Three imager technologies—CCD, APS CMOS, and ACS CMOS—offer choices for vision/imaging applications.
By Terry Zarnowski and Tom Vogelsong
The early market predictions that CMOS (complementary metal-oxide semiconductor) imagers would quickly overwhelm CCD (charge-coupled device) imagers for many vision and imaging applications have not been realized. Whereas early active-pixel-sensor (APS) CMOS devices have raised the bar on physical and functional attributes, many designers have been unwilling to compromise on video quality, particularly for demanding applications. The recently introduced active-column-sensor (ACS) CMOS imagers, however, hold the potential of overcoming video-quality limitations and finding their way into more-demanding applications.
According to industry analysts, the newer imaging technology involving CMOS-type image sensors has been promising for several years to replace the established technology of CCD-type image sensors for many vision/imaging applications. Numerous technical articles and research papers during the past few years have touted the demise of CCD sensor technology, but CCD sensors still reign as the leading imager for most applications. However, CMOS sensors are steadily making inroads in market share.
Due to their lower cost, CMOS imagers are selling into high-volume applications in the low-end market. These applications include PC cameras, toy cameras, low-resolution digital still cameras (DSCs), position detectors, and some machine-vision systems.
A performance examination of the three dominant imager technologies—CCD, APS CMOS, and ACS CMOS—sheds some light on how designers make a sensor selection for a given application. Important attributes of imagers can be divided into three groups: physical, functional, and video quality. Important physical attributes for designers to consider are cost, physical size, ease of use, and power consumption. Functional attributes include on-chip features such as analog-to-digital converters, subframe capability, binning, nondestructive readout, and imager timing and control, among others. Important video-quality attributes include noise, sensitivity, dynamic range, speed, and resolution.
To evaluate the defined attribute groups and see how the three imager types can be applied, four typical applications—a PC camera, a machine-vision system, a DSC, and an astronomical telescope—are useful examples from the viewpoint of a system designer. To help choose the best imaging system for these applications, the attribute groups can be compared for the three types of imaging systems.
Charge-coupled devices represent the present technology standard for imaging systems. During the past 30 years, these imagers have been steadily improved by manufacturers and now provide good image quality for a range of applications.
However, CCD imagers require an involved and relatively high-cost integrated-circuit manufacturing process. Consequently, all the imager design and development costs are factored into CCD-chip prices. The CCD-chip manufacturing volume is relatively low compared to standard integrated circuits, which affects device yield and cost. The CCD manufacturing process does not typically incorporate timing and control logic integration on the same chip or more complex functions such as analog-to-digital (A/D) conversion.
A typical CCD also requires complex external circuits to generate precisely timed multiple phases of high-power clocks. These clocks transfer the photon-generated charge from the pixel site to an amplifier. Multiple power supplies are also needed to supply the device with proper bias voltages. Moreover, the video output from the CCD must be conditioned before it can be input to an external A/D converter or sent to a monitor.
FIGURE 1. In an APS CMOS image sensor, a source-follower amplifier is used within a pixel. This setup decreases both the amount of pixel area available for light collection (fill factor) and its full-well capacity. This pixel amplifier converts the sensed photon-generated charge to a voltage that is then output to a column video bus. It must be sized to balance its operating speed with the bus capacitance. A higher speed requires a larger amplifier, thus decreasing the pixel's sensitivity and dynamic range.
Therefore, a high level of analog expertise is required by designers to develop a CCD imager assembly that optimizes the complex signals and timing required and minimizes system noise to deliver high-quality video output. As a result, numerous camera manufacturers are vigorously competing to design industry-acceptable CCD-sensor-based cameras.
Available CCD imagers are relatively costly, and their power consumption is typically high. Because multiple functions cannot be integrated on-chip, such functions must be added externally. As a result, CCD imagers are physically large and therefore are ranked "low" in physical attributes (see table). Moreover, because CCD imagers require off-chip circuits to perform the A/D function, they rate "poor" in functional attributes, as well. On the plus side, however, CCDs provide quality video performance, such as low noise, wide dynamic range, good sensitivity, fair antiblooming and smear-reduction capabilities, and operate at real-time rates (30 frames/s at standard video resolutions on a single video port). Consequently, CCDs achieve a "high" ranking in video quality.
APS CMOS-based systems
Complementary metal-oxide semiconductor fabrication has become the mainstream choice for integrated-circuit (IC) manufacturing. It offers high volume, low power, low cost, and a higher level of both analog and digital logic integration on the same silicon die. CMOS fabrication is now producing millions of highly integrated analog and digital complex devices such as microprocessors, modems, A/D converters, and ASICs, among others. It can also incorporate imaging-system functions on the same silicon die, permitting small, high-performance devices.
FIGURE 2. In an active-column sensor CMOS image sensor, a shared unity gain amplifier is used at the top of a column of pixels in place of the source-follower amplifier per pixel. This amplifier furnishes feedback at each column of pixels, leaving only the input element of the amplifier present at each pixel site. Unlike the APS pixel amplifier, the UGA input element is physically smaller. For a similar pixel size, the ACS imager offers greater dynamic range and sensitivity.
However, APS pixels are typically noisy, suffer from short dynamic range, and can be speed-limited. An APS pixel therefore uses an amplifier within the pixel (see Fig. 1), which decreases the amount of pixel area available for light collection (fill factor) and its full-well capacity. The pixel amplifier converts the sensed photon-generated charge to a voltage that is then output to a column video bus. This bus presents an inherent capacitance that increases as the vertical resolution increases. The pixel amplifier therefore must be sized to balance its operating speed with bus capacitance. Higher speed requires a larger amplifier, thus further decreasing the pixel's sensitivity and dynamic range.
The largest contributor to pixel noise is fixed pattern noise (FPN). It is caused by process variations during fabrication that produce gain and offset variations of the APS pixel amplifier from pixel to pixel. Because this noise is fixed in value, it can be subtracted out by using on-chip or off-chip circuits. Noise subtraction depends on the application and may not always need to be implemented. However, implementing noise subtraction contributes to increased chip and system size, complexity, cost, and power consumption. Consequently, the APS CMOS imager ranks "high" in physical and functional attributes but ranks "fair" to "poor" for video quality compared with CCD imagers due to decreased sensitivity, dynamic range, and speed, as well as increased noise.
ACS CMOS-based systems
The latest imager technology, ACS CMOS technology, eliminates the source follower amplifier per pixel and its associated problems by implementing a unity gain amplifier (UGA; see Fig. 2). This amplifier furnishes feedback at each column of pixels, leaving only the input element of the UGA present at each pixel site. Unlike the APS pixel amplifier, the UGA input element is physically small. For a similar pixel size, the ACS imager offers greater dynamic range and sensitivity.
Available APS imager products built with 7.5-µm square pixels provide fill factors of about 30%. At the same pixel size, an ACS imager provides fill factors of about 70% using the same foundry process. Alternatively, ACS imagers can be made using smaller pixels than APS imagers and achieve similar photosensitive areas and performance parameters, resulting in less-expensive imagers for high-volume applications.
An additional ACS imager benefit over APS types is that unity gain feedback assures uniform gain for every pixel, despite process variations. Moreover, correlated double sampling can be used to cancel offset variations. In addition, ACS technology does not convert the photon-generated charge into a voltage that drives a column bus, as done by APS imagers; instead, it produces a proportional current. As a result, ACS imagers can run much faster than APS imagers can using similar areas for pixel electronics. The ACS pixels, like the APS pixels, produce no smear since the charge stays within the pixel. Furthermore, they offer outstanding antiblooming capability in both rows and columns, which makes them well suited for high- and low-lighted scenes.
The ACS imagers offer other performance advantages over APS imagers. A key advantage involves inherent binning capabilities along the imager columns. As each pixel provides a current, the current output of multiple pixels can be binned (added and averaged) along a column by selecting multiple rows in any desired order. Row binning is also available. Nondestructive readout can be accomplished by CMOS imagers to extend the dynamic range by reading and selectively clearing pixels based on the video signal level. As a result, ACS imagers rank "high" in physical and functional attributes as do APS CMOS imagers. They also rank "high" in video quality as do CCD imagers.
Because of data-bandwidth issues with standard PC interfaces, such as universal serial bus, serial, or parallel, as well as speed issues with modems and the Internet, a high-resolution imager device is not well suited for very-low-cost PC videoconferencing applications, especially if trying to run at near video rates. Also, the pixel depth, or bits per pixel, also affects bandwidth; a higher depth parameter results in a higher bandwidth requirement. Market demands dictate that the PC camera must be inexpensive and physically small. Therefore, CCD imager technology is not a practical choice for this application. On the other hand, APS and ACS CMOS imagers can satisfy these application requirements.
FIGURE 3. Active-column-sensor (ACS) CMOS imaging sensor uses a unity gain amplifier to furnish feedback at each column of pixels, which results in improved dynamic range and sensitivity over CCD imagers. An ACS imager provides fill factors of about 70%, uses small pixels, and produces a proportional current.
Of the three types of machine-vision system applications—low-, medium-, and high-end—the low-end system share of the market is expanding rapidly. It is typically driven by special-purpose vision applications where physical attributes, such as cost and size parameters, and functional attributes outweigh video quality attributes. In the low-end market segment, APS CMOS imagers are finding increased acceptance.
However, medium- and high-end machine-vision applications require better video and higher speeds than low-end applications. They need high dynamic range to tolerate bright to dark images and high antiblooming response. Therefore, high-video quality is desired above physical and functional attributes, but lower cost and size parameters are also desirable.
In practice, CCD-imager-based systems offer high video quality and are currently widely used for machine-vision applications. Therefore, APS CMOS imagers have not been able to successfully penetrate this market segment. However, CMOS ACS imagers are finding increasing use because they provide good video quality, especially with its high dynamic range, anti-blooming ability, and sensitivity, and ranks "high" in physical and functional attributes.
The market for digital still cameras has grown quickly over the past few years. With the growth of this market has come segmented applications. Low-end digital still cameras and dual-use cameras are moving into high-volume production and are predominately driven by price. Resolutions of 640 x 480 pixels and 8-bit color appear acceptable to this segment, which is adequately served by APS CMOS imagers. Meanwhile, ACS CMOS imagers are expected to provide better performance and lower cost due to their increased fill factor.
Another application segment covers premium cameras. These multimegapixel designs deliver high image quality. The APS CMOS imagers have come up short in this market due to dynamic range limitations and low fill factor issues, which get worse for small pixels. The ACS CMOS imagers, however, can provide the required image quality since pixel sizes under four microns can be realized with acceptable fill factors.
In astronomy, imaging a star field is a challenging task, as very bright objects appear in the same scene with nearby dim objects. Yet, good imaging contrast is needed for each object without blooming. Because high resolution, ultralow noise, and high dynamic range are highly desirable parameters in astronomy applications, high-video-quality imaging is needed more than good physical and functional attributes. Therefore CCD imagers have been unchallenged in this field.
Currently, special CCD imagers are fabricated for these applications; they are known as backside-illuminated thinned devices. These CCDs are ground thin and are illuminated from the backside. This process greatly increases the quantum efficiency of these devices, while reducing the yield of low defect imager die. However, these devices are relatively expensive.
These CCDs also suffer some of the same problems as their front-lighted counterpart. Reading out a CCD requires that the photon-generated charge be moved out of the array; therefore, it is not possible to randomly access a pixel nor is it possible to perform a nondestructive readout. Nondestructive readout and random access provide astronomers with the ability to perform adaptive exposure and signal averaging on every pixel in the array. By doing so, read noise is greatly reduced through signal averaging, which provides a high signal-to-noise ratio.
In addition, the pixels collecting light on dark scenes can be exposed longer, thus improving contrast. However, longer exposure times do involve the issues of dark current. Dark current for ACS CMOS devices is generally slightly higher than that of high-end CCDs. On the other hand, ACS CMOS devices respond well with applied cooling. They perform satisfactorily at exposure times of more than 30 minutes at 0°C.
Therefore, for high-end applications, backside-thinned CCD imagers may be a better choice, especially when cooled.
TERRY ZARNOWSKI is director of sales and marketing and TOM VOGELSONG is president and chief executive officer of Photon Vision Systems, Cortland, NY 13045.