Officina Turini · Experimental photonics

Dark Count Performance &
Low-Light Detection Sensitivity

Two ten-minute laboratory evaluations of the otSPAD prototype with the onsemi MICROFC-30050-SMT-TR SiPM at a −25 °C cold-plate setpoint. Measured counting statistics, photometric threshold estimates and explicitly simulated C-Series alternatives.

Measurement: 09 November 20261 s and 10 s gates600 + 60 consecutive samplesClosed dark optical path
MEASURED DARK COUNT RATE8,420 cpsLast 300 s · not dark current in amperes
MEASURED DISPERSIONF = 1.001Sample Fano factor, 1 s gates
COLD-PLATE SETPOINT−25 °CSiPM junction temperature not directly measured
MEASURED CONTINUITY600 / 600No missing gates or FIFO overflow

Executive summary

The completed measurement shows a low-noise, stable dark-count stream after an initial thermal settling period. The full ten-minute run has 8,446 cps average counts and F = 1.329; using the final five minutes selected in advance of the sensitivity calculations yields 8,420 cps, a measured standard deviation of 91.8 counts per 1 s and F = 1.001. This closely matches a Poisson count process at the one-second time scale. An independent acquisition with 60 direct 10-second gates supports the same Poisson-compatible interpretation, within the larger variance uncertainty of the longer-gate sample.

Terminology matters. This experiment measures dark count rate (counts/s), not the device's electrical dark current (amperes). The report estimates the smallest additional optical signal above the dark background; it does not demonstrate an absolute incident-photon detection limit without a calibrated weak light source and the actual operating photon detection efficiency (PDE).
Recorded setupValueRecorded setupValue
DetectorMICROFC-30050-SMT-TRGate / run1 s / 600 s
Active sensor area3 × 3 mm²SiPM discriminatorTLV3501, 10 kΩ feedback
Cold-plate setpoint−25.0 °CMeasured plate, first → last−24.617 → −25.017 °C
Bias command28.727 VOvervoltage setting2.5 V (commanded)
Comparator threshold2.300 VCooling command, first → last73.5 → 76.4%
Acquisition clock80 MHzGate records / losses600 / 0

1. Direct observations MEASURED

2026-11-09T15:39:10.863767 image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ 0 100 200 300 400 500 600 Elapsed acquisition time (s) 8200 8300 8400 8500 8600 8700 8800 Counts in 1 s gate Reference interval Dark count time series — complete acquisition 1 s recorded counts 23 s running mean
One-second count totals. The 23-second running average visualizes the initial drift. The final 300 s are used for the reference dark rate and short-gate noise statistics; no data are deleted from the archived acquisition.
2026-11-09T15:39:10.951225 image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ 0 100 200 300 400 500 600 Elapsed acquisition time (s) −25.0 −24.9 −24.8 −24.7 −24.6 Cold-plate temperature (°C) Measured thermal settling — junction temperature not measured 73 74 75 76 77 TEC command (%)
The AD590 sensor monitors the cold plate. TEC drive is a command percentage, not measured electric power. The 0.8 mm sensor PCB can create a thermal offset and delay between the plate and junction.

Warm-up and baseline selection

The initial count rate decreases as the cold plate approaches its final reading and the TEC controller settles. To avoid conflating this drift with the intrinsic counting noise, only gate sequences 301–600 (300 full seconds) are used for the baseline sensitivity calculations.

The selected interval has mean 8,420 cps, 95% approximate confidence interval 8,409.5–8,430.4 cps for its mean under weak serial dependence, and a residual fitted trend of -0.093 counts/s² (linear-trend p = 0.128). The residual trend is not statistically significant at 5%; its possible systematic impact is not guaranteed negligible in later runs.

Statistical interpretation

At a one-second gate, the expected Poisson standard deviation is 91.8 counts. The measured value is 91.8. The Fano factor, Var(N)/E(N), is 1.001 with an approximate 95% interval of 0.858–1.183.

Lag-1 autocorrelation is +0.018, and no selected gate exceeds ±4 sample standard deviations. This supports—but does not prove—a near-Poisson counting model on this observation scale.

2026-11-09T15:39:11.046176 image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ 8200 8300 8400 8500 8600 8700 Counts per 1 s 0 10 20 30 40 50 Number of gate samples Stationary-window count histogram vs Poisson prediction Observed, final 300 one-second gates Poisson expectation at observed mean
Theoretical expected bin occupancy is computed from the Poisson CDF with mean fitted to the same selected window. Close agreement is descriptive, not proof of independent primary dark events.

Effect of longer integrations

Adjacent one-second gates were added without changing their order. The resulting variance can be compared with the Poisson expectation. Long integrations contain far fewer independent blocks and cannot be characterized to the same confidence from this ten-minute file.

2026-11-09T15:39:11.151230 image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ 1 2 5 10 20 30 60 120 Summed integration time (s) 0.0 0.5 1.0 1.5 2.0 Observed Fano factor n=5 n=2 Re-binned count variance (sample size shrinks at longer gates)
Measured Fano factors on non-overlapping groups of the final 300 samples. Annotated sample counts below ten are too few for robust inference; several fluctuations below 1 are not evidence that the detector beats Poisson noise.
IntegrationIndependent blocksObserved FanoReliability
13001.001Exploratory
21500.987Exploratory
5600.970Exploratory
10300.701Exploratory
20150.469Exploratory
30100.423Exploratory
6050.643Limited — few samples
12020.948Limited — few samples

Independent 10-second gate validation MEASURED

A second independent 600-second dark-count acquisition was conducted at the same −25 °C cold-plate setpoint, now using 60 direct 10-second gates. The new run started after the original one-second measurement and therefore did not exhibit its initial thermal settling. All 60 gate records were continuous, with no reported missing gates or FIFO overflows.

MetricOriginal 1 s runIndependent 10 s run
Samples / nominal run time600 × 1 s / 600 s60 × 10 s / 600 s
Full-run mean dark count rate8,445.7 cps8,436.1 cps
Mean rate, final 300 s8,420.0 cps8,428.8 cps
Final-window per-gate standard deviation91.8 counts / 1 s234.2 counts / 10 s
Final-window Fano factor1.001 (300 samples)0.651 (30 samples)
Approx. 95% Fano interval0.858–1.1830.413–1.176
Cold-plate reading at first / last gate-24.617 / -25.017 °C-25.017 / -25.017 °C
Missing gates / FIFO overflow0 / 00 / 0
2026-11-09T15:47:49.745969 image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ 0 100 200 300 400 500 600 Acquisition time in each run (s) 8400 8450 8500 8550 8600 8650 Dark count rate (counts/s) Independent 10-second cross-check: directly gated vs re-binned 1-second data Original 1 s records, re-binned to 10 s Independent direct 10 s gate Original stable reference 8,420.0 cps
These are separate experiments. The original series is summed in consecutive blocks of 10 one-second gates; the second series uses actual 10-second hardware gates. Their time axes are relative to their respective acquisitions, not synchronized physical times.

Comparison with the Poisson expectation

Using the original stable baseline of 8,420.0 counts/s, the ideal prediction for a 10-second gate is 84,200 counts with a standard deviation of 290.2 counts. The new stable 10-second gates contain 84,288.3 counts on average, with sample standard deviation 234.2 counts. The stable mean-rate difference between runs is 8.9 counts/s (0.105%).

2026-11-09T15:47:49.812440 image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ 1 s direct (n=300) 10 s re-binned (n=30) 10 s direct (n=30) 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 Fano factor ± approximate 95% CI Measured variance across integration scales — last 300 s Poisson expectation (F = 1)
Approximate confidence intervals assume stationary independent gates; estimates from 30 blocks are necessarily broad. The re-binned original 1-second data yield F = 0.701 over the final 300 seconds. The independent direct 10-second run yields F = 0.651 over the same relative interval.
Interpretation. The independent 10-second result is statistically compatible with a Poisson background. An observed Fano factor below 1 with only 30 stable ten-second gates does not establish sub-Poisson counting, lower detector noise or a better optical detection threshold. Both runs support retaining the original conservative Poisson-based S/N=1, 3 and 5 sensitivity estimates. Optical photon-flux sensitivity is still a model estimate until tested with a calibrated weak light source.

Independent run source: 25_10s.zip, session 20261109_153010_825_017fa285. The temperature sensor monitors the cold plate; SiPM junction temperature is not directly measured. The slightly different baseline rates need not reflect a change in photon detection efficiency.

2. Estimated optical detection threshold MODEL, NOT MEASURED

For a known stationary dark background of D = 8,420 counts/s, let t be the integration interval, p the detector PDE, and z the chosen ratio S/N. Defining S/N as the expected net signal counts divided by the standard deviation of dark counts, the idealized minimum photon flux incident on the active area is:

Background counts:             B = D × t
Ideal Poisson background noise: σB = √(D × t)
Required detected signal:      S = z × σB
Incident photon flux:          Φmin = z × √(D × t) / (p × t)
                                  = z × √D / (p × √t)

These are thresholds in photons arriving at the active surface per second for a source at the specified wavelength, not the number of photons that can be identified one by one. The 35% PDE is the onsemi datasheet's nominal 420 nm, 21 °C, +2.5 V overvoltage value; its exact value at the cold junction and in this comparator chain was not calibrated. At S/N=1 the signal is at the noise level—not a reliable detection declaration. S/N=3 or 5 is an amplitude criterion, not a guaranteed detection probability or formal detection limit.

Background subtraction matters. The ideal expression assumes a perfectly known background mean. If background is instead determined from a separate equal-duration dark exposure, its shot noise adds and the required flux is approximately √2 larger. A finite 300 s dark calibration adds an intermediate uncertainty under ideal stability. These alternatives can be selected below. No mode includes unknown junction offsets, light leakage, PDE drift or baseline changes between exposures.
—S/N=1 · incident photons/s
—S/N=3 · incident photons/s
—S/N=5 · incident photons/s

Photon flux matrix — S/N = 1, 3 and 5

Values in incident photons per second, across integrations. Defaults: 35% assumed PDE, known Poisson background. Changing PDE or the background treatment recomputes the table in your browser.

IntegrationS/N = 1S/N = 3S/N = 5
1 s262.2786.51,310.9
2 s185.4556.2926.9
5 s117.2351.7586.2
10 s82.9248.7414.5
20 s58.6175.9293.1
30 s47.9143.6239.3
60 s33.8101.5169.2
120 s23.971.8119.7
2026-11-09T15:39:11.521956 image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ 1 2 5 10 20 30 60 120 Integration time (s) 1 0 2 1 0 3 Estimated incident photons / s Idealized flux threshold — PDE 35% at 420 nm; Poisson background S/N = 1 S/N = 3 S/N = 5
Reference-model curves at nominal PDE=35% for a known background. They follow a 1/√t trend if the background remains stationary. The graph is static; the table above supports changing PDE and background calibration.

Absolute signal counts: the corresponding mean number of additional detected photoevents is simply S = p × Φmin × t. Long integration can reveal a weaker average flux, not eliminate dark events or establish that a particular individual count was optical.

3. Alternative C-Series sensors SIMULATED — NOT TESTED

onsemi specifies typical dark count rates at 21 °C and 2.5 V overvoltage of approximately 300 kcps for the 3 × 3 mm² C-Series and 30 kcps for the 1 × 1 mm² variants. For an illustrative projection to the present cold-plate setup, each candidate is scaled by its nominal room-temperature DCR ratio relative to the tested MICROFC-30050:

Dcandidate(−25 °C plate, simulated)
    = Dmeasured,30050 × DCRnominal,candidate(21 °C)
                         / DCRnominal,30050(21 °C)

Every candidate inherits the same relative temperature reduction, threshold acceptance, front-end counting behavior and 2.5 V overvoltage as the measured unit. None of those assumptions was experimentally validated for the other parts. PDE values refer to the nominal spectral peak at 420 nm under datasheet reference conditions, not measured cold PDE.

2026-11-09T15:39:11.619644 image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ 0 200 400 600 800 1000 1200 Photons incident on active area / s (S/N=3, 1 s) MICROFC-30050 MICROFC-30035 MICROFC-30020 MICROFC-10035 MICROFC-10020 MICROFC-10010 787 888 1,147 281 363 622 Other C-Series detectors — unverified extrapolations at −25 °C
Illustrative incident photon flux needed for an S/N of 3 with a 1 s gate and a perfectly known Poisson background. A small sensor must actually receive the modeled light; smaller dark area does not automatically improve system sensitivity under diffuse illumination.
onsemi SiPMArea mm²PDE at 420 nmNominal DCR, 21 °C (kcps)Projected DCR at plate −25 °C (cps)S/N=3, 1 s (photons/s)Flux density (photons/s/mm²)
MICROFC-30050 TESTED935.0%3008,420.0786.587.4
MICROFC-30035 SIMULATED931.0%3008,420.0888.098.7
MICROFC-30020 SIMULATED924.0%3008,420.01,147.0127.4
MICROFC-10035 SIMULATED131.0%30842.0280.8280.8
MICROFC-10020 SIMULATED124.0%30842.0362.7362.7
MICROFC-10010 SIMULATED114.000000000000002%30842.0621.8621.8
Optical coupling changes the ranking. For a focused beam entirely collected by the 1 mm² sensor, the simulated 10035 may give a lower count-rate threshold than the tested 30050. For uniformly illuminated surfaces, compare photons/s/mm² or total captured flux, not just the DCR. The small active area collects roughly one ninth as much uniform illumination as the 9 mm² detector at equal irradiance.

4. Measurement limits and reproducibility

What has been demonstrated

Two complete, independent 600-second dark-count acquisitions (1 s and 10 s gates), measured gate timing, temperature telemetry and stable final-window counting variance consistent with Poisson statistics. No missing gate sequence and no logged FIFO overflow. The stabilized measurement reflects the complete SiPM, TIA, comparator, bias and FPGA counting chain.

What remains unmeasured

Actual SiPM junction temperature; absolute PDE at −25 °C; calibrated photon flux at the sensor; the transient response to a weak illuminated source; true dark leakage current in amperes; dark count spectral characteristics on sub-second scales; and the minimum signal with explicitly verified false-alarm and detection probabilities.

Important qualification about temperature

The temperature measurement is on the cold side of the thermoelectric assembly. The SiPM is thermally coupled to this surface through a 0.8 mm printed circuit board. Thickness alone does not determine its thermal resistance: the copper layers, via density, contact pressure, interface material and heat load all matter. Therefore the label “−25 °C” throughout this report refers to the cold-plate setpoint and sensor measurement, not a verified semiconductor junction temperature.

Statistical and methodological choices

Entire 600-gate series included in this standalone HTML file.

Primary references

  1. onsemi, C-Series SiPM Sensors, MICROC-SERIES/D, Rev. 9 (February 2022) — active areas, nominal 21 °C DCR, peak PDE at 420 nm, overvoltage and temperature coefficients.
  2. Currie, L. A., Detection and quantification limits: basic concepts (2004) — distinction between critical level, detection limit and quantification limit.
  3. Officina Turini, SPAD/SiPM prototype — project description and instrument architecture.
  4. Experimental source: 25_1s.zip (samples.csv, events.csv, metadata.json), session 20261109_151913_128_5c72b3e9, started at 2026-11-09T15:19:13.128Z.
  5. Independent validation: 25_10s.zip (samples.csv, events.csv, metadata.json) with direct 10 s gates.