Dark current is unwanted current generated or collected without the intended illumination. Read noise is uncertainty introduced when the stored signal is converted and read. Dark current can produce both a mean dark signal and temporal fluctuations; it should not be used as a synonym for every noisy pixel.
Start with a pixel held in darkness
Even without the intended light, a sensor can accumulate charge from thermal generation and other leakage mechanisms. That accumulated charge contributes a mean dark signal. Its amount depends on the device and conditions, including the time available for accumulation.
The generation process also fluctuates. Under a simple independent-event model, the associated shot-noise variance follows the mean number of accumulated events. Removing a mean offset does not remove the random realization present in an individual exposure.
Process checkpoint
Understand HKD/AR1 AlO deposition in context
Explore backside surface passivation as one contribution to dark-current control.
Process context for “Dark Current vs Read Noise: Separate Charge and Readout”: 40nm BSI CMOS Image Sensor · BKPAS · Step 307
Readout adds another layer of uncertainty
Charge must be converted into an electrical signal and passed through a readout chain. Transistor noise, sampling behavior and later electronics can contribute to uncertainty. Their input-referred magnitude depends on the chain and its conversion gain; it is not inferred from dark-current magnitude alone.
| Component | What is varying or accumulating? | Useful distinction |
|---|---|---|
| Mean dark signal | Average unwanted accumulated charge | Offset is not the same as random noise |
| Dark-current shot noise | Event-to-event fluctuation in that accumulation | Mean subtraction does not erase it |
| Read noise | Uncertainty added during signal conversion/readout | Can remain when dark accumulation is small |
| Pixel-to-pixel dark variation | Different mean dark behavior among pixels | Spatial nonuniformity is not identical to temporal noise |
Why one dark frame cannot identify everything
A dark image can contain mean offsets, spatial nonuniformity and temporal fluctuations at once. A bright outlier could involve unusually high dark generation, a readout offset or another defect. The image alone does not establish one cause.
As a conceptual exercise, compare repeated dark observations while separating the average of each pixel from its frame-to-frame variation. Then ask how the result changes with accumulation time while keeping the readout conditions comparable. A time-dependent accumulation and a readout contribution are different hypotheses, although real devices can contain mixed or time-dependent effects.
This is a reasoning framework, not a calibration recipe. It does not assume that all noise is white or that every pixel follows the same model.
Process interfaces and circuit design meet here
Semiconductor defects and interfaces can influence unwanted charge generation. The readout transistor and sampling architecture introduce their own behavior. Improving one source may reveal another previously hidden contribution rather than making the image noiseless.
Likewise, higher conversion gain can change the significance of downstream voltage noise when referred back to input charge, but it does not establish that dark generation disappeared. A useful comparison states the signal reference and measurement conditions instead of mixing voltage noise, electron-referred noise and dark current.
Source links
References
Modern Semiconductor Devices for Integrated Circuits - MOS Transistor
Chenming Hu
Modern Semiconductor Devices for Integrated Circuits · Ch6 MOS Transistor
Physics of Semiconductor Devices - Full
S. M. Sze, Kwok K. Ng
Physics of Semiconductor Devices · ISBN 978-0-471-14323-9