by Doug George.
We’re often asked whether CMOS or CCD sensors are better. The simple answer is, “it depends.”
Both types of sensors detect light the exact same way. An incoming photon hits an atom of silicon, which is a semiconductor. When this happens one of the electrons in the atom is boosted to a higher energy level (orbital), referred to as the conduction band. Silicon normally behaves like an insulator, so its electrons can’t move around. But once an electron is boosted up to the conduction band, it is freed to move around to other adjacent atoms, as if the silicon was a metal. What was an insulator becomes a conductor – this is why silicon is called a semiconductor. In optical sensors these now-mobile electrons are referred to as photoelectrons.
Both types of sensors use pixels. Pixels are simply a tiny square region of silicon, which collect and hold these photoelectrons. The usual analogy is an array of rain buckets in a field, each collecting rain water. If you want to know how much it rained in any part of the field, you just have to measure how full each bucket is. So far everything is the same for CCD and CMOS; it’s the measuring process where things are very different.
The Bucket Analogy
The Charge Coupled Device, or CCD, is the older and more mature technology. The chips are fabricated using either NMOS or PMOS technology, which was popular in the 70’s but is otherwise rarely used today. During readout CCDs move the electrons from pixel to pixel, like a bucket brigade. They shuffle one-by-one out through a readout amplifier in the corner of the sensor. The big advantage of doing this is that every pixel is measured in the identical same way. The use of a single readout amplifier makes the readout process extremely consistent. This produces high quality data with low fixed pattern noise and read noise. There’s also no wasted space in the pixel, which is a problem with CMOS sensors. Shuffling all the photoelectrons to one corner of the device does limit the readout speed though; as a result some sensors have readout amplifiers in each corner for faster readout.
CCD Sensors Have One Readout In Corner,
CMOS Sensors Have Readout at Each Pixel
Most modern electronics are built using CMOS technology, or Complementary Metal Oxide Semiconductors. CMOS devices use both NMOS and PMOS transistors, which gives them excellent switching characteristics. Building sensors with CMOS technology allows them to incorporate additional electronics, such as analog-to-digital converters. Each pixel in a CMOS sensor has its own readout amplifier, and often sensors have A/D converters for each column; this makes it possible to read out the array extremely quickly. The transistors located at each pixel use up some space, resulting in less sensitivity and well depth. Aside from speed, the primary motivation in developing CMOS sensors was cost, not performance. As a result for many years the sensitivity, noise, and dark current performance of CMOS sensors was far inferior to CCD sensors.
CMOS Readout Consumes Some Pixel Area,
Reducing Well Depth and Sensitivity
CMOS sensors do not require complex external clock driver electronics that produce precise voltages and waveforms to move charges around the sensor. They do not require complex external readout electronics, double-correlated samplers, and A/D converters. All of the electronic components needed for readout are built right into the sensor. The single chip just needs clean power to provide a good image, and it is directly read out digitally. That is the reason that CMOS sensors have a big advantage in terms of cost. That said, for scientific applications the additional mechanical and electronic hardware required to support cooling the sensors is still a major cost driver, regardless of the sensor type.
Over time the more mature CCD technology was enhanced through many innovations, both big and small. Interline sensors were developed for higher speed and shutterless operation. Microlenses were added to improve sensitivity, by directing light around readout electrodes on top of the chip. Back-Illuminated Thinned sensors avoid passing light through electrodes and other structures, resulting in quantum efficiency approaching 100%; however it is difficult and expensive to accurately thin the sensors. Electron multiplication devices (EMCCDs) with incredibly low read noise were also developed. Some of these innovations can be also be applied to CMOS sensors, including back-illumination and microlens technologies. Others such as EMCCD techniques are specific to the CCD architecture. Naturally CMOS device manufacturers adopted the technologies that were applicable, improving the performance of their sensors. This has helped close the gap in imaging performance between CMOS and CCD.
Back-Illuminated Thinned Technology Dramatically Improves Sensitivity,
but is Expensive to Manufacture
In addition to adopting these technologies, the overall architecture of CMOS sensors has also improved; for example, various methods are used to reduce the impact of the readout transistors on the sensitive area of the sensor, and others to reduce electrical noise. Some modern CMOS sensors have comparable performance to CCD sensors, and in some cases exceed their performance in certain respects. Recent scientific-grade CMOS devices, such as the GPixel GENSE400BSI have extremely high quantum efficiency and incredibly low read noise, on the order of 1.5 electrons in certain operating modes. But large amplifier glow and limited well depth combine to significantly limit the maximum usable exposure times. This limitation can be overcome through stacking images, but that process brings the final read noise back to conventional CCD levels. Still, these sensors have greatly closed the gap, and depending on the application can be superior to a CCD solution.
Today CMOS sensors have become dominant over CCD sensors in video, smartphone, and DSLR applications due to their low cost and fast readout speed. They are also making inroads into scientific applications. We can expect over time that CMOS will gradually supplant CCD in higher-performance applications. CCD technology currently still has advantages in well depth, amplifier glow, and large array sensors. But the gap continues to narrow, and we expect in the next 5-10 years to see CMOS supplant CCD in many high-performance imaging applications. It is likely, though, that for some time the highest-performance applications will still utilize EMCCD or back-illuminated thinned CCD technology.