CCD and CMOS Sensor Architecture and Readout Modes

Two image sensor types widely used in cameras for microscopy are scientific grade Charge Coupled Devices (CCD) and scientific Complementary Metal Oxide Semiconductors (CMOS or sCMOS). There is a number of similarities between the two technologies, but one major distinction is the way each sensor reads the signal accumulated at a given pixel. This Tech Note will explain how the differences in readout modes impact the exposure timing, illumination, and triggering of cameras and light sources in microscopy experiments.

While many readout modes exist, CCD cameras popular in microscopy often use interline transfer CCDs in a Global Shutter mode. In Global Shutter mode, every pixel is exposed simultaneously at the same instant in time. This is particularly beneficial when the image is changing from frame to frame. The CCD however has an inherent disadvantage when it comes to frame rate. When the exposure is complete, the signal from each pixel is serially transferred to a single Analog-to-Digital Converter (A/D). The CCD’s ultimate frame rate is limited by the rate that individual pixels can be transferred and then digitized. The more pixels to transfer in a sensor, the slower the total frame rate of the camera.

Figure 1

Figure 1. Basic schematics of CCD (left) and CMOS (right) sensor architectures.

A CMOS chip eliminates this bottleneck by using an A/D for every column of pixels, which can number in the thousands. The total number of pixels digitized by any one converter is significantly reduced, enabling shorter readout times and consequently faster frame rates. While there are many parallel A/D’s sharing the workload, the entire sensor array must still be converted one row at a time. This results in a small time delay between each row’s readout.

Rather than waiting for an entire frame to complete readout, to further maximize frame rates, each individual row is typically able to begin the next frame’s exposure once completing the previous frame’s readout. While fast, the time delay between each row’s readout then translates to a delay between each row’s beginning of exposure, making them no longer simultaneous. The result is that each row in a frame will expose for the same amount of time but begin exposing at a different point in time, allowing overlapping exposures for two frames. The ultimate frame rate is determined by how quickly the rolling readout process can be completed. This Rolling Shutter mode is illustrated in Figure 2.

Figure 2

Figure 2. Diagram demonstrating the time delay between each row of pixels in a rolling shutter readout mode with a CMOS camera.

For a CMOS sensor in rolling shutter mode, the frame rate is now determined by the speed of the A/D (clocking frequency) and the number of rows on the sensor. For example, the QImaging optiMOS is a sCMOS camera that has a combined A/D speed of 283MHz with 1,080 rows of pixel. At this clocking frequency, a single line’s readout time, and consequently the delay between two adjacent rows, is approximately 8.7µs. With 1,080 rows, the exposure start and readout time delay from the top of the frame to the bottom is approximately 10ms. This also corresponds to the maximum frame rate of 100 Frames per Second (fps) and minimum temporal resolution of 10ms (at full frame).

Figure 3

Figure 3. Diagram demonstrating the overlap of multiple exposures in a sequence with a sCMOS sensor running 100fps.

 

Rolling Shutter Challenges

While rolling shutter offers the advantage of fast frame rates, the overlapping behavior and time delay between each row’s exposure may also be a disadvantage under some conditions. Rolling shutter’s challenges can be summarized by two categories:

1. Rolling shutter spatial distortions
Imaging moving objects requires consideration for the object’s size and velocity in order to properly sample and avoid motion blur. This is true for both global and rolling shutter readout, irrespective of whether a CCD or CMOS sensor is used. If an object moves a significant distance during a given exposure, the image of the object will still be subjected to motion blur. When imaging moving objects and assuming a blur of no more than 10% is acceptable, the exposure time required can be calculated as:

Equation
Where: T= exposure time; delta x= the object’s length; v = the object’s velocity

Rolling shutter specific spatial distortions, however, can occur for large, fast moving objects even with very short exposures that avoid motion blur. Due to the time delay between each row’s exposure, it is possible that if an object is large enough and fast enough, the relative structure of an object will appear to change. Under this scenario, a complete global illumination and exposure is required to avoid spatial distortions. It is important to note that this sort of distortion is far less common than the general motion blur for biological microscopic imaging. Nevertheless, a straightforward way of determining if an object will experience rolling shutter spatial distortion is to use the following logic:

- IF: Exposure time < Object Height (# of Rows) * 8.7µs
- THEN: Rolling shutter spatial distortion is expected

a. Example 1: slide scanning with the rolling shutter at 10fps

  • Sensor size: 1920 x 1080 with 6.5um pixels = 12.48mm x 7.02mm
  • At 10fps, scanning velocity is 124.8mm/s (moving horizontally)
  • Tissue image height of 7.02mm
  • To avoid a 10% blur, exposure time must be <5.6ms
  • Frame readout time and maximum time delay across the rows is
    1080 x 8.7µs, or ~ 10ms
  • 10ms > 5.6ms; therefore, rolling shutter spatial distortions are expected

b. Example 2: imaging fast axonal vesicle transport

  • 100nm vesicles moving at a velocity of 3µm/sec
  • At 100x magnification, vesicle diameter on sensor is approximately 4 pixels
  • To avoid a 10% blur, exposure time must be <3.3ms
  • Readout time and maximum time delay between rows covered by the object is
    4 x 8.7µs = 34.8µs
  • 34.8µs << 3.3ms; therefore, rolling shutter spatial distortions are NOT expected

2. Complexity when synchronizing with a light source
When the imaging application requires synchronizing with a triggered light source or switching between multiple excitation wavelengths for every exposure, the overlapping exposure of frames with a rolling shutter sensor can add a layer of complexity.

a. Single Excitation Wavelength
One common application of hardware triggering is to shutter a light source and synchronize the sample’s excitation with a camera’s acquisition. By doing this, light is only exposed to the sample when the camera is acquiring images, thereby reducing the photo-toxicity and photo-bleaching of the sample. Under this condition, a hardware triggering line from the camera is used to turn on a light source at the beginning of an exposure and off at the end of an exposure. However, with rolling shutter, since each row exposes at a slightly different point in time, the definition of “beginning” and “end” of exposure becomes ambiguous, as it can follow any individual rows exposure behavior.

The simplest triggering behavior on a rolling shutter sensor follows the beginning and end of the first row’s exposure and is shown in Figure 4. When viewing or collecting a live stream with just a single excitation channel, this mode is perfectly acceptable as it keeps the light source on for the duration of the stream and allows the fastest frame rates possible. However, acquiring a single frame in this mode means the light source is on only during the first row’s exposure. The result is that every row after the first row has a decreasing portion of its exposure time that is actually collecting light. This is increasingly evident for exposure times close to the readout time of the sensor (10ms) and results in a top-to-bottom gradient of intensity down the image.

Figure 4

Figure 4. Illustration of the vertical intensity gradient caused by the time delay between each row of pixels in a rolling shutter readout mode with a CMOS camera and a triggered light source.

In Figure 4, a user enters a 10ms exposure and selects snap, causing the camera to receive a software or hardware trigger to begin acquisition. Row 0 on the sensor (top row) begins exposing, and the “Expose Out” hardware trigger line from the camera goes high. This Expose Out signal is used to trigger an LED light source, which turns on and stays on the entire time the Expose Out line is high. Once Row 0 has finished its 10ms exposure, Row 0 is read out, the Expose Out line drops and the LED light source turns off.

While each row on the sensor exposes for the same length of time (10ms), the delay in exposure start between the subsequent rows reduces the amount of light actually captured per row. With an 8.7µs delay between each row’s exposure start, the length of time that a single row collects light while the excitation source is on can be calculated as follows:

Single Row’s Illumination Time = Exposure time – (8.7µs x Row Number)

For a 10ms exposure, Row 0 receives 10ms of light, Row 539 receives 5.3ms of light and Row 1079 receives 0.6ms of light. The result is a steady gradient running from the top to the bottom of the frame.

b. Multiple Excitation Wavelengths
For multi-channel fluorescence imaging, where the light source alternates between excitation wavelengths for every exposure, the gradient will appear consistently in each channel as long as the channels are slowly acquired so that each color capture is completely exposed before the next channel’s exposure begins.

Figure 5

Figure 5. Two 10ms exposures with the optiMOS and the X-Cite® XLED1 from Excelitas Technologies. With the standard rolling shutter exposure and with a triggered high speed light source, a single acquisition results in a top-bottom image gradient with Row 0 receiving 10ms of light exposure and Row 1080 receiving 0ms of light exposure.

However, for rapid capture and switching between multiple excitation channels, typical rolling shutter acquisitions can result in channel blending due to the overlapping nature of CMOS rolling shutter readout. For example, when using a 5ms exposure, a single acquisition results in a gradient from the top of the image down to the middle of the frame. As discussed above, this occurs because the light source is turned off after the first row is finished exposing and begins reading out, which for a 5ms exposure, occurs before the entire second half of the sensor begins exposing. However, when running a live focus loop or acquiring multiple frames rapidly in sequence, the second frame begins exposing simultaneously as the second half of the first frame is finishing its exposure.

Figure 6

Figure 6. One 5ms exposure taken from a multi-channel stream of images. With the standard rolling shutter exposure and a triggered high speed light source that’s alternating between excitation wavelengths for every exposure, a single frame results in a top to middle image gradient for one channel and a bottom to middle image gradient for the other channel.

When this occurs, the light source is again triggered and the second excitation channel turns on while both the second half of the first frame and the first half of the second frame are exposing. The resulting Frame 1 image with a 5ms exposure then has a channel 1 (red) gradient running top to middle and a channel 2 (green) gradient running bottom to middle. The resulting Frame 2 image is then just the inverse of frame 1, with a channel 2 gradient running top to middle and a channel 1 gradient running bottom to middle. One way to avoid these single channel and multiple channel artifacts is to run the CMOS camera in global shutter readout mode. However, many CMOS cameras do not operate in this mode or are subjected to other tradoffs that limit their benefit.

CMOS Global Shutter Challenges

Even though the architecture of a CMOS sensor requires that each row of pixels is digitized individually, on some chips it is possible to achieve a global shutter readout. Similar to a CCD, when a CMOS sensor is in global shutter mode, each pixel begins and ends its exposures simultaneously. Once the exposure is complete, each pixel then simultaneously transfers its charge to a non-photosensitive transistor where it waits to be digitized. The A/D’s then clock through the sensor, digitizing each row individually. Before beginning the next exposure, there is a global clear to eliminate any charge accumulated during readout and ensure every pixel achieves an equal and simultaneous exposure.

Under these conditions, global shutter readout eliminates the time delay across the frame for both sample illumination and detection. This also eliminates spatial aberrations from extremely fast moving objects, eliminates gradients with triggered illumination and allows fast excitation channel switching while avoiding overlapping channels. Global shutter readout on a CMOS, however, comes at the cost of significantly reduced frame rate. By holding charge as each row is individually digitized and clearing the sensor before the next exposure, the overlapping capability of the sensor is lost and the frame rate is effectively cut in half.

For example, all sCMOS sensors in rolling shutter mode are capable of reaching 100fps at full resolution. However, sCMOS sensors running in a global shutter mode are limited to a maximum full frame rate of 50fps. Additionally, the global shutter readout process introduces nearly double the read noise as well as increases the dark current noise. These constraints of slower speeds and increased noise can limit the practical application of global shutter with a sCMOS sensor.

Best of Both Modes - Achieving Global Exposures with Rolling Shutter

To maximize the performance of sCMOS sensors for microscopy, the ideal triggering implementation would achieve a global exposure of the sample that eliminates channel overlap and rolling shutter spatial distortions while avoiding the increased noise and reduced frame rates seen with traditional global shutter. The QImaging optiMOS sCMOS camera offers a custom triggering mode that is able to achieve a global exposure with a rolling shutter readout. This triggering mode allows rapid shuttering of a high speed light source (lasers, LEDs) which is pulsed only when all rows in the frame are exposing at the same instant in time to achieve a global exposure. Meanwhile, the camera is kept in rolling shutter mode for digitizing charge in order to maintain high frame rates and low read noise. This mode in the optiMOS is called “All Rows Mode.”

Figure 7

Figure 7. Diagram demonstrating artifact-free global illumination using 10ms exposures using the optiMOS All Rows Mode to trigger a single excitation channel light source.

All Rows Mode is described in Figure 7. In this example, the user enters a 10ms exposure, selects snap, and the camera receives a software or hardware trigger to begin acquisition. Row 0 on the sensor begins exposing, but initially the “Expose Out” hardware trigger signal from the camera stays low. Each row subsequently begins its exposure with the Expose Out signal remaining low until the very last row of the frame begins exposing. Once the last row starts its exposure, the Expose Out line goes high. This rising edge of the Expose Out signal is used to trigger a high speed light source such as a laser or LED, which turns on and stays on the entire time the Expose Out line is high. This event marks the beginning of the full frame’s exposure, as all rows have started their exposure and the Expose Out line has turned on the excitation source.

All rows continue to expose for the length of time defined by the user (10ms). Once this time has elapsed, the Expose Out line drops, turning off the excitation source and Row 0 begins readout, followed by the cascading readout of every remaining row. In this mode, the camera continues to expose and readout in a rolling shutter fashion, but the Expose Out line is used to gate the illumination so that all rows in a single frame collect light for the same length of time and at the same instant in time. The result is a global illumination, achieving the same effect of a traditional global shutter and avoiding any potential spatial distortions due to rolling shutter. For this mode to function properly, especially at high frame rates, precise synchronization between the camera and a high speed light source is required. It is essential that the time required to turn the light on and off is very small relative to the exposure time of the camera. New high speed light sources such as the X-Cite® XLED1 from Excelitas Technologies, are capable of shuttering the light on the order of 10µs, making it possible to both achieve a uniform global illumination in this All Rows Mode as well as rapidly switch between multiple excitation wavelengths.

This All Rows Mode is particularly useful for fluorescence imaging applications using multiple excitation wavelengths. The Expose Out line now synchronizes the camera with the light source so that all rows are capturing an image and exposed to a single excitation channel at the same instant in time, thus completely avoiding channel blending.

Figure 8

Figure 8.Two 5ms exposurestaken from a multi-channel stream of images with the optiMOS and the X-Cite® XLED1 from Excelitas Technologies. With the All Rows triggering mode and a triggered high speed light source that’s alternating between excitation wavelengths for every exposure, an artifact-free global illumination is achieved resulting in alternating channel exposures with no gradient and no channel blend.

With the All Rows Mode from the QImaging optiMOS, in combination with a high speed triggered light source, such as the X-Cite® XLED1 from Excelitas Technologies, it is possible to achieve over 90fps full resolution while maintaining low read noise and avoiding any rolling shutter artifacts.

While it is possible to achieve much higher frame rates with this custom triggering mode over traditional global shutter, there is still a limitation to the achievable frame rates. The time required to digitize all rows within a single frame of the optiMOS is approximately 10ms. This is a hard limit which is added to the user defined global exposure time.

To achieve a global exposure, the exposure time defined by the user does not begin until the last row of the frame starts its exposure. Since it takes approximately 10ms from the start of the first row to the start of the last row’s exposure, the achievable frame rate of the camera in this mode is now the exposure time plus 10ms. Below is a table outlining a few user-defined exposures and associated frame rates.

User Enters Each Line Exposes For Expose Out Goes High For Effective Exposure Expected Frame Rate
1ms 11ms 1ms 1ms 90fps
5ms 15ms 5ms 5ms 67fps
10ms 20ms 10ms 10ms 50fps
20ms 30ms 20ms 20ms 33fps

Of course, it is possible to further increase this frame rate by capturing smaller Regions of Interest (ROIs), as seen in the following table.

ROI Exposure Time Expected Frame Rate
1910x1080 1ms 90fps
1910x512 1ms 175fps
1910x128 1ms 460fps



 

Summary

CCD and CMOS (or sCMOS) image sensors each offer distinct advantages for scientific microscopy applications, but their different readout modes, rolling shutter and global shutter, have varying impacts on exposure timing, illumination synchronization, noise levels and spatial distortions. CMOS cameras, like the optiMOS, have a considerable speed advantage over many CCD cameras; however, their overlapping behavior and time delay between each row’s exposure may introduce rolling shutter artifacts under some conditions.

Although global shutter readout offers the simplest solution to avoiding rolling shutter artifacts, many CMOS sensors either do not offer global shutter readout or offer it with reduced sensor performance. One way described in this Tech Note to obtain the speed and noise benefits of rolling shutter while avoiding rolling shutter image artifacts, is to use a special triggering mode, like the optiMOS All Rows Mode in combination with a fast switching light source. The following table summarizes the advantages and disadvantages of the rolling shutter readout triggering in All Rows Mode and traditional global shutter readout with a sCMOS camera.

  All Rows Mode
with Rolling
Shutter Readout
sCMOS with
Global Shutter
Readout
Read Noise (rms) 1ms 90fps
Maximum Frame Rate (1ms exposure) 5ms 67fps
Eliminate multi-channel overlap 10ms 50fps
Eliminate rolling shutter spatial distortions 20ms 33fps
Global Exposure with non-triggered light source No Yes

 

One thing to note is, that while this special triggering mode used with rolling shutter has the advantage of higher frame rates and lower noise performance compared to global shutter, it does require a triggered high speed light source and shorter exposure times to achieve the >50fps frame rates. Shorter exposure times reduce signal-to-noise in the image, but it also reduces motion blur artifacts, a tradeoff that should be taken into consideration when designing an experiment. Ultimately, the researcher must decide which camera, light source, and triggering mode combination will capture the most useful data for a particular experiment. Consideration should be given to the impact of these different options in terms of signal-to-noise, temporal sampling and image artifacts, among others, for the objects being observed. The various combinations’ advantages and disadvantages mentioned above should assist in that decision.

 

 

Contact a QImaging Consultant

Thank you for your interest in QImaging Products. Please complete the form below and the QImaging Consultant in your area will contact you shortly.

As a standard practice, the information you submit is used only in connection with your interest in QImaging products and is not shared with third parties.

Back to Top