Comparison of CCD and CMOS image sensors (CMOS Sensor principle)

Infineon / Mitsubishi / Fuji / Semikron / Eupec / IXYS

Comparison of CCD and CMOS image sensors (CMOS Sensor principle)

Posted Date: 2024-01-28

The photosensitive element used by the camera to create images is called Image Sensor or Imager. The two types of Image Sensors currently widely used are CCD and CMOS Image Sensor (CIS).


CCD technology has been widely used in the digital camera market, but CCD requires a large amount of energy and a considerable number of supporting chips. So CMOS Image Sensor came into being. Each of its pixels can perform its own charge conversion, significantly reducing the energy and supporting circuitry required to produce an image. In addition, CMOS Image Sensor is manufactured using the same materials and technologies as most microprocessors and memory chips, making it easier to manufacture and more cost-effective, so it is widely used in mobile phones and other mobile products. Here we will mainly introduce the CMOS Image Sensor in detail.

1 CMOS Sensor Principle

CMOS is the abbreviation of Complementary Metal Oxide Semiconductor in English. This is a mainstream semiconductor process with the advantages of low power consumption and high speed. It is widely used in manufacturing CPUs, memories and various digital logic chips. The image sensor designed based on the CMOS process is called CMOS Image Sensor (CIS), which is more than 90% similar to the general semiconductor process, especially the memory process.

The CMOS image sensor uses CMOS switches to switch the signals of each photodiode. As shown in Figure 1, each pixel is composed of a photodiode and a CMOS switch. When the row driving circuit selects a certain row, the CMOS switch of the pixel output signal of this row is turned on, and the CMOS switches of the other rows are not turned on. The photoelectric signal of the pixel of this row is transmitted to the column selection multiplexer. The column select multiplexer is also composed of a series of CMOS switches, which sequentially connect the signals of the rows to the output terminal to complete the sequential reading of the signals of a column.

Figure 1 Principle of CMOS image sensor 2 CCD image sensor

A CCD device is essentially an analog shift register. The principle is to transfer the signal charge from one potential well to another potential well under the control of the clock, thereby realizing the shift transfer of the analog signal. The CCD image sensor utilizes the signal transfer function of the CCD device to implement electrical scanning. Figure 2 shows the working diagram of the CCD image sensor. The signal multiplexing consists of two parts: vertical transfer CCD device and horizontal transfer CCD device. In an actual CCD image sensor, the photoelectric conversion device and the vertical CCD device are integrated together. The vertical CCD device sequentially transfers the photodiode signals of each row to the horizontal CCD device; and the horizontal CCD device transfers the signal of this row to the output terminal. Repeat the above process to form an image signal.

Figure 2 CCD image sensor principle

3 Comparison of CCD and CMOS image sensors

The main difference between CCD and CMOS is shown in Figure 3 below. CCD devices usually have only one charge-voltage converter (Charge-Voltage Converter). When the Sensor reads out pixel data, the charges accumulated in each row of pixels need to be controlled by the row voltage. The next step "creeps" to the next row until it finally reaches the row buffer (Row Buffer) to which the array belongs, and then begins to "creep" step by step to the charge-voltage converter at the exit of the array under the control of the column voltage to complete the readout. process. A major advantage of CCDs is that all pixels share the same charge-to-voltage converter, so pixel uniformity is very good. In contrast, each pixel of CMOS has its own dedicated charge-to-voltage converter, and the consistency is difficult to control. When the number of CCD pixels exceeds 2 million, sharing a charge-voltage converter for all pixels will seriously affect the readout speed, so at this time, consideration will be given to designing the pixels into two or four arrays, with each array equipped with a dedicated line buffer. and charge-to-voltage converters that speed up readout exponentially.

Figure 3 The main differences between CCD and CMOS

4 CMOS Sensor pixel structure

Passive Pixel

The simplest Pixel structure has only one PN junction as the photosensitive structure, and a reset transistor (RS) connected to it as a switch, as shown in Figure 4:

Figure 4 Single PN node image passive element structure

The Passive Pixel structure works as follows:

Before starting the exposure, the row selection address of the pixel will be powered on, so RS is enabled, connecting the PN junction and the column selector (Column Bus). At the same time, the column selector will be powered on, causing a high reverse voltage to be added to the PN junction (such as 3.3 V), after a short delay, the electron-hole pairs in the PN junction reach equilibrium, so the reset operation is completed, the RS signal becomes invalid, and the connection between the PN junction and the columnbus is cut off.

When exposure begins, the silicon in the PN junction absorbs photons and excites electron-hole pairs. Affected by the electric field in the PN junction, electrons will flow to the n+ end of the PN junction, and holes will flow to the p-substrate of the PN junction. Therefore, the reverse voltage of the PN junction will decrease after exposure.

After the exposure, RS is enabled again, and the readout circuit will measure the voltage in the PN junction. The difference between this voltage and the original reverse voltage is proportional to the number of photons received by the PN junction.

After reading the photosensitive signal, the PN junction will be reset again to prepare for the next exposure.

When RS is enabled and the column selector is passed to high level, in circuit principle it is equivalent to charging the capacitance of the PN junction, but the voltage value obtained after charging has a certain degree of randomness. On the one hand, the actual capacitance of each PN junction The size will obey a certain probability distribution, and there will be a fixed deviation between junctions, which will constitute a fixed pattern noise (FPN); on the other hand, due to the existence of dark current noise in the circuit, even the same junction The actual voltage obtained after each charge is not exactly the same, which constitutes another mode of noise, which is related to the structure, temperature and junction capacitance of the PN junction, and is called kTC noise.

Active Pixel

Currently, mainstream CMOS sensors adopt Active Pixel structure design. The Active Pixel structure shown in Figure 5 is called a 3T structure. Each pixel contains a photosensitive PN junction and three transistors, namely a reset tube RST, a row selector RS, and an amplifier SF.

Figure 5 3T active pixel structure

The Active Pixel structure works as follows:

reset. Enable RST to load the reverse voltage to the PN junction, and cancel RST after the reset is completed.

exposure. The same principle as Passive Pixel.

read out. After the exposure is completed, RS will be activated, and the signal in the PN junction will be amplified by SF and read out.

cycle. After the signal is read out, it is reset, exposed, read out, and image signals are continuously output.

Active Pixel based on PN junction was popular in the mid-1990s, and it solved many noise problems. However, the kTC noise introduced by PN junction reset has not been solved.

PPD structure

The emergence of PPD (Pinned Photodiode Pixel) is a huge breakthrough in CMOS performance. It allows the introduction of correlated double sampling (CDS) circuits, eliminating kTC noise introduced by reset, 1/f noise and offset noise introduced by op amps, as shown in the figure 6 shown.

Figure 6 PPD structure

The PPD structure works as follows:


reset. Enable RST at the end of the exposure to reset the readout area (n+ area) to high level.

Read the reset level. Read the level of the n+ area, which includes the offset noise of the op amp, the 1/f noise and the kTC noise introduced by the reset, and store the read signal in the first capacitor.

Charge transfer. Enable TX to completely transfer the charge from the photosensitive area to the n+ area to prepare for reading. The mechanism here is similar to the charge transfer in CCD.

Read the signal level. Read the voltage signal in the n+ region to the second capacitor. The signals here include: signals generated by photoelectric conversion, offset generated by the op amp, 1/f noise, and kTC noise introduced by reset.

signal output. The signals stored in the two capacitors are subtracted (if CDS is used, the main noise in the Pixel can be eliminated), the resulting signal is analog amplified, and then sampled by the ADC, and then the digital signal can be output.

The PPD pixel structure has the following advantages:

The kTC noise of the readout structure (n+ region) is completely eliminated by CDS.

The offset and 1/f noise of the op amp will be significantly improved by CDS.

The kTC noise caused by the reset of the photosensitive structure no longer exists due to the total transfer of PPD charges.

Photosensitivity, which directly depends on the width of the depletion region. Since the depletion region of PPD extends to near the Si-SiO2 interface, the photosensitivity of PPD is higher.

Due to the double-junction structure of pnp, PPD has a higher capacitance and can produce a higher dynamic range.

Since the Si−SiO2 interface is covered by a layer of p+, the dark current is reduced.

Correlation Doubly Sampling (CDS)

Correlated Double Sampling, the basic idea is to perform two samplings, first sample a reference signal to evaluate the background noise, then collect the target signal after a short delay, and subtract the reference signal from the second sampling. The target signal with most of the background noise removed is obtained, and its principle model is shown in Figure 7 below.

Figure 7 Denoising principle model

The condition for CDS to be established is that the amplitude of the background noise does not change much between two samplings, so it is ideal for removing fixed noise (FPN) and low-frequency noise, such as 1/f noise, kTC noise, etc.

Figure 8 CDS circuit model

5 CMOS Sensor characteristics

The essence of a CMOS Sensor is a linear sensor that measures photoelectric conversion events. In a certain sense, it can be said to be a photon counter. The reading value of each pixel on the Sensor reflects the number of photons captured by the pixel within a specified time. An ideal Sensor should have the following characteristics:

The output is always proportional to the input (no Sensor noise, only the noise of the signal itself)

Input and output can be infinite

High sensitivity, small input stimulates large output

high frame rate

high resolution

Low power consumption

Simple process

low cost

The response characteristics of an ideal CMOS Sensor are shown in the figure below:

Figure 9 Ideal sensor response characteristics

The slope of the straight line in Figure 10 determines the response size that a unit input can excite. This slope is called the gain coefficient (Gain). Sensor will provide a set of interfaces for adjusting the actual gain value.

Figure 10 Ideal Sensor response characteristic curve

The actual Sensor can only maintain a linear response within a limited interval, and it will not be able to faithfully represent input signals with too small or too large amplitudes.

Figure 11 Actual sensor response characteristics

6 CMOS Sensor Noise (Noise)

Assuming constant, uniform illumination intensity, the noise in the image captured by the camera is the sum of spatial and temporal vibrations in the measured signal. The figure below summarizes the CMOS Sensor optical and electrical conversion model and the mathematical models of several major noises in the form of transfer functions.

Figure 12 Transfer Noise Model Figure 13 below describes in more detail the sources and locations of various noises during the CMOS Sensor imaging process.

Figure 13 Source paths of noise

Dark Noise

The thermal movement of electrons in the silicon chip will cause some valence electrons to be randomly excited into the conduction band to form a dark current (Dark Current), so even if no photons are incident at all, the Sensor will have a certain signal output. During the exposure process, random changes in dark current form dark shot noise. The main reason for the change in dark current is that when electrons pass through the PN junction, they encounter the potential barrier (Darrier) of the PN junction. Electrons need to undergo a conversion process of kinetic energy-potential energy-kinetic energy to cross the barrier, so it takes some time. Dark shot noise statistically obeys the Poisson distribution and has nothing to do with the level of the light signal, but is related to the temperature of the sensor. The general rule is that the dark current doubles every 8°C increase in temperature. Therefore, when designing the circuit, attention must be paid to arranging the electronic components that are prone to heat as far away from the Sensor as possible.

Figure 14 Dark current changes with temperature

read noise

This noise is generated when electronic signals are generated. The Sensor uses an AD converter (ADC) to sample the analog voltage output from the analog amplifier into a digital voltage. Since the accuracy of digital signals is always limited, usually 10 bits to 14 bits, analog signals with amplitudes between two adjacent numbers will be rounded to the nearest value, so this process will introduce quantization noise, which is the read an important part of the noise. This noise is determined by the design of the sensor, which means at least how many electrons are needed to drive the ADC of the readout circuit to change one bit. It is independent of signal high and low levels and sensor temperature.

Photocurrent Noise (Shot Noise)

This noise is statistical noise associated with photons falling on the sensor pixels. At the microscopic scale, the behavior of the photon flow reaching the sensor is non-uniform in time and space, just like counting the traffic flow on a highway. Sometimes the traffic flow is dense, and after a while it becomes sparse. Sometimes the lanes on the left are dense, and after a while the lanes on the right are dense. The overall statistical pattern conforms to the Poisson distribution. Photon shot noise is related to the level of the measured signal and has nothing to do with the sensor temperature.

Fixed Pattern Noise (FPN)

This noise is caused by the spatial non-uniformity of the pixels. Each pixel of the CMOS Sensor is configured with a charge voltage amplifier. There are some transistors in each row and column to control the reset and readout of the pixels. The operating parameters of these devices The drift relative to the theoretical value constitutes a fixed pattern noise. In addition, bad pixels and defective pixels can also be regarded as a kind of fixed pattern noise, and its effect can be roughly simulated using the schematic diagram 15 below.

Figure 15 FPN model

reset noise

The rolling shutter exposure method requires resetting the potential well first to release all the charges freely accumulated in the potential well to prepare for subsequent readout. However, due to the existence of dark current, some noise signals of random size will remain after each reset, that is, reset noise. Its size is related to the pixel structure, chip temperature, and PN junction capacitance, so it is also called kTC noise. It takes a certain amount of time to reset the pixels. Quantitative research shows that even if a larger reset current is used, it generally takes more than 1ms to completely release the charge, as shown in Figure 16 below.

Figure 16 Charge release curve

The actual reset control signal is usually shorter than 1ms, so the next frame image will more or less retain some shadow of the previous frame image. This afterimage is called Image Lag, which is also a form of noise.


In the field of communications, it refers to the signal coupling between two signal lines due to poor shielding. The signal on one line is fed to the nearby signal line through the mutual inductance and mutual capacitance existing between the cables. In the era of analog communication, it may cause hearing loss. to someone else's call. In the sensor field, crosstalk refers to the fact that the light signal incident on a pixel A is not captured in this pixel, but is captured by the surrounding pixel B, causing B to generate an unexpected signal.

In the example in Figure 17 below, pink represents opaque pixels, which should not have any output, and yellow represents normal pixels, which should have output. In fact, photons can penetrate a certain distance in the silicon chip, and thus have the opportunity to enter the photosensitive area of ​​the pink pixel, thereby turning into the signal of the pink pixel. This is the crosstalk mechanism of the CMOS Sensor.

Figure 17 Crosstalk model

As can be seen from Figure 18 below, the longer the wavelength, the more serious the crosstalk, and the crosstalk energy at some pixel positions can reach 5%.

Figure 18 Wavelength and crosstalk energy curve

7 CMOS Sensor process structure

Front-illuminated process (FSI)

A major shortcoming of the traditional FSI process is that several layers of circuit structures need to be manufactured between the photosensitive PN junction, the filter film and the microlens. Due to the circuit height problem, the area and angle of the PN junction that can collect light will be limited. At the same time, the light is moving forward. It will absorb and scatter with the circuit structure, so it will increase the loss of light energy, as shown in Figure 19 below.

Figure 19 FSI process structure

Backside illuminated process (BSI)

With the advancement of semiconductor technology, people have discovered that the wafer can actually be polished very thin, allowing light to penetrate the wafer and enter the photosensitive PN junction from the back. This idea has become feasible both technically and cost-wise, so the idea was born The backside illumination process (BackSide Illumination, BSI) is adopted.

Figure 20 BSI process structure

8 CMOS Sensor Progress Sony launched the first generation Exmor series CMOS image sensor in 2007. Compared with traditional CIS technology, the main feature of Exmor is a dedicated ADC and additional CDS configured for each column of pixels. Because the ADC unit is physically closer to the pixel, and because massive parallelization can reduce the operating frequency of a single device, the noise characteristics of the Sensor are greatly improved. The newly added CDS further suppresses digital noise.

Figure 21 Comparison of CDS improvements

With the further evolution of manufacturing technology, the stacked process has been developed based on the back-illuminated process. As the name suggests, the stacking process stacks two or more silicon wafers up and down. The uppermost silicon wafer is all used to manufacture the photosensitive area of ​​the pixel, and the analog and digital logic required for sensor control are all moved to the lower silicon wafer, so The ratio of the photosensitive area to the sensor target surface size can be close to 100%, finally reaching the peak of sensor efficiency.

Figure 22 Comparison of FSI, BSI, and Stacked structures

In fact, CMOS Sensor can also be designed to support Global Shutter exposure method. Similar to CCD, the implementation principle of Gobal Shutter is that each exposed pixel is accompanied by a storage capacitor. All pixels on the photosensitive array are exposed simultaneously, and then the photoelectrons are immediately transferred to the storage capacitor and locked, waiting for readout by the readout circuit. The picture below is a newer Global Shutter pixel design, which supports two different gain coefficients and therefore supports HDR functionality.

Figure 23 Global Shutter Pixel structure

DCG This mode was proposed by OmniVision. The principle is that each pixel can control the gain independently (using the CG signal in the picture below). When working in HDR mode, only one exposure is performed, but it is read out in two times, one using HCG (High Conversion Gain) to capture dark information, and once using LCG (Low Conversion Gain) to capture bright information.

Figure 24 DCG Pixel structure

Review Editor: Huang Fei

#Comparison #CCD #CMOS #image #sensors #CMOS #Sensor #principle