Avoid image degradation in digital camera designs
Article By : Richard Crisp
A closer look at the 3T Pixel architecture, examine some undesirable characteristics and then introduce the charge transfer pixel mechanism and see how it can be used to avoid many of these limitations...
In part 3, we learned how the basic 3T Pixel architecture works and is used in arrays to develop CMOS image sensors. We learned why the electronic rolling shutter has different exposure start and stop times for every scanline in the sensor, leading to focal plane distortion when motion is present in the scene.
We also learned how the 3T Pixel can be used without focal plane distortion by altering the timing of the reset and readout of each scanline and combining with either a synchronized pulsed light source or a mechanical shutter in progressive scan mode.
We also learned that the charge in each pixel that forms the image must be dynamically stored within the pixel for up to almost a full frame time when used in this mode. However, due to naturally occurring thermally generated charge, noise can increase in the image when using the progressive scan mode.
In this article, we will a closer look at the 3T Pixel architecture, examine some undesirable characteristics and then introduce the charge transfer pixel mechanism and see how it can be used to avoid many of these limitations.
Issues with 3T Pixel
In its most basic form, a column in a CMOS image sensor is made from a collection of pixels arranged in a vertical column and sharing a common video line. This video line carries a raw analog video signal from the selected pixel to signal processing circuitry comprising a sample-and-hold amplifier and an analog-to-digital converter (ADC).
Figure 1 The signal-processing circuitry used in a CMOS image sensor design. Source: Etron
In its simplest mode of operation, the exposure time is set by the frame rate. Therefore, to measure the signal, the reset level and video level are sampled for each frame. Because of the way the amplifiers work in Figure 1, when the video is sampled, the reference reset level available is only after the immediately-following reset pulse, as shown in Figure 2.
Figure 2 A pixel operation timing diagram showing the video sampling. Source: Etron
However, the reset level sampled at that time is the reset level for the next frame versus the reset level for the frame just captured. So long as each reset operation provides the same signal level on the sense node, this is not an issue. Unfortunately, that is not the situation.
As will be shown in the next section, the reset level can change frame to frame and pixel to pixel.
Pixel reset level
Inside the pixel is a transistor used to precharge/reset the sense node by driving the node high when the reset signal is pulsed. The transistor is used in a source follower configuration. Unless the reset signal is driven higher than Vdd+—the threshold voltage of the reset transistor—the sense node reset voltage will be less than Vdd.
Figure 3 The circuit simulation shows a source follower charging a capacitive load with modeled photodiode leakage. Source: Etron
As can be seen, the high level on the sense node is lower than the Vdd level (1.5 V) and the amount of time it takes for the signal to asymptotically reach the final signal level can be long. Finally, since light continues to shine on the sensor, there’s a current load on the reset transistor that results in a voltage divider formed by the reset transistor’s drain-source resistance in series with the equivalent resistance of this photo-induced current load.
Additionally, depending on how discharged the sense node was after the previous exposure, and how long is allocated to reset the node, additional variation of the reset voltage is observed frame to frame. The combined effect of these factors is known as incomplete reset. It can cause increased noise in images, leaving a remnant of prior exposures in a subsequent image, which is one cause of image lag.
Due to random semiconductor process variations, slight differences in operating current of otherwise identical transistors is also observed, leading to an overall spatial and temporal variation of pixel reset values.
Figure 4 The simulation shows variation of sense-node reset levels in response to prior exposure. It also provides an example of image lag. Source: Etron
The incomplete reset problem can be solved by using a reset pulse that has a higher voltage to compensate for the transistor threshold. In that case, a 3.5-V reset signal is used to provide a hard reset, clamping the sense node to the power supply voltage, 1.5 V in this case.
For the temporal reset problem, the charge transfer pixel can allow the reset level of the current exposure to be used when reading out the pixel’s video signal by adding to a 3T pixel a transfer gate to separate the sense node from the integration node as well as a reset transistor for the integration node’s photodiode.
Charge transfer pixel
Charge transfer pixel has four main functional parts to it: a reset circuit, a light capture region, a charge transfer switch, and a sense/readout circuit. The photodiode/light sensitive region is reset separately from the sense node. During normal operation, the sense node is read twice per pixel: once for the reset value and once for the video.
Usually, the sense node capacitance is smaller than the photodiode’s capacitance, so a given amount of charge transferred through the transfer gate to the sense node will make a larger potential change than in the photodiode, as shown at the bottom in Figure 5.
Figure 5 The saturation level for the pixel usually is set by the sense node capacity unless specifically designed otherwise. Source: Etron
The structure allows the sense node to be read while integration is occurring. Usually the sense node is reset, read out and then the transfer gate is opened, allowing charge to rush in from the photodiode, where it can be measured as illustrated at the bottom of Figure 6 and Figure 7.
Figure 6 The 5T charge transfer type pixel is shown along with an equivalent circuit and a potential/charge motion. Source: Etron
In Figure 7, a timing diagram is showing how the charge transfer pixel can capture the reset level for the same exposure as the video. As the left-most part of Figure 7 shows, the sense node is first reset while the pixel is selected. The readout source follower in the pixel impresses this reset level on the analog video signal passing through the column.
Figure 7 The timing diagram shows how the charge transfer pixel can capture the reset level. Source: Etron
Then the transfer gate is pulsed and the charge from the ongoing integration is transferred to this freshly-reset sense node, where it’s also read out immediately via the same analog video path through the array where it too is sampled. The differenced video is then amplified and converted to a digital number completing the readout of a given pixel. The photodiode is then reset and a new integration begins.
In the event of an incomplete sense node reset, any charge remaining from the previous exposure will be included in the reset level used for the current exposure.
Figure 8 The reset level is stored by the sample-and-hold amplifier. Source: Etron
Figure 8 is a summary of the operation of the array using a 5T pixel in a rolling shutter mode. If combined with a synchronized pulsed light source, focal plane distortion can be avoided if the light is pulsed during the exposure time and not during the beginning or ending part of the exposure. When used in this way, the beginning and end of the photon capture is set by the light or a mechanical shutter, thereby avoiding the undesirable rolling shutter artifacts.
The basic array global timing is similar to the 3T pixel array; each line is reset and read out at a different time. The array is scrolled through top to bottom and then repeats. Each pixel in each line is read out and then the line progresses to the next. But the charge transfer pixel works a little differently than the 3T architecture.
When reading the 5T pixel’s video level from the photodiode, the sense node is first reset and read out and the analog level sampled and held. Next the transfer transistor is switched on and charge from the photodiode flows onto the sense node which is read out, sampled and amplified using the reset level as the reference. It can be seen in the simulations shown in the upper middle plot in Figure 9.
Figure 9 The lower left plot shows a larger time-slice, highlighting the operation of two rows of pixels in several cycles. Source: Etron
It’s clear that the photodiode integrations begin and end at different times for different rows, arising from the operating method of the rolling shutter. The key advantage offered by the 5T Pixel architecture in this case is reset noise minimization, which can offer significant SNR benefit under low-light exposure regimes.
Figure 10 A high-level view illustrating how the array is cycled in rolling shutter mode. Source: Etron
Figure 10 shows when the beginning and end time of the exposure of each row are not time-aligned row-to-row, leading to the focal plane distortion. It shows how during a single frame readout, a video and a reset signal are both read out of each pixel in the array.
A key advantage of this architecture is that it is feasible to perform the differencing of the analog video and analog reset level on-chip in analog circuitry before the A/D conversion. Different circuits can be used to sample and amplify the analog video data from the pixel array. It’s also common to use multiplexing circuits to permit a single A/D converter to be shared among a group of columns.
Global snap shutter
As the name implies, the global snap shutter begins and ends the exposure of all pixels time-aligned. The 5T Pixel can be used to implement a global snap shutter mode of operation. There are different schemes for achieving the snap shutter operation and one is illustrated below.
In this scheme, the pixel photodiode array is globally reset at the beginning of a frame. Near the end of the frame exposure, all sense nodes are reset and read out, digitized and stored in memory, as shown in the top diagram in Figure 11. At the end of the exposure, the transfer gate for all pixels is pulsed, transferring image charge from all photodiodes to their corresponding sense nodes simultaneously. At that time, the pixel photodiode array is globally reset, another exposure begins and the sense nodes holding charge from the previous image are read out, digitized and stored in memory.
Figure 11 A high-level view of global snap shutter operation. Source: Etron
Note that the start and stop times of the pixels are time-aligned for each pixel row. This eliminates the focal plane distortion of the rolling shutter mode; however, it comes at the expense of a more complicated pixel and the need for digital memory.
The image is created by differencing the video data with the reset level image previously stored in memory. In this method of operation, the worst-case analog charge storage time for the sense node reset level readout and for the integrated signal charge is one frame time. During this charge storage time, the image charge is contaminated by thermally generated charge as explained in part 3 of this article series. Once again, low light exposure regimes are the first to be degraded by charge generated thermally within the pixel array.
Figure 12 A high-level view of how the operation of a global snap shutter pixel array is constructed using 5T Pixel architecture. Source: Etron
As noted in the simulations in Figure 11, the exposure’s start and stop time for all pixels is time-aligned. During each frame readout, both a reset image and a video image frame are read out in digital form to be differenced by a computer creating the final image. Charge is dynamically stored within the pixel array for both the reset level and for the video signal during readout for as long as a frame time in the worst case. This can lead to image degradation caused by dark signal inside the image sensor.
Motion artifacts arising from the rolling shutter can be avoided without the need for a pulsed synchronized light source of a mechanical shutter using this electronic shuttering method. For motion video, the global snap shutter has many attractive features.
In this article, once again we touched on noise, mentioning image degradation due to thermally generated charge inside the image sensor. We looked in detail at the operation of the 5T Pixel architecture in rolling shutter and global snap shutter array configurations. The next article in this series will look at noise in more detail.
This article was originally published on EDN.
Richard Crisp is VP of New Product Development for Etron Technology America.
Other articles in this series:
- Engineering a digital-camera system
- Digital camera design: Capturing images that include motion
- Designing image sensors for a digital camera
Related articles:
- Image sensor defects can ruin an image
- Modern ADCs improve CMOS image sensors
- Challenges for designers of digital-camera audio subsystems
- Powering image sensors in automotive camera applications
- Under the Hood: Sensor, processing at heart of DSLR
- Teardown: Blink XT security camera
- Teardown: High-quality and inexpensive security camera