Whether it's the latest cell phone or a large space telescope, solid-state imaging devices now fill nearly every image-capture need. Smaller pixel sizes enable existing VGA and multimegapixel sensors to shrink, while still larger sensors with tens of millions of pixels are readily manufacturable. Over the last few years, CMOS-based image sensors became the technology of choice for consumer products. They've won the economic and performance battle over charge-coupled-device (CCD) sensors for imaging devices with resolutions ranging from sub-VGA to about 8 Mpixels. Above the 8-Mpixel mark, though, CCDs still dominate due to their lower noise and better sensitivity. See Figure
CCD sensors reign supreme in industrial and medical applications, too, since many applications require high frame rates instead of high resolution. Chip architectures span the range from simple linear arrays of a few thousand pixels to multimegapixel arrays. Fairchild Imaging, Fraunhofer-IMS, Hamamatsu, Kodak, and Sarnoff Labs all offer solutions in this market.
CMOS-based sensors leverage the process scaling of CMOS technology and the ability to better integrate logic functions, such as image processors and analog-to-digital converters (ADCs), to create a full "camera-on-a-chip" solution. With pixel sizes for CMOS sensors already dropping to below 3 µm on a side, designers can craft smaller VGA-resolution sensors or multimegapixel sensors with the same chip size as previous-generation VGA sensors.
Also, automotive safety applications will start to consume significant numbers of low-cost imaging devices over the next few years. Backup cameras, drowsy driver alerts, airbag deployment, and other applications will use image data to better protect drivers.
Further advances in lithography and pixel design will allow for additional scaling, enabling designers to craft devices with even higher resolutions. The challenge will be in maintaining pixel-cell sensitivity as the light-capturing area shrinks. Also, if the captured light energy is lower, then the amount of background noise must be lowered to effectively maintain a sufficient signal-to-noise ratio. As a result, process developers must pay careful attention to reducing thermal noise and other noise sources inherent in the semiconductor materials to improve the signal-to-noise ratio.
In a CMOS sensor, each pixel has its own charge-to-voltage conversion. The sensor often includes amplifiers, noise-correction, and digitization circuits so the chip outputs digital bits. These other functions increase the design complexity and may reduce the area available for light capture. With each pixel doing its own conversion, pixel-to-pixel uniformity is lower. But by using on-chip logic, the chip can be built to require less off-chip circuitry for basic operation.
Processes used for CCD sensors aren't as flexible, and most CCD sensors require considerable external support circuitry. CCDs traditionally provide the performance benchmarks in the photographic, scientific, and industrial applications that demand the highest image quality (as measured in quantum efficiency and noise) at the expense of system size.
CMOS VERSUS CCD TRADEOFFS
No clear line divides the types of applications that use CMOS and CCD sensors (see "Comparison Of CCD And CMOS Imaging Sensors" at Drill Deeper 11898, www.elecdesign.com). While CMOS designers devote much effort toward achieving high image quality, CCD designers focus on reducing power requirements and pixel sizes to compete with CMOS devices at the lower end of the product spectrum. The main advantage of CMOS sensors will be low cost, since the sensors can leverage mainstream CMOS manufacturing capabilities.
At the high-end of the imaging area are CCD imaging devices with capacities ranging from 14 to over 81 Mpixels. (For more about these high-end solutions, go to "CCDs Replace 35-mm Film" at www.elecdesign.com, Drill Deeper 11899.) Between 14 Mpixels and 5 Mpixels, designers have a choice of both CMOS and CCD imagers, with the bulk of the solutions in CMOS. And there still are a few CCD imagers below 5 Mpixels, but they're getting harder to find as CMOS imagers totally dominate that portion of the market.
CMOS: UP TO THE TASK
One of the more unusual approaches in creating a CMOS imager comes from Foveon. Rather then use a singe layer of pixels covered by a layer of color filters, the X3 architecture employs three layers of pixels in the silicon (Fig. 1). This direct image-sensing approach directly captures the red, green, and blue light at each point in an image during a single exposure. A single pixel region then can capture all three primary colors. In contrast, most CCD and CMOS sensors use layers of color filters on top of the pixels that form a mosaic of three-pixel clusters to capture the three primary colors. The Foveon scheme exploits the fact that different wavelengths of light are absorbed to different depths in the silicon. Therefore, each vertical stack of red, green, and blue pixels directly records all of the light at each point in the image.
The largest X3 sensor packs 10 Mpixels. Its vertical stacking of the pixels is considerably smaller than a traditional X-Y array that uses the color filters. The sensor also uses little power, suiting it to many digital still cameras (DSCs). With a 2.5-V supply, it draws 50 mW during readout, 10 mW during standby, and 0.1 mW during power-down.
One pixel region does the job of three. And thanks to a variable pixel size feature, the sensor can seamlessly switch between capturing still images at maximum resolution and digital video at reduced resolution. That tradeoff is achieved via control signals that group adjacent pixels into clusters such as 1-by-2, 2-by-2, or 4-by-4 blocks.
The bigger the block, the greater the sensitivity, since more pixels collect light for the same point on the image. In its full-resolution mode, the sensor can capture 4.4 frames/s. In 576-by 384-pixel resolution mode, it captures 25 frames/s.
Additional circuitry on the imaging chip gives system designers a highly flexible on-chip readout system that simplifies the implementation of digital zoom, scene metering, and other functions. The chip also lets software implement a feature Foveon calls Fill Light, which considerably improves the quality of images affected by challenging lighting conditions. In the scheme, software simulates the photographic method of adding extra light to shadow regions while preserving highlight details.
Although these high-resolution sensors demonstrate the best the industry can offer, their price eliminates them from mass-market products such as cell phones, Web cameras, and consumer-grade still-image cameras (typically under 6 Mpixels). Advances in traditional CMOS sensor technology, though, continue to enhance resolution and sensitivity while further reducing chip size in efforts to lower chip cost.
Cypress Semiconductor's CYIHDSC9000AA, a 9-Mpixel color sensor for high-end consumer DSCs, banks on 130-nm design rules to deliver a pixel pitch of 6.4 µm. Designed to meet the requirements of the Advanced Photographic Standard, the imaging array consists of 3710 by 2434 pixels and occupies an area of 23.3 by 15.5 mm. This produces an effective focal length multiplier of 1.5 compared to a full-frame 35-mm camera. A monochrome version of the sensor also is available. The color sensor can deliver 5 frames/s at full resolution and 20 frames/s with VGA resolution.
With details expected at next month's IEEE International Solid State Circuits Conference (ISSCC), a Sony image sensor with 6.4 Mpixels promises some of the fastest frame rates for a consumer imager—60 frames/s. The chip's zigzag-shaped, 4-pixel sharing scheme results in an effective 1.75-transistor/pixel architecture. An on-chip, 10-bit, counter-type column-parallel ADC delivers digitized pixel data. Fabricated with 180-nm design rules, the pixels are just 2.5 µm square. Moreover, the imaging array can switch between full-frame and a 2-by-2 binning mode without inserting an extra invalid frame to avoid integration time inconsistency.
Also at ISSCC, Samsung Electronics Co. Ltd. will unveil a slightly larger sensor with 7.2 Mpixels. It will use a four-shared pixel structure but leverages 130-nm design rules and a copper damascene process, which reduces the pixel height and improves optical efficiency.
5 MPIXELS GO MAINSTREAM
For mainstream DSC applications, sensors with 5-Mpixel resolution ride atop the consumer mass market. At the same time, camera phones are leveraging the advances in CMOS sensors, which means top-of-the-line cell phones in 2006 will pack 5-Mpixel imagers.
Kodak, Micron, and Omnivision Technology are the main players competing at the 5-Mpixel resolution mark. And earlier this month at the Consumer Electronics Show in Las Vegas, newcomer Planet82 unveiled a novel nanotechnology-based 5-Mpixel sensor that can work with extremely low light levels.
Late last year, Kodak released the KAC-5000. This 5-Mpixel sensor offers a 1/1.8-in. optical format. Taking aim at mainstream DSCs, it employs 2.7-µm square pixels and Kodak's novel Pixelux technology, which combines the use of pinned photodiodes, four-transistor pixels, and a shared pixel architecture for high sensitivity under low-light conditions. It can capture 6 frames/s with full resolution and over 30 frames/s at VGA resolution.
Referencing the pinned photodiodes back to ground can reduce the dark current. To improve spectral response, the sensor's true correlated double sampling scheme removes thermally generated noise. Because the shared pixel architecture allows binning, four adjacent pixels can be grouped to create a larger pixel to better capture images in low-light situations. Dynamic power is a modest 150 mW. But the 0.5-mW standby power is about 10 times that of the Micron and Omnivision devices.
Micron's MT9P001 5-Mpixel sensor uses a larger, 1/2.5-in. optical format. It can deliver full-resolution images at 12 frames/s or VGA-resolution video at 30 frames/s. Based on the company's Digital-Clarity technology, the MT9P001consumes less than 260 mW, suiting it for DSCs as well as high-end cell phones.
Small, 2.2-µm square pixels configured as 2592 by 1944 elements keep the chip's dark current to just 20 electrons per second, minimizing background noise. The small pixel size also translates into a small imaging area of just 5.7 by 4.28 mm. However, it delivers a better dynamic range than the Kodak device—60 dB versus 52 dB.
Thanks to an on-chip 12-bit ADC, the chip can deliver digital data to the host system. An electronic rolling shutter enables the chip to take quick snapshots or capture continuous video. Sophisticated camera functions such as programmable gain, frame rate, exposure time, image mirroring, and viewfinder and snapshot modes are directly incorporated into the imaging chip.
The Omnivision OV5610 5.17-Mpixel camera chip packs a similar-sized pixel array and an on-chip ADC (10 bits versus Micron's 12 bits). But like the Kodak chip, it uses a 1/1.8-in. optical format and similar-sized pixels. It's the slowest of the three, though, delivering 4 full-resolution frames/s. On-chip circuitry and algorithms cancel fixed-pattern noise, eliminate smearing, and drastically reduce blooming and dark current. And, it can perform optical black-level calibration to achieve a 60-dB dynamic range comparable to the Micron sensor.
Active power consumption is about 140 mW while standby current is below 35 µW, suiting the chip for standalone cameras and camera phones. Control registers let designers manipulate the timing, polarity, and chip functions such as programmable auto-exposure, gain control, and auto white balance.
Planet82's approach uses a scheme the company calls single-carrier modulation photo detectors (SMPD). This technology creates an image sensor that functions like an artificial eye, capturing images without a flash in almost total darkness. The pixel element is based on a quantum transistor structure rather than a PN diode, and it has a sensitivity more than three orders of magnitude higher than CMOS or CCD-based sensors.
It can capture pictures without a flash—even with light levels below 1 lux, which is better than what the human eye can distinguish. The scheme also minimizes the aperture ratio of the pixel region, allowing more pixels per unit area on the chip. This shrinks the size of the chip compared to CMOS sensors using the same design rules. Power consumption also will be lower, typically about 82 mW for the 5-Mpixel sensor. Planet82 expects to start sampling the sensors in mid-2006.
As resolution drops to 3 Mpixels and below, more vendors now compete at today's sweet spots of 3- and 1-Mpixel resolution. Cypress Semiconductor, Kodak, Magnachip, Micron, Omnivision, and Toshiba join the fray at the 3-Mpixel level. Avago Technologies (formerly Agilent), Sharp, and STMicroelectronics can be found at 2 Mpixels and below too.
SENSORS TAKE DIFFERENT PATHS
Vendors in this market are moving in two directions. On the one hand, they design basic sensors with minimal on-chip logic. On the other, they're creating highly integrated camera-on-a-chip solutions with functions such as JPEG image processors, autofocus control, flash strobe control, and other image and video support functions. These approaches let cell-phone designers better match their phone architectures to the imaging subsystem. In addition to offering bare sensors, the vendors can offer value-added modules that combine the sensor, a fixed or variable-focus lens, and some control logic. (For more, see "Module Or Discrete" at www.elecdesign.com, Drill Deeper 11881.)
At 3 Mpixels, most imaging chips don't include high-level processing functions. But they usually can offer higher frame rates than larger sensor arrays. For instance, Micron's MT9T012 leverages the same 2.2-µm square pixels as the company's 5-Mpixel chip. It can deliver 15 frames/s at full resolution and up to 30 frames/s at lower resolution.
Targeting mobile applications, it uses a 1/3.2-in. optical format and includes programmable snapshot and flash control.
Kodak's KAC-3100 also leverages the 2.7-µm pixels of its larger brother, the KAC-5000. At 12 frames/s, it's twice as fast as the 5-Mpixel sensor. However, it's still slower than Micron's 3-Mpixel chip.
The ICM320T image sensor developed by Magnachip ( formerly IC Media) pushes speed a bit further. It uses 2.57-µm square pixels and a 1/2.7-in. optical format to deliver a full-resolution frame rate of 16 frames/s and progressively higher rates when the array is subsampled. The chip is frugal with power, consuming 70 mW at 15 frames/s and less than 20 µW on standby. Like the Micron sensors, a two-wire interface controls the various operating modes (exposure time, frame rate, subsampling window size, analog and digital gain, horizontal and vertical image inversion, and dead pixel removal).
Also delivering 15 frames/s, the OV3630 from Omnivision and the PS1320 from PixelPlus perform at levels similar to the Micron MT9T012. Based on a proprietary pixel structure the company calls Omnipixel2, the OV3630 sensor can cancel fixed pattern noise and considerably reduce smearing and blooming. Meanwhile, the PS1320 incorporates an on-chip image signal processor that lets users program various windows and frame rates, handle a video preview mode, and perform black-level compensation.
Toshiba's entry into the 3.2-Mpixel market, the ET8E99-AS, uses a 2.7-µm pixel and a 1/2.6-in. optical format. In its full-resolution mode, it too can deliver 15 frames/s and over 30 frames/s when pixels are binned (3-to-1 vertical binning). An on-chip ADC delivers raw digital data to the host over a serial differential interface (Fig. 2). It supplements the company's existing 2-Mpixel, 1.3-Mpixel, and VGA sensors and modules.
Designers can choose from many 2- and 1.3-Mpixel standalone sensors under the 3-Mpixel level. But these devices require external processors to take the image data and deliver JPEG still images or video. To lower system costs, a few companies are creating single-chip cameras that offer a more highly integrated solution.
Such chips include Micron's 2-Mpixel MT9D111 and 1-Mpixel MT9M111, as well as Avago's 1.3-Mpixel ADCC-3960. They include JPEG image processors and other system support logic that offload the host processor in a cell phone or camera. Thus, it becomes easier to add the camera function to an existing chip-set solution.
For a complete list of vendors, go to www.elecdesign.com, Drill Deeper 11900.