While designing image sensors for a digital camera, engineers can employ the 3T pixel array architecture in both rolling shutter and global reset modes.
Part 2 of this series on digital camera design looked at some of the issues associated with taking images of scenes that include motion, explaining the different types of mechanical shutters used to start and stop exposures and outlining different types of electronic shuttering available in CMOS image sensors. This article, Part 3 of the series, will explain CMOS shuttering in more detail.
CMOS pixel architecture basics
The three-transistor design or 3T Pixel is the simplest CMOS pixel architecture (Figure 1). One transistor is used to reset or precharge the photodiode while two more are used for video readout: one is a source follower and the other is used to select the pixel. The RESET signal is pulsed high and that precharges or resets the N-type electrode of the photodiode to a voltage set by Vdd and the high level on the RESET signal. After RESET is taken low, the photodiode begins to respond to incident light, draining the charge as photo-generated electrons are accumulated on the photodiode’s cathode.
Figure 1 In this 3T Pixel architecture, one transistor is used to reset or precharge the photodiode while two more are used for video readout. Source: Etron
Once the integration time has elapsed, the row select signal is driven high and the source follower current amplifier is coupled to the video node for sensing. Note how the pixel, if exposed to light, continues to collect light during this process. When being reset, light still shines on the photodiode but the reset transistor clamps the photodiode’s cathode to the voltage supply, Vdd.
To understand the operation of the pixel exposure and readout in a practical array, it is useful to examine a column of pixels on the sensor. For an M x N pixel sensor, N pixels are arranged in M vertical columns. In addition to the pixels, circuitry is usually added to each column to build a sensing and buffering scheme for the video data, including a way to sample-and-hold (S/H) the video data. An example of a pixel column is shown in Figure 2.
Rolling shutter operation
For rolling shutter operation, each line begins and ends its exposure at a slightly different time. In Figure 2, the video signal from the array can be read from only one pixel at a time per column; the pixel is selected and the video signal forms on the signal line labeled Video(m) in Figure 2. That signal is then sampled by the sample-and-hold circuit, amplified, and sent to an analog-to-digital converter (ADC) to be turned into a digital number and then driven off-chip as digital video.
The readout time for transferring the digital video from a single line off-chip is the same amount of time that the exposure from line to line is offset; only one line can be selected at a time and the line must be selected during the readout when using the rolling shutter scheme. The sensor is constantly exposed to light and there is no mechanical shutter.
The end of the reset pulse for a given line determines the beginning of the exposure and the exposure ends when the sample-and-hold signal is fired to capture the state of the Video(m) line after the exposed pixel is selected via its Select(N) line. Figure 3 shows how the operation of the pixel array for multiple lines is being exposed and sampled in rolling-shutter mode.
The image sensor is constructed using a group of columns, as shown in Figure 2, which are arranged such that there’s an N x M array of pixels (Figure 4). Several additional functional blocks are used to provide control and sequencing of the signals used to operate the pixel array, such as the reset and select signals, video sample-and-hold, and the multiplexers used to feed the ADCs that convert the analog video to digital video.
The output of the image sensor is digital video data and may be provided from a single ADC or multiple ADCs operating in parallel. With parallel ADCs, the readout time for each line can be reduced, leading to less exposure timing offset line to line—less distortion—as well as higher frame rates for the image sensor. This comes at the expense of power, silicon area, package pin count, and cost.
Global reset scheme
The 3T pixel array can be operated in a different mode by altering the timing of the reset signals used for each line or row. Instead of pulsing the reset signals for each row at a different time, this mode pulses all at the same time, a.k.a. global reset. The key benefit is that the exposure of each line begins at the same time. If a synchronized light source or a suitable mechanical shutter is used in a way that the exposure ends as the light is extinguished, then motion artifacts can be avoided or minimized to a tolerable level.
For very high pixel-count CMOS sensors with frame rate limited by low-speed digital links, the line-to-line exposure timing offset will increase as pixel count increases and can reach a point of being unsuitable for even snapshot photography if relying on a rolling shutter. For that reason, a high-speed mechanical shutter may be used in combination with a 3T pixel array operated with global reset.
When progressive scan readout is combined with global reset, the readout is performed one line at a time while the sensor is kept dark via synchronized light or mechanical shutter. Once the sensor has been read out, another exposure can begin. So even with slow line readout, as may occur with a very high pixel count sensor using a single digital link, action photography may still be supported by using a global reset combined with an appropriate mechanical shutter.
If the decision is made to use a mechanical shutter, the artifacts described in Part 2 should still be considered, given the application requirements as well as the shutter types and timing parameters.
Noise considerations for global reset
Since either a synchronized light transitioning from light to dark or a mechanical shutter closing ends an exposure for the entire array at the same time, and because the sensor is read out one line at a time, the image must be stored dynamically within unread pixels that are not illuminated during the readout process. Once a line is read, the stored image data within its pixels is no longer of interest; however, remaining lines that have not been read must store the image until read. The last line to be read holds the image the longest and that may be almost a frame time for short exposures.
This storage requirement imposed by global reset used with progressive scan readout can reduce low light sensitivity, thanks to non-ideal operational characteristics of real-world CMOS image sensors arising from the electronic properties of silicon.
In future articles of this series, noise sources will be discussed in detail, but a few words on the topic are appropriate at this point without a significant digression from the main topic of pixel operation in arrays.
Image degradation from dark signal
Within each 3T Pixel in the image sensor array is a reverse-biased photodiode exposed to focused light, which creates hole-electron pairs—photo-generated charge—within the illuminated region of the silicon. This mechanism for photo-generating charge is known as the photoelectric effect, discovered by Albert Einstein, for which he was awarded a Nobel Prize in physics.
Each reverse-biased photodiode can be thought of as a capacitor that can be used to collect and hold electronic charge. The amount of photo-generated charge held in each pixel is a function of the exposure time and brightness of the scene as focused on the pixel under study. Light interacting with the silicon photodiode is one way that charge can be generated and is the fundamental mechanism by which solid-state image sensors form images.
Another way charge is generated in silicon devices is by random thermal generation within the silicon itself and is not caused by exposure to light. For a given temperature, the dark current or units of charge/time is constant and is exponentially dependent on temperature. The dark current isn’t uniform over an image sensor surface due to imperfections in the silicon crystal structure and low concentrations of impurities. A three-dimensional plot of dark current across an image sensor surface exhibits significant areal nonuniformity (Figure 5).
The photodiodes collect and hold the charge generated by the dark current equally just like the photo-generated charge. Once this charge is collected in a pixel, it’s indistinguishable from photo-generated charge. The net result is that the signal-to-noise (S/N) ratio of the image is degraded by this dark signal. So, under low light conditions, an image may be formed from only a small amount of photo-generated charge. In other words, despite a normal exposure time, dark charge can seriously degrade an image taken under low-light conditions (Figure 6).
When used in global reset mode, since the image is stored longer on some lines than others, the S/N ratio may vary top to bottom in a low-light image taken using global reset with a slow progressive scan readout. The slower the readout, the more time is there for dark signal to accumulate, aggravating the thermal degradation in the last lines read.
Design choices to reduce this noise source include speeding up the readout, which means using more digital bandwidth from sensor if possible. Then there is the option of increasing speed of optics to focus more light flux onto the sensor or cool the sensor or subtract a dark frame of equal exposure length from the image (Figure 6).
The next article of this series, Part 4, will explain the charge transfer pixel concept and examine global snap shutter mode of operation in some detail.
Richard Crisp is VP of New Product Development for Etron Technology America.
This article was originally published on EDN.
Other articles in this series: