It was mainly economic reasons that prompted Sony to discontinue its CCD production lines and complex CCD manufacturing in 2015. In terms of quality, CMOS sensors have already overtaken CCD technology in some areas - with low noise, high sensitivity and high frame rates, they achieve comparable or better values even for demanding applications.

The evolutionary development of CMOS sensors is based on the pixel readout mechanism. Each area sensor consists of a matrix array with photodiodes in which the incident photons are converted into electrons. In CCD sensors, the charges of the individual photodiodes are conducted via horizontal and vertical shift registers to a readout amplifier outside the active area. There, the pixel charges are read out centrally and converted to an analog voltage. CMOS technology, on the other hand, reads each pixel directly via transistors located on the pixel. The signal is converted via the read-out circuit, digitized with low noise and finally transmitted in parallel via LVDS wires (Low Voltage Differential Signaling).

Pixel structure with frontside illumination versus backside illumination: The backside illumination and microlens arrays increase the sensitivity of the CMOS sensors.

© Sony/Framos

For a long time, the decentralized readout of CMOS pixels was a disadvantage: the exact readout value at the pixel depends on the physical characteristics of the pixel-individual transistors. However, as these always exhibited slight differences in the earlier history of CMOS, a high level of 'fixed pattern noise' was generated. The increased background noise also resulted in poor sensitivity. In addition, the transistors were relatively large compared to the pixel size. The need for three transistors to realize a rolling shutter and five to six transistors for a global shutter led to a reduction in the active pixel area and thus to a significant reduction in light output. The high space requirement of the transistors was at the expense of image quality and sensitivity. For a long time, the focus of development was therefore on fast rolling shutter sensors. However, this was precisely the limitation of CMOS technology at the time and, to some extent, of image processing for machine vision applications - because industrial production and applications in the automotive sector, medical technology and many virtual reality applications require high-quality global shutter sensors without artefacts for fast-moving processes in real time and outstanding image quality.

Global instead of rolling shutter

The decisive step for CMOS development was the combination of image capture based on the global shutter principle and significantly increased speed and image quality. With the development of the first CMOS global shutter generation and the IMX174, Sony laid the foundations for the CMOS era in the industrial sector. In contrast to the rolling shutter process, the simultaneous exposure and emptying of all pixels means that there is no time delay between the pixel lines. Fast-moving objects can be captured simultaneously with global shutter without artifacts and with high sharpness.

Based on their architecture and the parallel readout of all pixels, CMOS sensors are many times faster than conventional CCD sensors. Whereas previously the frames per second (fps) that could be achieved limited the application, very high frame rates are possible with CMOS technology; the main limitation for maximum speed is now the interfaces.

In addition, the progressive miniaturization of transistors has enabled a greatly improved ratio between sensor size and active pixel area. At the same time, the use of incident light has been maximized. This means that CMOS sensors can also be used in low-light applications. And in the opposite case, with extremely high incident light, for example in direct sunlight or reflections, CCD pixels emit excess converted electrons to their neighboring pixels (overflow) or pass them on even after exposure has already taken place. This results in negative blooming or smearing effects for image analysis. CMOS sensors, on the other hand, do not allow such effects thanks to their direct pixel readout. Technical features such as the backside illumination of the sensor and microlens arrays mounted on the chip to amplify the incident light also mean that CMOS sensors achieve higher sensitivities compared to CCD sensors.

The third CMOS generation

After Sony achieved further improvements in the pixel architecture in the second global shutter generation and implemented additional functions relevant to machine vision applications, there is one major innovation in particular in the current third generation: SLVS-EC (Scalable Low Voltage Signaling with Embedded Clock) introduces a sensor interface that meets the ever-increasing demands for resolution and speed. According to industry insiders, these sensors can achieve twice the bandwidth of the second-generation sub-LVDS-based sensors.

The first generation Sony CMOS global shutter set a first standard for the machine vision market with the IMX174. Suddenly, applications could be realized with an image quality and speed that had been unthinkable with CCD technology. With a pixel area of 5.86 µm, a very high saturation of 30,000 e- was achieved, so that the dynamic range already reached 75 dB 'despite' the still high readout noise of 5 e-. In the second generation, Sony has focused on the requirements of the machine vision sector: The additional bit depth of 8 bits requires less bandwidth, the double number of channels results in a doubling of the output speed to 9.5 Gbps. Functions such as additional trigger modes have also been integrated. With a reduced pixel size of 3.45 µm, the saturation is reduced to 11,000 e-, while at the same time the readout noise, which has been reduced to 2 e-, ensures a consistently high dynamic value of 74 dB.

The innovations of the third generation will bring improvements, which should have a particular impact on image quality and speed. Applications with moving objects in particular, such as moving production lines, robotics applications and in the ITS and automotive sectors, will benefit from an increase in performance thanks to improved detection quality. It is expected that the further increase in pixel size will result in a significant increase in saturation compared to the second generation and will almost reach the values of the first generation. In combination with low readout noise, the maximum dynamic range may even reach a new high compared to the previous generations. This means that improved light/dark detection can also be implemented in difficult lighting conditions. However, as it would no longer be possible to transfer the increased image data at a higher speed with previous standard interfaces based on the various quality improvements, Sony has developed the SLVS-EC standard with eight channels.

As integrated circuits, CMOS sensors are relatively easy to design and install in image processing systems. However, the challenge for the application lies less in the actual installation than in achieving the best possible image quality.