Technical field note · November 11, 2026 · Instrument evaluation

Measuring light
when it seems to have vanished.

A real-world test of otSPAD: phosphorescent paint, a cooled SiPM, and more than nine hours of experimental observation, ending with a direct sample-removal background check. The objective is instrument performance, not a chemical study of the pigment.

Experimental prototype · Officina TuriniGate: 100 msSiPM onsemi MICROFC-30050Primary data: 3 CSV acquisitions
Verified samples
111,414
100 ms gates; three CSV acquisition files
Observation span
9 h 11 min
From first to last recorded point, including breaks
Reference dark count
11,800cps
SiPM at approximately −20 °C
Sample-removal control
767 cps
Measured difference, last 5 min of each phase
01 / The test

A small experiment with a precise instrumental question

How long can a laboratory-built, SiPM-based photon detector recognize the residual light from a very weak source as its flux decreases by orders of magnitude?

The source was a small spot of phosphorescent paint,strongly excited by light before the measurement beganand then kept in darkness. In this experiment,no radioactive source was used. The sample was small enough to fit within the calculated optical field of view, but alignment was performed by eye: no optical bench or micrometric positioning stages were available.

Measurement chain

SiPM → comparator → FPGA → DSP → PC

onsemi MICROFC-30050-SMT-TR SiPM, 3 × 3 mm active area; pulse detection, controlled electronic threshold, FPGA digital counting, BF518 DSP supervision, and PC acquisition with otSPAD. The hardware counter supports event rates up to 150 MHz; the rates observed here are far lower.

Operating conditions

Controlled temperature, improvised optical geometry

Cold plate near −20 °C, set overvoltage of 2.5 V, and comparator threshold of 2.300 V. The lens has a 55 mm focal length and approximately 31.75 mm diameter; lens-to-sample distance is estimated at 30 cm. The recorded temperature refers to thecold plate, not the SiPM junction.

100 ms / gate50 MHz + PLL / FPGA16-bit / DAC adjustmentsCooled SiPMRelative measurement
02 / Measurements

From a strong signal to a nearly invisible afterglow

The first acquisition, from07:41 to 08:41, starts at about591,655 cpsand ends with a final-minute average of approximately22,660 cps. A further reading of about16,400 cps at 10:13was noted during the morning. The second complete acquisition available to us, from14:09 to 15:58, documents the continuing decline and, notably, several temporary increases.

Figure 1 · Earlier decay sessions; the independent final control appears in Figure 5
2026-11-11T15:31:56.966629 image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ 0 1 2 3 4 5 6 7 8 Hours since the start of the first acquisition 10k 100k Counts per second (log scale) After intense optical excitation of the sample Acquisition 1 · 07:41–08:41 Acquisition 2 · 14:09–15:58 10:13 reading · ~16,400 cps Reference dark count · 11,800 cps Empirical 1/t model (for comparison only)
Time evolution of the count rate averaged over 30-second intervals. The colored segments represent the two acquisition files actually analyzed; the orange diamond is a reading noted during the experiment, not a point recovered from a CSV file. No line artificially bridges unmeasured intervals. The 1/t model is illustrative rather than a complete physical fit.
Figure 2 · First hour: a count-rate reduction exceeding one order of magnitude
2026-11-11T15:31:57.106214 image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ 0 10 20 30 40 50 60 Minutes since the start of the first acquisition 100k Counts per second 30 s average R(t) = 11,800 + 4.17 × 10⁷/t 11,800 cps background
The steep initial drop gives way to a slow tail. The empirical expression shown, estimated from the less steep portion (t ≳ 400 s), does not describe the initial excitation. The time origin is the first recorded gate, not the exact moment when the excitation light was switched off.
RecordLocal time (CET)Valid gatesKey result
First hour · fosfori1h.zip07:41:22–08:41:2336.007~592,000 cps initially → ~22,660 cps in the final minute
Intermediate check· noted reading10:13—~16,400 cps observed during the experiment
Long acquisition · longRun(1).zip14:09:18–15:58:1865.401~13,580 cps in the first minute; ~13,032 cps in the final 10 minutes
Independent end-of-day control · controprovafinale.zip16:32:50–16:52:43 (with pause)10,006Sample present ~12,429 cps; removed ~11,668 cps overall; +767 cps last-five-minute differential

The first two complete files contain 101,408 valid gates, corresponding to about 2 h 49 min of actual sampling; the end-of-day sample-removal control adds 10,006 more gates, for a total of 111,414 gates (~3 h 06 min of active integration). Every gate has a nominal 100 ms integration. The three files span more than nine hours of clock time, including intervals when no measurements were recorded. The control run contains an intentional recording pause and its skipped sequence interval; no values are interpolated.

Verified result:at the end of the 15:58 session, approximately8 h 17 min after the first recording began, the average over the last ten minutes is13,032 cps. Relative to the previously measured background, about1,232 cpscan be attributed, as a first approximation, to the additional signal observed with the sample in place.

A successful numerical prediction, not a proven physical law

A simple approximation had been proposed from the early points along the decay tail:R(t) ≈ 11,800 + 4.17 × 10⁷/t, with t in seconds. It predicted approximately16,370 cps at 10:13, compared with the noted reading of about 16,400 cps. At the beginning of the 14:09 session, it predicted about13,590 cps, compared with a recorded first-minute average of approximately13,580 cps. The close agreement is striking, but it does not establish a specific physical decay exponent.

R(t) ≈ D + A/t · D = 11,800 cps · A = 4.17 × 10⁷ counts

The model is used only as a comparison over the hours following the initial phase; it is undefined at t = 0 and is not presented as the unique physical model.

03 / What the detector sees

The curve is not perfectly monotonic

The second recording includes severalslow increases in count rate, described during the experiment as small “hills.” The most prominent occur approximately9, 34, and 70 minutesafter the start. Changes of several hundred cps are far too large to be explained solely by Poisson fluctuations in multi-minute averages. However, no stable 120-second periodicity is evident.

Figure 3 · Long acquisition: three broad rises and a more regular final section
2026-11-11T15:31:57.202840 image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ 0 20 40 60 80 100 Minutes since the 14:09 restart 12000 12500 13000 13500 14000 14500 Counts per second Slow, nonperiodic rises More regular final section 30 s average ~120 s moving average Reference background
The thin line represents 30-second averages, and the heavier line a moving average of approximately 120 seconds. Vertical lines mark the intervals containing the largest rises; they do not indicate periodic oscillations. The final portion, where the curve settles toward 13,000 cps, is noticeably more regular over multi-minute timescales.

Between the first five minutes and the 10–15-minute interval, the average rises from approximately13,586to14,104 cps. During the closing segment, the final ten minutes average approximately13,032 cps, with much smaller variations over multi-minute intervals. We do not interpret these features as a newly discovered property of the phosphor: we simply report what the instrument measured.

Figure 4 · Cold-plate temperature during the long acquisition
2026-11-11T15:31:57.292198 image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ 0 20 40 60 80 100 Minutes since the 14:09 restart −20.2 −20.1 −20.0 −19.9 −19.8 −19.7 Cold-plate temperature (°C) Cold plate
The recorded cold-plate temperature ranged from −20.084 to −19.817 °C. The SiPM junction and sample temperatures were not measured directly; the absence of a clear correspondence with the large count-rate rises does not rule out every possible temperature dependence.
Electrical stability

Constant settings within each session

Commanded threshold: 2.300 V. Set overvoltage: 2.5 V. Commanded bias: 28.819 V in the first recording and 28.842 V in the second (a 23 mV difference between sessions). These arerecorded setpoints, not continuous independent voltage measurements at the circuit nodes.

Data integrity

Digital counting with no reported losses

In both files:zero missing gatesandzero FIFO overflowsamong the valid samples. Gate timing comes from the device clock; the PC UTC timestamp indicates reception time. The gate cadence is not inferred from the arrival of USB packets.

04 / Independent end-of-day check

Removing the paint: the detector returns to background

A final control recording was made on 11 November 2026. Unlike a background value taken from another session, this recording includes both the light-source measurement and the sample-out check in the same file, separated by a deliberate pause. The nominal integration time was 100 milliseconds for every gate: raw CSV values are integer counts per 0.1 s, subsequently converted into counts per second (cps) using the gate clock.

Figure 5 · Final test: raw 100 ms gates, smoothed rates, and count distribution
2026-11-11T15:58:04.909198 image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ 0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 20.0 Minutes since the start of the control recording 11000 11500 12000 12500 13000 13500 Count rate (cps) PAUSE ~3 min 12 s 100 ms gates · light points show 1 in every 10 gates Paint in place · 30 s rolling mean Paint removed; bias recalibrated · 30 s rolling mean 1050 1100 1150 1200 1250 1300 1350 1400 Raw counts in each 100 ms gate 0.000 0.002 0.004 0.006 0.008 0.010 0.012 Probability density Paint present: 1241.2 counts/100 ms (last 5 min) Paint removed: 1164.5 counts/100 ms (last 5 min)
The top panel uses a 30-second rolling mean on each uninterrupted segment; faint dots show one gate per second to keep the SVG readable, while every 100 ms gate contributes to the analysis. The shaded interval is the intentional pause (no measured values are interpolated). The lower panel shows the distributions of the actual 100 ms integer counts from the last five minutes of each phase.
Recorded phaseLocal clock (CET)Valid 100 ms gatesOverall meanLast 5 min mean
Paint present16:32:50–16:42:536,026 (10 min 02 s)12,429 cps12,412 cps
Paint removed, after recalibration16:46:05–16:52:433,980 (6 min 38 s)11,668 cps11,645 cps
Difference (last 5 minutes)Two independent phasesApproximately 3,000 gates per phase—+767 cps

The operator paused recording at 16:42:53, disabled the bias, removed the paint sample, recalibrated the detector at approximately 16:44:02–16:44:05, allowed about two minutes for settling, and resumed gate recording at 16:46:05. The commanded bias was 28.822 V before removal and 28.819 V afterwards, a difference of only 3 mV. The set overvoltage (2.5 V), comparator threshold (2.300 V), and cold-plate temperature near −20 °C were essentially unchanged. The bias is a software-commanded value, not a separately calibrated voltage trace.

Key result: the source-present and sample-removed recordings differ by about 767 cps when comparing the closing five minutes of each phase. This is equivalent to approximately 76.7 additional raw counts in every 100 ms gate. The independently re-measured background at the end is about 11,645 cps; thus we need not infer the entire excess from the older rounded 11,800 cps reference.

The difference is far larger than the Poisson scatter of the multi-minute averages. The post-removal count rate nevertheless drifts downward by roughly 150 cps over its short recording, so the measurement should not be represented as an absolute optical calibration or an uncertainty of only a few cps. Removal, recalibration, and the non-simultaneity of the two phases remain potential systematic contributions. Taken together with the earlier dark measurements, the control strongly supports a real sample-associated excess in the count rate.

Data-quality note: the sequence gap at the restart is intentional, not an interpolation. The paused interval has 1,922 skipped gate-sequence positions and an accumulated FIFO overflow marker in the source file; these are excluded from analysis. Within each of the two actively recorded segments, the 100 ms gate sequences are contiguous. Source: controprovafinale.zip, samples.csv, events.csv, and metadata.json.

05 / Order-of-magnitude estimate

How many photons might we be measuring?

A count-rate measurement is not automatically a radiometric measurement. Nevertheless, we can convert the independently measured sample-present minus sample-removed difference from the final control into an illustrative photon flux. Earlier sections use the historic 11,800 cps reference; the new paired measurement provides a stronger differential check.

Measured excess in the final control

~767 counts/s · 100 ms gates

Difference between the closing five-minute averages. At 100 ms/gate, this is ~76.7 additional detected counts per gate.

Illustrative photon-equivalent rate

~3,837 photons/s · ~1.44 fW

Assuming 20% effective PDE and 530 nm photons; optical power incident on the sensor, not absolute detector sensitivity.

With the same illustrative optics used below (30 cm distance, Ø31.75 mm lens, 80% transmission, Lambertian emission, and complete imaging of the paint spot), the sample-out differential corresponds to approximately 1.71 million photons/s emitted into the forward hemisphere, or about 0.64 picowatt. The latter figures are especially uncertain because the spot was aligned by eye.

The original single-run estimate remains below as an earlier measurement at 15:58: ~1,232 cps above a previously measured 11,800 cps background, or ~2.31 fW under the same photon assumptions. It is not a direct contemporaneous source-out comparison; the ~1.44 fW estimate above is based on the later, separate end-of-day control.

Photons incident on SiPM ≈ (Cmeasured− Cdark) / PDE

The manufacturer specifies for the MicroFC-30050 a35% peak PDE at 420 nm with 2.5 V overvoltage. For an assumed green emission near 530 nm, we use20% as an illustrative value: the paint spectrum was not measured, and the effective PDE of the complete counting chain was not calibrated. The plot shows how the estimate changes under different assumed PDE values.

Figure 6 · Equivalent optical power at the sensor versus assumed PDE
2026-11-11T15:31:57.368973 image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ 10% 15% 20% 25% 30% Assumed PDE at the emission wavelength 0 1 2 3 4 5 6 Equivalent power at SiPM (fW) 4.62 3.08 2.31 1.85 1.54
Conversion of the net count rate over the final ten minutes, assuming photons at 530 nm. The middle bar (20%) represents the working scenario, not a SiPM calibration.
Optical flux at the sensor

~6,162 photons per second

With an assumed PDE of 20%, an excess of ~1,232 cps corresponds to approximately 6,162 incident photons/s. Each green photon at 530 nm carries ~3.75 × 10⁻¹⁹ joule; the equivalent incident optical power is~2.31 femtowatts.

Emission from the observed paint spot

~2.75 million photons/s

Assuming Lambertian emission, a 30 cm distance, a 31.75 mm lens diameter, 80% transmission, and the entire spot being collected, the estimated power emitted into the forward hemisphere is approximately1.03 picowatts. The eye-based alignment makes this latter estimate particularly uncertain.

Geometry used for the estimate

Paint spotLens Ø 31.75 mmSiPM spot not precisely centered estimated distance: 300 mm 3 × 3 mm

At the assumed distance, with a 55 mm focal length and a 3 × 3 mm sensitive area, the calculated nominal field of view is approximately13.4 × 13.4 mm. The paint spot could fit within this field, but we do not know whether it is centered: some of the emitted light may miss the sensitive area.

fcollection≈ D²/(4r²) = 0.280% · Nemitted≈ Cnet/(PDE · T · fcollection· ηarea)

D = lens diameter; r = lens-to-spot distance; T = transmission; ηarea= fraction of the paint spot’s light actually imaged onto the sensitive area (unknown). The collection expression assumes Lambertian emission into the forward hemisphere and the small-angle geometrical approximation. It is not an absolute optical power measurement.

Assumed green-light PDEIncident photons/sEquivalent power at SiPMEmission into hemisphere*
10%~12,323~4.62 fW~2.06 pW
20% (central scenario)~6,162~2.31 fW~1.03 pW
30%~4,108~1.54 fW~0.69 pW

* Emission from the observed region, assuming the entire paint spot is imaged, Lambertian emission, T = 80%, λ = 530 nm, and no corrections for electronic threshold, afterpulsing, or crosstalk. If ηareais below 1, the estimated total emission increases in proportion to 1/ηarea. Uncertainty in the reference dark count also propagates directly into the count-rate difference.

What these numbers do NOT mean.“~2.31 fW” denotes anestimated equivalent optical powerthat would produce the observed count rate under the stated assumptions. It is not a calibrated radiometric measurement, nor the detection limit of the instrument. In particular, the differential count rate must not be confused with a certified NEP or absolute-sensitivity figure.
06 / Open questions

The “hills”: possible explanations, no premature conclusions

This is aninstrument test, not a specialist study of phosphors. The paint’s precise chemical composition, grain-size distribution, and emission spectrum are unknown. The rises in the curve are measured; their mechanism remains unresolved.

Hypothesis A · Material

Trap populations and nonuniform release

Persistent luminescence can involve processes with widely different relaxation times. Some emitting centers might show transient increases; nonmonotonic behavior of individual spectral components has been reported for certain strontium-aluminate materials.We have neither identified the pigment nor verified that these processes apply to it.

Hypothesis B · Spectrum

Changes in wavelength distribution

The SiPM weights detected photons according to its PDE(λ). A spectral shift can therefore change the count rate without a proportional change in total optical power. Without spectrometric data, this remains a hypothesis rather than a verified interpretation.

Hypothesis C · Geometry

Nonuniform paint spot and imperfect alignment

The surface is not homogeneous, and the field of view was not calibrated. Different regions may contribute with different weights. However, a mere sum of components that are all decreasing cannot produce a rise; either the contributions of the regions or their emission would also need to change with time.

Hypothesis D · Instrument and environment

Effects not covered by telemetry

Cold-plate temperature and recorded setpoints are stable, but we do not directly monitor junction or paint temperature, short-term stray light, geometrical drift, or spectral characteristics. These remain open possibilities, not demonstrated instrumental faults.

A 2025 paper, in particular, reports nonmonotonic time responses of some RGB components in specific strontium aluminates. It is relevant to the general plausibility of such a process,not proof that it occurs in our paint.

07 / Evidence and limitations

What this test demonstrates — and what it does not

Well-supported observations

Counting, duration, data integrity, and method repeatability

The instrument records more than 111,000 time gates across three source files, tracks a long decay, and then reproduces a lower background upon removal of the sample. Multi-minute rises occur in the data, not merely in the on-screen display. The final test explicitly uses 100 ms hardware-timed gates.

Acknowledged limitations

No optical bench was conjured up on paper

The test lacks radiometric calibration, a measured emission spectrum, independent voltage measurements, and verified optical alignment. The final same-file removal control supports the differential count-rate measurement, but does not eliminate these limitations or determine the exact photon emission rate of the paint.

Earlier decay segments were compared with an approximately 11,800 cps background from a separate measurement. The end-of-day sample-out control adds a direct, independently recorded reference in the same acquisition file: the detector returned to approximately 11,645 cps in the closing five minutes after sample removal and recalibration. These two phases were sequential, not simultaneous, and their stability cannot be certified beyond the recorded telemetry.

The principal result is not the discovery of a new property of phosphorescent paint.It is the demonstration that a cooled-SiPM prototype, with digital acquisition and telemetry, can document the tail of an extremely weak optical emission over many hours — and resolve temporal structures that would otherwise go unnoticed.

Data and reproducibility

This article uses fosfori1h.zip, longRun(1).zip, and controprovafinale.zip (each containing samples.csv, events.csv and metadata.json). All three archives declare 100 ms integration for the sample gates. Graphs report cps by dividing counted events by hardware-clocked gate duration; cumulative running-average display values were not used as raw data. The publication package includes the 30-second aggregate tables, plus the original final-control archive for independent checking.

08 / References

Documentation and references

  1. onsemi, C-Series SiPM Sensors, datasheet MICROC-SERIES/D. Active area and peak PDE for the C-Series family; spectral curves in the datasheet are representative of the series and do not constitute a calibration of the specific device used.
  2. Rise and decay time responses of Sr aluminate phosphorescent materials, Journal of Luminescence, 235 (2021), 117985. Discussion of non-exponential decay and spectral dependence.
  3. Temperature- and Emission Wavelength-Dependent Time Responses of Strontium Aluminates, Crystals, 15 (2025), 744. Examples of differing temporal responses among spectral components.
  4. Officina Turini — otSPAD. Overview of the prototype and its capabilities.