Skip to main content

CCD & CMOS Imaging Sensors — Abridged Guide

Quick-reference guide to imaging sensors — CCD/CMOS architecture, noise, QE, binning, shutters, and selection. For full derivations and worked examples, see the Comprehensive Guide.

Comprehensive Imaging Sensors Guide

1.Introduction

CCD and CMOS are both silicon-based sensors exploiting the photoelectric effect. They differ in how charge is transferred and read out: CCD transfers charge sequentially to a shared amplifier; CMOS converts charge to voltage at each pixel and reads in parallel.
When comparing sensor specs, focus on four numbers first: read noise (e⁻ rms), quantum efficiency (%), dark current (e⁻/p/s at stated temperature), and frame rate (fps).

2.CCD Architecture

Three CCD variants exist: full-frame (highest fill factor, needs mechanical shutter), frame-transfer (continuous imaging, double silicon area), and interline-transfer (electronic shutter, reduced fill factor). Full-frame and frame-transfer dominate scientific applications.
Full-frame CCDs offer the best QE and lowest noise but require a shutter. If continuous acquisition without a shutter is needed, frame-transfer or interline-transfer architectures are required.

3.CMOS Architecture

The 4-transistor (4T) pinned-photodiode pixel with correlated double sampling (CDS) is the architecture that brought CMOS read noise to sub-electron levels, enabling sCMOS to compete with and surpass CCDs in most imaging applications.
CMOS sensors with column-parallel ADCs read entire rows simultaneously, enabling frame rates 10–100× faster than CCDs at equivalent pixel counts.

4.Noise & SNR

Full SNR
SNR=SS+Idt+σr2\text{SNR} = \frac{S}{\sqrt{S + I_d \cdot t + \sigma_r^2}}
S = signal (e⁻), I_d = dark current (e⁻/p/s), t = integration time (s), σ_r = read noise (e⁻ rms)
Three noise regimes determine sensor choice: read-noise limited (low signal → reduce read noise or use EMCCD), shot-noise limited (moderate signal → ideal regime, collect more photons), dark-noise limited (long integration → cool the sensor).
To quickly estimate if read noise matters: if the signal exceeds ~10× the read noise squared, the measurement is shot-noise limited. For a 3 e⁻ sensor, that threshold is about 90 photoelectrons.
Noise SourceDepends OnTypical ValuesMitigation
Photon shot noiseSignal level√SCollect more photons
Read noiseAmplifier, readout speed1–15 e⁻ rmsSlow scan, sCMOS, EMCCD
Dark noiseTemperature, integration time√(I_d · t)Cool the sensor
Fixed-pattern noisePixel-to-pixel variationCalibration-dependentDark subtraction, flat-field
Imaging Sensor SNR Calculator

5.QE & Spectral Response

Back-illuminated sensors achieve 90–95% peak QE by eliminating wiring obstruction. Front-illuminated sensors with microlenses reach only 55–65%. For any photon-starved application, BSI is the first specification to check.
Deep-depletion CCDs extend useful sensitivity to ~1050 nm by using thick, high-resistivity silicon. They require aggressive cooling due to higher dark current.
ArchitecturePeak QE (%)UVNIRUse
FSI CCD50–60PoorModerateLegacy
FSI CMOS + microlens55–65PoorModerateConsumer, machine vision
BSI CCD85–95GoodLimitedAstronomy, spectroscopy
BSI sCMOS90–95GoodLimitedMicroscopy, general scientific
Deep-depletion CCD85–90ModerateExcellentNIR spectroscopy

6.Dynamic Range & Binning

Dynamic Range
DR (dB)=20log10 ⁣(Nsatσnoise)\text{DR (dB)} = 20 \cdot \log_{10}\!\left(\frac{N_{\text{sat}}}{\sigma_{\text{noise}}}\right)
On-chip binning (CCD) improves SNR by N in the read-noise-limited regime. Software binning (CMOS) improves by only √N. This is why CCDs persist in spectroscopy: full vertical binning delivers 256× the signal with 1× the read noise.
Dynamic range ≠ bit depth. A 16-bit ADC does not give 16 bits of DR unless N_sat / σ_noise ≥ 65,536.

7.Cooling

Dark Current Halving Rule
Id(T2)Id(T1)2(T1T2)/6I_d(T_2) \approx I_d(T_1) \cdot 2^{-(T_1 - T_2)/6}
Dark current approximately halves for every 5–7 °C of cooling. TE cooling to −70 to −100 °C is sufficient for most scientific imaging. LN₂ cooling is reserved for deep-depletion NIR sensors and ultra-long integrations.
If dark noise is less than read noise, further cooling provides diminishing returns. Compare √(I_d · t) to the read noise spec first.

8.Shutters

Global shutter (all pixels exposed simultaneously) is inherent to most CCDs and essential for fast-moving objects. Rolling shutter (row-by-row readout) is standard in sCMOS and introduces skew artifacts for fast motion, but is negligible for static specimens.
For rolling shutter, total frame skew = N_rows × line_time. If object displacement during this time is less than one pixel, artifacts are undetectable.

9.Specialized Sensors

Five specialized sensor technologies: EMCCD (single-photon sensitivity), ICCD (nanosecond gating), sCMOS (lowest noise at highest speed), scientific CCD (on-chip binning for spectroscopy), high-speed CMOS (>10⁵ fps).
NeedBest TechnologyKey Advantage
Single-photon detectionEMCCDEM gain eliminates read noise
Nanosecond gatingICCDPhotocathode shutter <2 ns
Widefield imaging, 10–100 fpssCMOS (BSI)Low noise + high speed + large FOV
Long-integration spectroscopyScientific CCD (LN₂)On-chip FVB + negligible dark current
>1,000 fpsHigh-speed CMOSGigapixel/s throughput, global shutter
Photon-number resolvingqCMOS<0.3 e⁻ read noise, no excess noise
EMCCD is superior to ICCD in almost every metric except gating speed. If nanosecond time resolution is not required, EMCCD is the better choice.

10.Selection Workflow

Start with the photon budget, then consider temporal requirements, spectral range, and whether the application is spectroscopy (favors CCD) or imaging (favors sCMOS). sCMOS is the default for modern scientific imaging unless a specialized need dictates otherwise.
Beware of dynamic range claims using non-linear full well. Always request EMVA 1288 characterization data for standardized, comparable measurements.
Continue Learning

The Comprehensive Guide includes 7 worked examples, 7 SVG diagrams, 2 data tables, and 10 references.

All information, equations, and calculations have been compiled and verified to the best of our ability. For mission-critical applications, we recommend independent verification of all values. If you find an error, please let us know.