Increasing CMOS Camera Sensitivity Through Back-Illumination

Author: Dan Croucher, Applications Team Manager, Photometrics

Introduction

Scientific camera sensitivity is determined by three main factors; quantum efficiency (QE), pixel size and noise characteristics. Quantum efficiency is the measure of the effectiveness of the camera to produce electronic charge (electrons) from incident photons, where a higher QE results in the conversion of more photons to electrons of signal. Electrons go on to be converted into a digital signal that can be read by a computer and visualized. Pixel size relates to the physical area of the pixel, where a larger pixel can collect more photons and therefore deliver more electrons of signal. Noise characteristics, particularly read noise at low-light levels, determine how much the electron signal can fluctuate per pixel. The higher the signal over the noise, the higher the signal-to-noise ratio and therefore image quality. There will be no sample detection if noise exceeds signal.

This technical note will focus on quantum efficiency and how it was made possible to increase sensitivity on CMOS cameras by increasing QE to an almost perfect, 95% through the process of back-illumination. An easily repeatable experiment is also outlined to evaluate camera sensitivity.

Quantum Efficiency

Quantum efficiency can be defined as the percentage of electrons produced from the number of incident photons. For example, if 100 photons hit a 95% quantum efficient sensor, 95 electrons would be theoretically generated. Likewise, if 100 photons hit a 65% quantum efficient sensor, 65 electrons would be theoretically generated.

This process is a property of the photovoltaic effect, where light energy (photons) incident on the silicon substrate of a pixel creates electron-hole pairs. These electrons are then read out by the device and converted into a digital signal that can be interpreted by a computer.

There are many conditions that affect the photovoltaic effect and thereby determine the number of electrons generated by a single photon. Of these, the two most important conditions are the absorption coefficient and the chemical and physical properties of the material on the sensor surface. As these conditions determine the number of electrons that can be generated by a single photon, they directly influence the quantum efficiency of the camera.

Absorption Coefficient

The absorption of photons into the silicon substrate of the pixel is wavelength dependent. This is the reason why quantum efficiency is shown on camera datasheets as a curve, such as the curves shown in Figure 1.

Figure 1: Quantum Efficiency / Wavelength graph
Figure 1: Comparison of quantum efficiency curves of some typical CMOS cameras.
Adapted from Princeton Instruments, Kuro sCMOS

Quantum efficiency is higher in the green and yellow region (500 nm – 600 nm) because these wavelengths penetrate well into the region of the silicon substrate of the pixel where the photovoltaic effect takes place (Figure 2).

Figure 2: Light wavelength and silicon penetration.
Figure 2: Light wavelength and silicon penetration.
Adapted from Sony, back-illuminated CMOS image sensor

Shorter wavelengths do not penetrate deep enough so many photons are lost before reaching the silicon substrate. At the other end of the spectrum, longer wavelengths penetrate too far so photons pass straight through the silicon substrate.

There is usually a quantum efficiency cut-off at around 400 nm where the majority of the photons are lost before they can reach the silicon substrate.

There is also a critical wavelength, usually at around 1100 nm, where incident photons have insufficient energy to produce an electron-hole pair so no signal can be generated.

The Sensor Surface

On CMOS sensors, a certain fraction of the pixel surface is covered in the metal tracks, wiring and transistors (the circuitry) necessary to collect and transport charge (Figure 3). This has the unfortunate side effect of making that area completely light insensitive.

Figure 3: Photons unable to pass the metal tracks, wiring and transistors present in front of the CMOS sensor.
Figure 3: Photons unable to pass the metal tracks, wiring and transistors present in front of the CMOS sensor.
Figure 4: The use of microlenses to increase quantum efficiency of CMOS sensors.
Figure 4: The use of microlenses to increase quantum efficiency of CMOS sensors.

The photons landing on this area can’t reach the silicon substrate because they are physically impeded. These photons, therefore, won’t be converted into electrons and so the quantum efficiency will be negatively affected. CMOS sensors using this architecture typically have a peak QE of 82%, so almost a fifth of the photons arriving at the pixel never make it to the silicon substrate.

The highest quantum efficiency sensors using this architecture was made possible through the addition of microlenses on the sensor surface (Figure 4). The microlenses are designed to focus the incident light away from the circuitry and onto the silicon substrate. This effectively increases the number of photons reaching the silicon substrate and therefore increases QE. CMOS Sensors using this architecture claim a QE of up to 82%.

A downside of microlenses, however, is that they are most effective when the incident angle of light is normal to the sensor surface. When light enters the sensor from any other angle, the effectiveness of the microlenses can become severely reduced. This means that the reported QE increase of a CMOS camera with microlenses may not accurately reflect the real QE increase.

Regardless of the issues with microlenses, the real problem to overcome is clearly the position of the circuitry. To address this, sensor manufacturers have recently started creating back-illuminated CMOS sensors. By inverting the sensor and bringing light in from the back, the circuitry can be avoided completely.


Back-Illumination

A back-illuminated sensor is one that has essentially been flipped over so light enters directly into the silicon substrate rather than having to pass through the circuitry (Figure 5). Any light loss due to objects on the sensor surface is thereby eliminated.

Figure 5: Comparison of front and back illuminated sensors
Figure 5: Comparison of front and back illuminated sensors.

To allow the photons to penetrate deep enough into the silicon substrate to be converted into electrons, the silicon must also be thinned at the back. For this reason, a back-illuminated sensor may also be referred to as a back-thinned sensor.

The result is a sensor with an almost perfect, 95% quantum efficiency at its optimum wavelength. This can be seen in Figure 1, which shows the QE curve of the Prime 95B back-illuminated CMOS compared to front-illuminated CMOS devices, such as the 82% CMOS camera with microlenses.

Another advantage of back-illumination, highlighted in Figure 1, is the ability to achieve a high QE with shorter, UV wavelengths of light. It’s possible to achieve UV light detection on other CMOS devices but, as stated earlier, the fill- factor becomes limiting. This problem is completely overcome with a back-illuminated sensor. Moving the circuitry below the silicon substrate allows the fill-factor to reach 100%, granting both a high UV response and a high QE over a wide spectral range.

To test the theoretical increase in quantum efficiency of a back-illuminated sensor over a front-illuminated sensor, we performed a simple experiment designed to compare the sensitivity of multiple cameras.

Experimental Analysis of the Sensitivity of a Back-Illuminated CMOS Camera

Experimental Design

To demonstrate the sensitivity of a back-illuminated camera over a front-illuminated camera, experimental analysis was performed. The nature of biological samples often renders them unreliable for accurate measures of sensitivity, for this reason, a known standard was preferred. To this end, experiments were performed using the Argo-HM slide from Argolight. This slide is designed for calibrating and monitoring fluorescence systems through the use of stable fluorescence patterns of known size and fluorescence intensity.

One of the patterns on the Argo-HM slide is specifically designed for quantifying sensitivity (Figure 6). This pattern consists of two rows of 16 lines of increasing and decreasing intensity.

Figure 6: Intensity gradations on the Argolight Argo-HM slide using the Prime BSI camera at 60x magnification
Figure 6: Intensity gradations on the Argolight Argo-HM slide using the Prime BSI camera at 60x magnification.
The pattern consists of 16 lines of 22.5 µm x 1.5 µm rectangles having different fluorescence intensity levels following a linear evolution.

Using this slide, it’s possible to accurately quantify the sensitivity of a camera. Firstly, by observing the lowest intensity line that can be detected and secondly, by calculating the number of detected electrons at each line.

The back-illuminated CMOS (Prime BSI - 95% QE, 6.5 µm pixels) and the front-illuminated CMOS (82% QE, 6.5 µm pixels) were connected to a microscope with a Cairn TwinCam 50/50 image splitter to acquire images simultaneously on both cameras. The Argo-HM slide was brought into focus and parfocality was checked and corrected.

Images were acquired on both cameras with a 60x, 1.35 NA oil objective, excited with 450 nm LED light. The two CMOS cameras have the same 6.5x6.5 µm pixel size to control for sensitivity differences resulting from a larger pixel.

The data was analysed in ImageJ. 100 frames of raw data were uploaded and averaged then mathematically corrected for camera bias and gain to convert grey level signal into electron signal. This is performed using the following equation:

$$ \mathrm{\textit{Signal in electrons}} = \mathsf{(\text{Signal in grey levels} - \text{Bias})} * \textit{Gain} $$

Camera bias is the background offset applied to every scientific camera that gives every pixel a non-zero value and must be removed from every calculation. Camera gain represents the conversion factor of electrons to grey levels which occurs when detected electrons are converted on the camera into a digital signal to be read by a computer. Unlike grey levels, electrons are a quantifiable unit of signal measurement and should be used for sensitivity comparison. The bias and gain values of the cameras compared were are as follows:

Back-illuminated CMOS (Prime BSI): Bias = 100, Gain = 0.67
Front-illuminated CMOS (82% CMOS with microlenses): Bias = 100, Gain = 0.45

Camera gains were calculated using mean-variance analysis immediately before data collection.

Sensitivity Comparison of the Prime BSI and Front-Illuminated CMOS

The expected sensitivity increase of a 95% quantum efficient camera over an 82% quantum efficient camera would be expected to be ~1.15x increase in signal:

$$ \mathsf{95 \over 82} = \mathsf{1.15} $$

To test this, the Prime BSI and front-illuminated CMOS were compared under the same conditions, using the same exposure time (Figure 7).

Figure 7: Comparison of Prime BSI and front-illuminated CMOS signal with an 80 ms exposure time
Figure 7: Comparison of Prime BSI and front-illuminated CMOS signal with an 80 ms exposure time.
Top Argolight slide intensity gradation patterns imaged on both cameras
Bottom Box profile across the region marked in yellow on the intensity gradation pattern showing detected electron signal. The Prime BSI detected a peak 563 e- and the front-illuminated CMOS detected 442 e-.

The data presented in Figure 7 shows that, when using the same exposure time, the back-illuminated Prime BSI detected more electron signal than the front-illuminated CMOS. This result was expected with the increase in QE. However, the difference in peak signal is greater than the 1.15x increase that would be expected of the difference between an 82% QE sensor and a 95% QE sensor:

$$ \mathsf{563 \over 442} = \mathsf{1.27} $$

The actual result suggests a 1.27x increase in signal, effectively an above 25% increase in detected electrons on the Prime BSI. To confirm this, we also tested the Prime BSI with a 25% lower exposure time (Figure 8).

Figure 8: Comparison of Prime BSI and front-illuminated CMOS signal with a 60 ms exposure time on the Prime BSI and 80 ms exposure time on the front-illuminated CMOS
Figure 8: Comparison of Prime BSI and front-illuminated CMOS signal with a 60 ms exposure time on the Prime BSI and 80 ms exposure time on the front-illuminated CMOS.
Top Argolight slide intensity gradation patterns imaged on both cameras
Bottom Box profile across the region marked in yellow on the intensity gradation pattern showing detected electron signal. The Prime BSI detected a peak 441 e- and the front-illuminated CMOS detected 442 e-

The data presented in Figure 8 confirms the result seen in Figure 7. Reducing the exposure time on the Prime BSI by 25% to 60 ms results in equivalent detection to the front-illuminated CMOS at 80 ms.

These results successfully demonstrate the sensitivity increases made possible through CMOS back-illumination. Exposure times could be reduced by up to 25% and retain the same level of detection. This is great news for live cell biologists who want to image faster and reduce the impact of photobleaching and photodamage to their samples.

This result also suggests that the quantum efficiency claim of 82% on the front-illuminated CMOS may be overstated and is, in fact, closer to 75%:

$$ \mathsf{95 \over 75} = \mathsf{1.26} $$

The reason for this overestimation of QE may be due to the angular dependency of the microlenses discussed previously.

Conclusion

Camera sensitivity is determined by quantum efficiency, pixel size and noise characteristics. Increasing pixel size comes with the disadvantage of reducing resolution so increasing quantum efficiency is a more attractive method of increasing camera sensitivity. Recently, back-illuminated CMOS sensors have been developed which allow quantum efficiency to reach up to 95%. An analysis of the back-illuminated Prime BSI CMOS camera compared to a typical front-illuminated CMOS camera was performed which confirmed the increase in sensitivity. We conclude that using the Prime BSI, which has an equal pixel size to typical front-illuminated CMOS cameras, would allow exposure times to be reduced by up to 25%.

References

Argolight (http://argolight.com/)
Cairn Research (https://www.cairn-research.co.uk/)
Princeton instruments ( https://www.princetoninstruments.com/ )