Why CMOS over CCD?

[David Delaney]

I am still not sure of all the particulars, but I am wondering why mfcter are using CMOS over CCD especially with the strobbing and rolling shutter. I guess in the past, cost was an issue, but not the cost is relatively the same according to this article:
CCD vs. CMOS

Any ideas why they keep trying to use this tech when it has these issues?

[Jon Fairhurst]

I think the answer is on the very last point on that page. CMOS sensors can be produced on legacy chip lines, so they lend themselves to high volume manufacturing. Couple that with the push for larger and larger sensors, and that increases the volume of wafers that need to be produced for a given number of sensors. As sensors grow, the chance for a flaw in each sensor grows. That reduces yield, which further increases the needed wafer volume.

Personally, I'd like to see CMOS chips developed with an additional transistor per pixel that would allow a global shutter. Unfortunately, this would increase noise slightly. But someday, the noise will be low enough that the additional transistor noise would be acceptable.

[Daniel Browning]

Originally Posted by David Delaney

Why CMOS over CCD?

There are three main reasons.

• Features, features, features.
  First, CMOS allows you to have tens of millions of tiny computers (transistors) inside your image sensor, one inside each of your pixels. This tremendous capability can be harnessed for all sorts of different purposes: from improving image quality to providing any number of advanced features, like HDR, ADC, high frame rates, etc.
• Everyone is doing it.
  The majority of image sensor designers are CMOS, so even if a particular camera would have been better served by a CCD design, it will be CMOS if that's what the designer knows. But if you look at the reason *why* CMOS is now so popular, I think you'll find it's because of the first reason: features.
• Same or better image quality.
  The top-performing sensors in almost every sensor size from mobile phone up to 35mm are CMOS. They have the same or better QE, FWC, and especially read noise. I think this is caused by two things: 1) CMOS is more popular, so there is more brainpower going into its improvements than CCD, and 2) IS designers are using the inherent advantage of CMOS (transistors inside the pixels) to improve image quality in ways that CCD simply cannot (e.g. ADC).

Originally Posted by David Delaney

especially with the strobbing and rolling shutter.

Whether a sensor has a rolling or global (snap) shutter has very little to do with CMOS/CCD. In both cases, it takes up half the photodiode space, with all the downsides that entails (e.g. more noise, as Jon mentioned). They can (and do) design CCD with rolling shutter, CMOS with global, and vice-versa.

Originally Posted by David Delaney

I guess in the past, cost was an issue, but not the cost is relatively the same according to this article:
CCD vs. CMOS

I agree that it's not cost, and the article covers that very well. Overall I think it was a good article, I but I disagree with it on a few points.

For example, it says "CCDs have traditionally provided 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."

But it misses the fact that state of the art CMOS sensors equal CCD for QE (60-70% at 550nm *after* the CFA) and surpass it (by a wide margin) for read noise.

So why is CCD still so popular for the ultra-high end $40,000 Medium Format Digital Backs, astrophotography image sensors, and other scientific/industrial applications? Because you can buy very low quantities off the shelf with full service and support from Kodak and DALSA.

Of all the CMOS sensor designers, no one provides the same level of service for small quantities. They wont even return your phone calls unless you're into it for at least $100,000. As those markets grow, hopefully the other CMOS providers (I'm looking at you, Sony) will start improving their SDK, documentation, and integration pricing (e.g. test bench equipment, etc.) so that the low-profit markets like Medium Format and Astrophotography can finally get access to the state of the art technology.

[Daniel Browning]

Originally Posted by Jon Fairhurst

As sensors grow, the chance for a flaw in each sensor grows. That reduces yield, which further increases the needed wafer volume.

Agreed.

Originally Posted by Jon Fairhurst

Personally, I'd like to see CMOS chips developed with an additional transistor per pixel that would allow a global shutter. Unfortunately, this would increase noise slightly. But someday, the noise will be low enough that the additional transistor noise would be acceptable.

Agreed. I wonder if it's possible to put in a transistor that allowed user-customizable read times. Then you could actually use the rolling shutter for creative effects. One shot might have it set to global shutter, then another would be set to "rotary film shutter" speed, then another time you might set it to "equals frame rate" (or 1/24 second). There's probably only two settings that people would ever really want: global and film-like.

Originally Posted by Jon Fairhurst

I think the answer is on the very last point on that page. CMOS sensors can be produced on legacy chip lines, so they lend themselves to high volume manufacturing.

I think if you re-read the article you'll see that it's saying the opposite of what you are saying here. It correctly says that it costs just as much to convert a legacy chip line to CMOS as it does to CCD. So the hoped-for cost benefits never materialized: CMOS is at least equally expensive. The only companies to really attempt low-cost re-use (with very little modification) failed badly, e.g. Fill Factory.

[Chris Lawes]

I didn't realize that there are CMOS sensors with global shutter? If this is so, why don't the $20k+ ENG news cameras with CMOS sensors have a global shutter? I imagine that any tv station live event news cameraman would rather trade noise/etc for wobble and skew ???

I sure know I would. The only options out there right now that are skew-free (that I know of) and under $10k are the old slightly soft Pansonic HPX170s and the JVC HM700 series.

Aside from my above question, can you recommend some ENG CMOS sensors that have a global shutter or are less prone to skew/wobble?

[Brian Drysdale]

Most of the 2/3 ENG cameras currently have CCD sensors rather than CMOS, which currently seem to be used in the 1/2" and large sensor cameras. Although, the announced Aaton 4k camera uses a Dalsa CCD sensor and the more up market Sony large sensor cameras (apart from the F65) have CCDs .

I haven't yet heard of a camera with a CMOS global shutter, although the new cameras from RED and ARRI have hardly any noticeable skew and the SONY F65 has an optional mechanical shutter which seems to give good motion.

There certainly seems to a push towards CMOS sensors on a number of cameras.

[Daniel Browning]

Originally Posted by Chris Lawes

I didn't realize that there are CMOS sensors with global shutter?

The article in the first post mentions a great example. Vision Research used a global shutter in their high frame rate cameras, but they used a normal rolling shutter for their cinema-oriented cameras.

Originally Posted by Chris Lawes

If this is so, why don't the $20k+ ENG news cameras with CMOS sensors have a global shutter? I imagine that any tv station live event news cameraman would rather trade noise/etc for wobble and skew?

I guess it's because the camera manufacturers think news stations are less bothered by rolling shutter artifacts than they would be by global shutter downsides (e.g. more noise).

Originally Posted by Chris Lawes

Aside from my above question, can you recommend some ENG CMOS sensors that have a global shutter or are less prone to skew/wobble?

I don't know.

[James Chen]

Originally Posted by Brian Drysdale

Most of the 2/3 ENG cameras currently have CCD sensors rather than CMOS, which currently seem to be used in the 1/2" and large sensor cameras. Although, the announced Aaton 4k camera uses a Dalsa CCD sensor and the more up market Sony large sensor cameras (apart from the F65) have CCDs .

I haven't yet heard of a camera with a CMOS global shutter, although the new cameras from RED and ARRI have hardly any noticeable skew and the SONY F65 has an optional mechanical shutter which seems to give good motion.

There certainly seems to a push towards CMOS sensors on a number of cameras.

Commercial CMOS global shutter cameras do exist, but are very expensive. One example is the Phantom Flex uses a global shutter recording 2560x1600 at 1455fps. It's also a $120k+ camera.

To my understanding, most prosumer and professional CMOS sensors are developed with 4 transistors per pixel (4T). An extra transistor would be needed to store the photosite data prior to line scanning to achieve global shutter (5T). Doing so greatly increases manufacturing costs (considering how many transistors would need to be added at such high pixel densities) and would also introduce more noise, heat and power consumption, so smarter ways to deliver the same IQ was 4T imagers might be needed.

[Brian Drysdale]

Interesting to hear that Phantom Flex is global shutter. Just looking at the Vision Research site, this will allow "shooting super-fast-changing events such as lightning strikes"..Flash guns etc being a weakness of the standard CMOS, something that could restrict the subject mater of a high speed camera.

[Daniel Browning]

Originally Posted by James Chen

It's also a $120k+ camera.

There are plenty of machine vision (industrial) CMOS sensors with global shutter (e.g. CMV2000 - CMOSIS) for a lot cheaper than that. The 1/3" to 2/3" machine vision cameras based global shutter cmos sensors are available for $1500 - $3000. The market is a lot smaller than normal camcorders (and is often customized for each customer), so the prices are a lot higher than they would be if they were selling to millions of consumers. I would guess that there are similarly priced models with your choice of either global shutter or rolling shutter, depending on the needs of the application (noise vs rolling shutter artifacts).

Originally Posted by James Chen

An extra transistor would be needed to store the photosite data prior to line scanning to achieve global shutter (5T). Doing so greatly increases manufacturing costs (considering how many transistors would need to be added at such high pixel densities)

Adding an extra transistor does not increase manufacturing costs. The materials are the same, the labor is the same, yeild is the same, so cost is the same. The only possible cost increase is when a new and unproven process is used, such as BSI. But that is separate from the number of transistors. For a given process, 4T and 5T will have the same cost. But 4T on one process (BSI) may be a lot more expensive than 4T on a different process (early BSI). Of course, the fixed cost of the design itself may be more expensive, if you are comparing the cost of re-using an old, proven design (4T) without any changes against the cost of designing a new one (5T).

Originally Posted by James Chen

and would also introduce more noise, heat and power consumption, so smarter ways to deliver the same IQ was 4T imagers might be needed.

I can believe that.

[Alister Chapman]

As well as the extra transistor at the photo site, which may also reduce sensitivity as it will cover some of the photo site area, is there not also the need for extra address wires? Isn't one of the keys issues with CMOS shuttering down to the the way you address the pixels. While you could address all the pixels in one hit with the normal horizontal and vertical address wires for blanking or shifting the signal to the buffer transistor, when you read you must still address one pixel at a time and that takes time. Faster reads will mean more noise and heat and power consumption.

I'm sure if global shuttered CMOS was really that straight forward that Sony would have included it on the F65 rather than going to all the expense of a mechanical shutter.

[Sander Vreuls]

Grass Valley has been useing a mechanical shutter on their CCD camera's for years.. But now they have CMOS models on their low budget LDK3000 (and Infinity) and I can't find out if they still use the mechanical shutter on those.. Anyone know?

[Chris Lawes]

Well here's a question, is it still preferable to use an older CCD such as a Canon XH G1S vs. a newer CMOS such as a Panasonic HPX370 (or HPX250) for high-motion stuff like car-chase, dancing/stumbling through a crowd, fast-pans etc?

Meaning, will the CMOS sensors ever advance enough with a rolling shutter to be comparable to a 2006-era CCD?

If not it starts to become a hard question to use older/less sharp/inferior IQ chips or a modern wobbly chip depending on the application..