CCD versus CMOS: Which is Better?Doug George
by Doug George.
The Charge Coupled Device (known as a CCD) has dominated astronomy and consumer electronics for nearly five decades. That is changing.
The Nobel prize winning CCD was invented in 1969 and became a mature technology after about 20 years. CCD cameras gained wide acceptance for still imaging, video, and photometric measurements, replacing the previous generations of bulky vacuum tube equipment. The Hubble Space Telescope, launched in 1990, famously uses CCD technology to produce its stunning vistas and science data. On the home front, consumers bought CCD-based handheld camcorders possible to record family life, and businesses used them for security cameras and inspection equipment.
In the mid-1980’s, Active Pixel sensors were invented as a low-cost alternative to the dominant CCD technology. In the early 2000’s these sensors were updated to use the now-standard CMOS transistor technology. Although early CMOS Active Pixel sensors were only used in low-performance applications, the advent of smart phones pushed manufacturers to rapidly improve their performance. By 2007 CMOS had achieved market parity with CCD sensors, and by 2019 the first sensors capable of surpassing CCD performance appeared. Today, CMOS has matured to the point where it is replacing CCD in all but the most specialized applications.
Let’s first compare how the two sensor types work.
Camera sensors use picture elements known as “pixels” to detect light. A common analogy when talking about pixels is to imagine an array of buckets collecting rainwater. You could determine the shape and density of the cloud overhead by how much water ends up in each bucket.
The Bucket Analogy
CMOS and CCD both use arrays of silicon pixels (“buckets”) to detect light. When a photon of light hits a silicon atom, it knocks an electron into a higher energy state. This frees the electron to move through the material. It is now referred to as a photoelectron (“rain drop”).
The big difference happens when you read out the sensor. In a Charge Coupled Device (CCD), special electrodes attract and repel electrons, shuffling them out one-by-one to a corner of the chip. In our analogy, water is poured from one bucket to the next, like an old-fashioned fire brigade, until it reaches a corner of the array where it is measured. In a real sensor a couple of on-board transistors make this measurement by converting the number of electrons from a pixel into a voltage. It then goes to some electronics outside of the sensor, which include an analog-to-digital converter. The result is a number for each pixel, describing how much light was detected. Since all the pixels are measured by the exact same electronics, CCD cameras can be made very consistent and accurate.
CCD sensors are built using either NMOS or PMOS technology, which was popular in the 70’s but is rarely used today. Most modern electronics are built using Complementary Metal Oxide Semiconductor (CMOS) technology, which is a combination of NMOS and PMOS. By using CMOS it is much easier to build complex electronics right into the sensor itself. This can be a major cost and space savings, especially for a miniaturized cell phone camera.
In a CMOS detector, there are transistors at every single pixel. They convert the signal to a voltage, which connects via internal wires to some complex on-board electronics. Typical CMOS sensors have one or two analog-to-digital converters for each column in the sensor. Instead of a couple of transistors on board, there can be millions.
CCD Sensors Have One Readout In Corner,
CMOS Sensors Have Readout at Each Pixel
By incorporating all these electronics into the sensor, the chip itself is made much more complex, but the camera is greatly simplified. CCD sensors only have one, two, or sometimes four readouts – potentially one in each corner. CMOS sensors have thousands. This means that CMOS cameras can read out incredibly fast, even 100X faster than a comparable CCD. For long-exposure applications that is not so important, but it is especially important for video cameras.
These thousands of readouts in a CMOS sensor have a huge speed advantage, but there is a high price to be paid in terms of amplifier glow and pattern noise. CCD users have seen a little glow in the corners of the sensor; early users of CMOS sensors were overwhelmed by the glow and long exposure problems of these new sensors.
In the last few years, the best CMOS sensors are finally approaching or even exceeding CCD performance levels, but not in every aspect. Let’s compare CCD to the highest-performing CMOS sensors available today:
|Availability||Some major CCD sensor lines are being obsoleted. Very expensive specialty sensors made by companies like Teledyne e2v are here to stay.||Companies are making major investments, and the technology has been improving rapidly. New sensors appear all the time.||CMOS is the future for most applications. CCD will be continue to serve specialty niches such as scientific instruments.|
|Cost – both the sensor and the camera itself.||Large CCD sensors are expensive, and external analog and digital camera electronics are complex.||Large CMOS sensors are similarly expensive. Analog electronics are eliminated but digital electronics are more complex.||For simple cameras, CMOS is much cheaper. For cooled low-light imaging cameras, there is little or no difference.|
|Sensitivity||60% – 95%, though high QE sensors are very expensive||75% – 95%||Bang for buck, CMOS|
|Speed – read out in megapixels per second (MPS)||1 to 40 MPS||100 to 400 MPS||CMOS|
|Read Noise – how much noise in electrons is produced at each pixel when the sensor is read||5-10 electrons for standard CCDs, 1 electron for more complex electron multiplying devices (EMCCD)||1-3 electrons is common for modern CMOS sensors||CMOS or EMCCD|
|Cooling||High cooling is relatively easily achieved||Sensors generate a large amount of heat and cannot operate at extreme cold temperatures||CCD|
|Electronic Shutter||Interline and frame transfer sensors only||Rolling shutter is less complex but pixels expose at different times; global shutter is more expensive||No major advantage|
|Mechanical Shutter||Required for full-frame sensors; very helpful for image calibration||Very helpful for image calibration||No major advantage|
|Pixel Size||3 to 25 microns||2 to 9 microns||Larger pixels are a better match for long focal length telescopes. Most CMOS sensors have small pixels, but some larger pixel models are appearing.|
|Well Depth – how many electrons can each pixel hold – very important for photometry||40,000 to 200,000||30,000 to 75,000. Can be mitigated via stacking given low read noise.||CCD, but stacking can give CMOS the advantage.|
|A/D Converter bits||16 bits||Usually 12; some chips now use dual gain to create 16-bit images but with some pitfalls||CCD|
|Binning – combining pixels for sensitivity or resolution matching||Easily achieved at an analog level with zero added noise, extremely high binning levels possible||On-chip analog binning is extremely limited; most available sensors can only perform 2×1||CCD|
|Amp Glow – on-board electronics create some light via LED effect||Easily mitigated by powering down readout transistors||This is a bigger problem with CMOS, since there can be millions of on-board transistors.||CCD, though CMOS has improved substantially|
|Infrared Imaging||Deep Depletion sensors can achieve high QE at 650 to 1000 nm||Currently not possible with CMOS||CCD|
|Fixed Pattern Noise||Occasional hot columns, easily mitigated||Fixed pattern noise can be a significant problem, but technology is improving rapidly||No major advantage with newer sensors|
|Calibration – how “clean” an image can be created||Techniques for CCDs are well established and effective||Can be more complex, e.g. HDR modes, lack of overscan data; techniques are still being perfected||CCD|
As you can see, CCDs still have some significant advantages for high-performance, low light level imaging – although these advantages are slowly being chipped away at by new CMOS technology.
Some of our customers need to detect extremely faint light sources, requiring either hour-long exposures or very high binning factors to achieve sufficient signal-to-noise ratio. For these applications CCD sensors have a massive advantage over the newer CMOS technology; they have far less “amp glow” and have far better analog binning capabilities. CMOS sensor simply don’t work in these applications.
Why, then, are major companies switching to CMOS now? The reality is that most (non-scientific) imaging applications require video or short exposures; in those situations, CMOS is superior in both cost and performance. This has undermined the economic proposition for manufacturing CCD sensors in volume.
As a result, ON Semiconductor began discontinuing the former Kodak / Truesense devices in 2019. But it is not the end of CCD technology. Certain SONY CCD sensors will be available until 2026. For the high-end astronomy and spectroscopy markets, companies like Teledyne e2v will continue to manufacture very expensive, extreme-performance CCD sensors for years to come.
Serious astronomical applications such as photometry and spectroscope or life sciences applications such bioluminescence and fluorescence will continue to need CCD technology for the near term. Less demanding imaging or those needing higher speed imaging will all switch to CMOS sensors. Within 5 years, we predict the state-of-the-art in CMOS will supplant even more applications. To meet your needs for today and tomorrow, Diffraction Limited’s SBIG line of cameras now include both high-performance CCD and modern CMOS sensors.
Addendum: The technology, and the imaging market in general, is changing rapidly. I will periodically update this article with the latest trends.