cosmic muons	keV-equivalent threshold >1500 keV	16.8 ± 4.0 ps FWHM

For cosmic-ray muons, which typically have energies above 1 GeV, the estimated energy loss in a 3.2 mm crystal was approximately 3–4 MeV for both BaF$\times$3 and BGO. Because there was only one detector pair and no external tracking telescope, an energy threshold of >1500 keV equivalent was applied to bias toward central traversals and larger deposited energy (Ota et al., 3 Oct 2025).

The BGO timing distribution required special treatment. At the 0 mm source position, the ratio of Cherenkov-photon-triggered coincidence events was 8.5 ± 2.1%, and the FWHM of the Cherenkov-only sharp peak was 38.2 ps. For BGO-IMP B, the average detected photons in the 511 keV experiment were 0.862 Cherenkov and 134 scintillation, while in the cosmic-ray muon experiment they were 10.5 Cherenkov and 762 scintillation. Applying a photon-density threshold defined as the charge detected within the first 200 ps of the waveform improved fast-event classification: at 0.085 pC, the fraction of fast events increased, and at 0.095 pC, the CTR improved further to 28.4 ps FWHM (Ota et al., 3 Oct 2025).

These results establish two distinct timing regimes. At 511 keV, both detector types achieve approximately 50 ps FWHM pairwise coincidence timing while retaining energy measurement. At GeV-scale muon energies, prompt-photon statistics improve sufficiently that the same integration strategy reaches 25.1 ps and 16.8 ps FWHM coincidence timing.

5. Relation to adjacent MCP–scintillator architectures

SCI-IMPs are part of a broader family of scintillator readout systems involving MCP-based photodetectors, but not all such systems are direct integrations. A closely related precursor is the time and position sensitive single photon detector for scintillator read-out, which does not present a monolithic scintillator-integrated MCP-PMT in the strict SCI-IMP sense. Instead, it demonstrates a closely related scintillator-to-MCP-PMT optical readout architecture for high-rate neutron/$\times$4 radiography. There, a scintillation flash is split into two optical branches: a standard PMT branch for fast timing and pulse-height measurement, and an RS-PMT imaging branch in which a triple MCP stack terminates on a resistive screen and the induced image charge is capacitively coupled through the ceramic wall to an external delay-line anode. The system showed $\times$5 mm FWHM or better spatial resolution, gain up to $\times$6 at rates up to 1 MHz, gain about $\times$7 at 10 MHz, and a novel pulse-height-to-time converter with 20–30 ns dead time and pulse-height resolution well below 5% over 10% to 90% of full scale (Schössler et al., 2012).

That architecture matches several functional attributes later associated with SCI-IMPs—scintillator flash readout by an MCP-based photomultiplier, single-photon/counting-style operation, ultrafast timing, position-sensitive anode, and high-rate imaging—but differs in the decisive respect that the scintillator is not directly integrated onto the MCP-PMT entrance window. Light is transported by mirror + lens to the RS-PMT, and energy information comes from a separate side-coupled standard PMT rather than from the MCP imager alone. This suggests that SCI-IMPs can be viewed as a further step in integration: they internalize within the detector package what earlier systems achieved through distributed optics and coincidence electronics (Schössler et al., 2012).

A second adjacent example is JUNO-OSIRIS, which is also not a true SCI-IMP. It is a liquid scintillator detector instrumented with externally mounted large-area MCP-PMTs: 64 20-inch MCP-PMTs around an 18-ton liquid scintillator target in a 3 m diameter acrylic vessel, plus 12 additional 20-inch MCP-PMTs for the water Cherenkov muon veto. OSIRIS was designed to reach a sensitivity of $\times$8 g/g for U/Th and, during water-filling commissioning, operated with a trigger threshold of 11 PMTs fired in 64 ns, producing 7–10 Hz event rates; muon-like events with well over 100 p.e. total appeared at a rate of about 0.3 Hz (Rodphai et al., 2024).

OSIRIS is relevant at the system level rather than at the device-physics level. It shows how large-area MCP-PMTs are integrated around a scintillator target, including cones offering magnetic shielding, a front cylinder reflector, common JUNO 1F3 electronics, waveform acquisition, and a triggerless stream containing only Start time and Charge. The distinction is again essential: SCI-IMPs are integrated scintillator–MCP devices, whereas OSIRIS is a scintillator detector instrumented with MCP-PMTs (Rodphai et al., 2024).

6. Applications, limitations, and future development

The principal application domains identified so far are high-energy physics, radiation detection, and time-of-flight PET. For HEP timing layers and particle-identification systems, the reported 25.1 ps FWHM for a BaF$\times$9-IMP pair and 16.8 ps FWHM for a BaF$_2$0-IMP + BGO-IMP pair place SCI-IMPs in the regime of a few tens of picoseconds. For TOF-PET, the relevant result is that the detectors preserve energy measurement while achieving roughly 46–55 ps FWHM coincidence timing at 511 keV. The paper also links the concept to direct positron emission imaging (dPEI), where ultrafast timing may enable reconstruction-free or reduced-reconstruction paradigms (Ota et al., 3 Oct 2025).

The current devices also exhibit clear limitations. Although the scintillator faceplate is 22 mm diameter, the photocathode covers only 11 mm diameter, and for BaF$_2$1 the estimated light collection efficiency from the crystal center is only ~35%. The present single-channel format has substantial dead area around the active region, reducing packing fraction for both HEP tiling and TOF-PET arrays. Fabrication variability is visible in the significantly different QE curves and energy resolutions of the two BGO-IMPs. In BGO at 511 keV, most events are not Cherenkov-triggered, and the long timing tails remain a challenge; the exact reason why four exponentials, rather than two, were needed in the phenomenological fit remains under investigation. In muon measurements, saturated waveforms occurred for events crossing the scintillator center, producing bumps on the left side of the timing histograms (Ota et al., 3 Oct 2025).

The proposed development path is correspondingly explicit. The clearest hardware direction is toward two-dimensional, position-sensitive, multi-anode SCI-IMPs, with expected benefits of larger effective photosensitive area, denser packing, more compact arrays, improved spatial resolution, and better light collection and timing. Photocathode optimization could improve QE by 30% or more. For BGO, better Cherenkov-event extraction is a specific priority; prompt-charge thresholding already increased the fast-event ratio from 8.5% to 20% and improved CTR to 28.4 ps FWHM. Future material directions mentioned include photonic-crystal scintillators, composite scintillators, and perovskite scintillators (Ota et al., 3 Oct 2025).

Taken together, these results define SCI-IMPs as a distinct detector class rather than a narrow implementation detail. Their key contribution is the direct integration of scintillator and MCP-PMT into a single vacuum sensor that preserves energy measurement while exploiting prompt-photon transport and MCP-class timing. A plausible implication is that the long-term significance of SCI-IMPs will depend less on proving the basic concept—which has already been demonstrated—and more on solving photosensitive-area scaling, fabrication uniformity, and prompt-photon classification in multi-channel, array-compatible formats.