In high-density optical density filtration—such as pairing a 10-stop (ND1000 / Optical Density 3.0) or 15-stop (ND32000 / Optical Density 4.5) filter with low-aperture cinema glass—exposure durations frequently extend from standard fractional shutter speeds into the range of 120 to 600 seconds. While chemical photographic emulsion suffered from the classic Schwarzschild effect (reciprocity failure caused by low-intensity reciprocity breakdown in silver halide crystals), silicon-based CMOS sensors exhibit an entirely different physical failure mode: thermal dark current accumulation and fixed-pattern noise (FPN) scaling.
As exposure duration lengthens, thermally agitated electrons spontaneously migrate across the silicon depletion boundary even in the total absence of incident photons. This study documents empirical dark-current charge leakage curves, evaluates the Arrhenius temperature relationship, and details operational workarounds for maintaining dynamic range in extreme long-exposure cinematography.
1. Physics of Dark Current in Silicon Depletion Regions
Within a pinned photodiode (PPD) CMOS sensor, charge accumulation should ideally reflect only photoelectric conversion: photons striking the valence band to promote electrons into the conduction band. However, at temperatures above absolute zero, thermal lattice vibrations (phonons) continually excite electron-hole pairs via Shockley-Read-Hall (SRH) generation-recombination centers.
The rate of dark current generation ($I_{dark}$) scales exponentially with junction temperature ($T_k$, in Kelvin) according to the empirical Arrhenius equation:
Where E_g is the bandgap energy of silicon (~1.12 eV at 300K), k_B is Boltzmann's constant ($8.617 \times 10^{-5}\text{ eV/K}$), and C is a structural proportionality constant. In practical engineering terms, dark current approximately doubles for every 6°C to 8°C increase in internal sensor silicon temperature.
| Sensor Core Temp (°C) | Exposure Duration | Mean Dark Current Signal | Dynamic Range Degradation | Hot Pixel Count (>50% Saturation) |
|---|---|---|---|---|
| 12°C (Cold Ambient) | 60 seconds | 0.42 e⁻ | -0.08 EV | 14 pixels |
| 24°C (Nominal Lab) | 60 seconds | 1.38 e⁻ | -0.21 EV | 68 pixels |
| 36°C (Thermal Soak) | 60 seconds | 4.52 e⁻ | -0.65 EV | 412 pixels |
| 36°C (Thermal Soak) | 300 seconds | 22.60 e⁻ | -1.84 EV | 2,840 pixels |
| 42°C (Processor Load) | 600 seconds | 94.80 e⁻ | -3.15 EV | 11,350 pixels |
As documented in our controlled thermal chamber telemetry, allowing the sensor die to reach 42°C during a 600-second ND1000 daylight exposure consumes nearly 95 electrons of full-well capacity purely from thermal leakage, compressing highlight headroom and elevating shadow noise floors beyond the capability of standard debayer denoisers.
Empirical ND Filter Shutter Speed Calculator
Calculate precise optical attenuation, reciprocal exposure compensations, and EV reductions across all standard ND density ratings from ND0.3 through ND4.5.
Launch ND Filter Exposure Engine →2. Fixed-Pattern Noise (FPN) vs. Shot Noise
The total noise profile during ultra-long exposures decomposes into two distinct components:
- Temporal Dark Shot Noise ($\sigma_{dshot}$): The statistical variance of thermal electron arrival governed by Poisson statistics ($\sigma_{dshot} = \sqrt{I_{dark} \times t}$). This noise is completely random across frames.
- Dark Current Non-Uniformity (DSNU / FPN): Microscopic manufacturing variations in individual photodiode silicon dopant concentrations cause certain pixels ("warm" or "hot" pixels) to accumulate dark current at rates 10× to 100× higher than their neighbors. Unlike shot noise, DSNU is spatially invariant.
3. Laboratory Mitigation Protocols
To maintain signal-to-noise ratio (SNR) above 42 dB during long-exposure architectural and cinematic captures, production workflows must enforce two primary field mitigation strategies:
Protocol A: Long Exposure Noise Reduction (LENR / Dark Frame Subtraction)
The camera firmware records a secondary exposure of identical duration immediately following the primary exposure with the physical shutter closed. The resulting dark frame captures the exact spatial footprint of the DSNU fixed-pattern noise at the prevailing internal temperature. Firmware subtracts the dark frame pixel-by-pixel from the raw light frame:
While this mathematically eliminates 100% of spatial fixed-pattern noise, it increases total temporal shot noise by a factor of $\sqrt{2}$ (~1.414×) and doubles total shooting cadence.
Protocol B: Post-Production Median Bias Stacking
For continuous motion or rapid light shifts where a 1:1 dark frame delay is prohibitive, optical teams record five master dark frames at the conclusion of the shoot under matched thermal equilibrium. Computing a temporal median pixel rejection stack across the five dark frames creates a pure master bias reference without injecting the temporal Poisson shot noise penalty of single-frame LENR.