Full well capacity (FWC) is the maximum number of photoelectrons a single pixel can hold before it saturates, and dynamic range in dB is defined as 20×log10(FWC divided by total noise), with noise being the sum of dark and read noise components [S1][S3].
Industrial machine-vision CMOS sensors typically deliver 60-75 dB of dynamic range, which is sufficient for standard factory-automation tasks [S5]. Cooled scientific CCDs can push the figure to roughly 100,000:1, about 100 dB, when read noise is suppressed to 2-5 e- per pixel with Peltier cooling [S3].
How Full Well Capacity and Noise Set the Dynamic Range Number
Dynamic range is the ratio of the largest storable signal to the smallest detectable signal, and it is reported either in dB (20×log10) or as a linear ratio such as 4,490:1 [S3]. Photodiode size is the dominant physical lever: Hamamatsu cites modern CCD photodiodes from 4.5 to 24 microns with corresponding well capacities of 20,000 to 600,000 electrons, while read noise typically runs 10-20 e- at room temperature and 2-5 e- on Peltier-cooled chips [S3].
Industrial camera datasheets must therefore be read as a paired spec: a 2 MP CMOSIS CMV2000 lists 13,500 e- FWC and 13 e- dark noise for 60 dB, while the high-full-well CSI2100 lists roughly 2 Me- FWC and 945 e- noise, also about 66 dB, demonstrating that 148× more charge capacity moves the dB number by only 6 dB [S2]. The same point is reinforced by Teledyne's framing that dynamic range describes the ratio between the highest and lowest signals the camera can display [S1].
High Full Well vs High Dynamic Range: Two Different Optimization Targets
High full well capacity and high dynamic range are not the same design goal, because dynamic range is FWC divided by noise, not FWC alone [S2]. A high-FWC sensor (for example, Adimec's Q-2HFW with the CMOSIS CSI2100 at about 2 Me-) is engineered for shot-noise-limited measurements where a small signal rides on top of a bright background, with white-light metrology, heterodyne detection, and optical coherence tomography listed as typical applications [S2].
High dynamic range (HDR) mode, by contrast, is built for contrast in a single frame, bright and dim features side by side, such as reflective metal next to matte plastic on a production line [S2]. The CMV2000's HDR mode pushes the dynamic range to 90 dB, equivalent in signal capacity to a 400,000 e- well even though the actual pixel well stays at 13,500 e-; the gain comes from pixel-electronics tricks, not from physically enlarging the well [S2]. For broader machine-vision deployment, the practical 60-75 dB band from standard industrial CMOS covers most factory-automation scenes without the extra cost of cooled scientific devices [S5].
Reading Datasheets: FWC, Linear FWC, and Bit-Depth Traps
Full well capacity is bounded by two independent limits: the physical potential well inside the pixel silicon, and the digital grayscale ceiling set by the ADC bit depth [S4]. The term linear full well capacity (linear FWC) is the highest signal level at which the sensor still responds linearly to light, and most scientific datasheets publish that figure rather than the absolute saturation capacity, because the linear region is the only one that yields quantitative measurements [S4].
Once the pixel approaches saturation, the residual electric field inside the well collapses, quantum efficiency drops, and the response becomes non-linear, which is why simply cranking exposure time is not a substitute for a larger FWC [S4]. Many scientific cameras ship with multiple readout modes; selecting a lower bit depth can access only a slice of the available physical FWC, even though the silicon can hold more charge than the ADC can code [S4]. Pairing a 12-bit ADC (4,096 levels) with a 100,000:1 dynamic range device is a common mismatch, with the Hamamatsu CCD39-01 example requiring a 17-bit, 131,072-level converter to fully digitize its 100,000:1 range [S3].
Sensor Class Comparison: Industrial CMOS, Cooled CCD, EMCCD
Three sensor classes cover most industrial and scientific workloads, and they split cleanly on the FWC-versus-noise trade-off. Standard industrial CMOS lands in the 60-75 dB band with FWC typically in the 10,000-100,000 e- range and read noise in single-digit to low double-digit electrons [S5]. Cooled scientific CCDs reach roughly 100 dB by trading integration speed for noise suppression: a 24 micron pixel device like the Marconi CCD39-01 stores up to 300,000 e- with 3 e- RMS read noise at 20 kHz when cooled, giving 100,000:1 dynamic range [S3].
Electron-multiplying CCDs (EMCCDs) swap some FWC for sub-electron read noise through on-chip gain registers: a 100,000 e- full-well with 40 e- conventional noise and EM gain on still produces a 2,500:1 dynamic range, which a 12-bit ADC (4,096 levels) can already capture [S9]. Across these three classes, the EMCCD favours low-light, high-frame-rate imaging, the cooled CCD favours quantitative photometry, and the standard industrial CMOS favours cost-sensitive factory-automation lines that do not need more than 75 dB [S3][S5][S9].
Selection Criteria for Industrial Specifiers
Selecting a camera for an industrial line should start with the contrast ratio in the scene, not with the headline dB number. For uniform-bright scenes with a faint modulation, such as laser-line profiling or interferometric metrology, a high-FWC CMOS part in the 60-66 dB band is the right pick, because more charge per pixel suppresses shot noise on the bright background [S2]. For mixed-reflectivity scenes, such as inspecting shiny metal alongside dark plastic, an HDR-mode sensor at 90 dB or a cooled CCD at roughly 100 dB is the right pick, because the limiting factor is read noise on the dark side of the frame [S2][S3].
Specifiers should also verify three numbers together: pixel pitch in microns, FWC in electrons, and read noise in electrons RMS, then compute 20×log10(FWC/noise) themselves rather than trust the marketing dB figure, because that calculation exposes whether the sensor is high-FWC or low-noise optimised [S2][S3]. Cooling helps: dark current roughly doubles every 8-10 °C, so a Peltier-cooled stage is often the cheapest way to lift dynamic range by 6-12 dB in process environments with stable ambient temperature [S3]. For factory-automation deployments where 75 dB is adequate, the machine-vision glossary entry on dynamic range is a useful starting point for the lighting-side assumptions that pair with the sensor spec [S5].
Limitations, Failure Modes, and Multi-Bit Sampling Workarounds
The single-exposure dynamic range of any image sensor is bounded by FWC at the top and total noise at the bottom, so a single frame cannot exceed that ratio regardless of how good the optics are [S3][S6]. Pixel saturation destroys quantitative information, and even before hard saturation the response becomes non-linear, which is why linear FWC is the spec engineers should track on a datasheet rather than the absolute saturation value [S4].
Multi-bit temporal sampling is one documented workaround: a 2013 International Image Sensor Workshop paper by T. Vogelsang showed that the dynamic range of an image sensor need not be limited by FWC if the sampling policy is changed to choose when and how to sample, effectively trading frame rate for dynamic range [S7]. In industrial practice this corresponds to multi-frame HDR bracketing or split-pixel gain architectures, both of which sit on top of the underlying FWC-versus-noise physics, not in place of it [S2][S7]. For the broader machine-vision and lighting context, the industrial camera and lighting overview covers the optical side assumptions that feed into these sensor decisions, while process engineers weighing measurement-quality trade-offs can also reference OEE framing in continuous process industries when justifying the cost of a cooled scientific camera on a production line.
The underlying component specifications are covered under dynamic compactor, and dynamic balancing machine.