The Future of CCD Sensors

[Jim Kasson]

Regarding ADCs, they just take an analogue signal like 10 micro volts and convert in a number. There are only two ways to do it, correctly or not.

Uh, Erik, I know you were trying to simplify, but, if that was your aim, I think you succeeded altogether too well. As somebody who's put together a lot of analog/digital systems, I can testify that no ADC is perfect. There are static and dynamic errors, and they are often noisy, so that error has both a stochastic and a predictable component. Even a perfect ADC has an error of plus or minus half the least-significant bit, and practical ones, especially in the rarefied atmosphere of 14 bits, have more. When I have modeled cameras, I often use mean ADC errors of 0.5 to 1.5 LSBs, in addition to the worst-case half-LSB theoretical error.

That aside, you're doing a great job of explaining complex things very well in this thread, as usual.

Jim

[eronald]

Is the only reason there is no AD on chip CCD, that they didn't invest the time in r&d?

Other than the location of the AD, is there any other reason CCD will produce a different image than CMOS? Maybe the famous differences everybody are talking about, come from the different ADC?

Interesting question.

ADs have been off-chip on all except the most recent CMOS cameras; in fact they are one of the  expensive chips in the high speed cameras like the D3, because the faster you want to acquire (accurately), the more of them you need.

I think Canon pioneered the use of CDS pixel-sampling design  around the time the 1DII was released; AFAIK the AD was still off chip. Before that CDS was routinely employed in the AD stage of CCD designs. Initially the results from Canon were really smeary. This site explains why CMOS designs have some intrinsic noise issues..

I don't understand this stuff as well as many of the posters here, but I really feel that CCD has been the victim of underinvestment while CMOS has been pushed relentlessly by the Japanese for a variety of reasons, one of which is the low power consumption at standby - power consumed during use is proportional to frequency.

Usually a working technology can be perfected, and I think that it is sad to see CCD imagers abandoned when the replacement is not really a substitute.

Edmund

[Jim Kasson]

I don't understand this stuff as well as many of the posters here, but I really feel that CCD has been the victim of underinvestment while CMOS has been pushed relentlessly by the Japanese for a variety of reasons, one of which is the low power consumption at standby - power consumed during use is proportional to frequency.

Usually a working technology can be perfected, and I think that it is sad to see CCD imagers abandoned when the replacement is not really a substitute.

Edmund, aside from clever circuit design in CMOS sensors, the technology has a huge advantage in that it is a twist on mainstream semiconductor technology used for all kinds of things, while the CCD process technology has become kind of an orphan, used almost (entirely?) for image capture. At one, time, ages ago, CCDs were used as memory devices, but they lost out to DRAM and CMOS static RAM.

That's my perspective as an integrated circuit user starting with DCTL (which was really hard to work with, but seemed like a miracle at the time) in 1963. I'd welcome the perspective of a real semiconductor engineer. I know you're out there...

Jim

[BJL]

Phase and Hassy killed CCD.

They ... underinvested in new chip and body developments ... and pressured their chip suppliers for the best price.  The suppliers reacted by recycling their existing designs as far as they could take them without major reinvestment.

During the same period, Sony used the money from their sensor foundry to create a fountain of new chips which are actually drop-in modules, and are used by the whole industry - nobody was stopping Dalsa and Kodak from going the same way.

 As a result we have a super-modern CMOS all-singing all-dancing design fighting it out with geriatric 5 year old MF CCDs. There is no state of the art all-digital MF CCD with on-chip A/D available - in fact one can argue that Phase and Hassy wanted there to be none because it would have been a drop-in camera. Set me right if I'm wrong. I for one believe that MF-CCD was a perfectly good technology that was pushed into irrelevance by greed.

Firstly, AFAIK, MFD camera and back makers account for only a small fraction of the demand for CCDs from their sensor suppliers, with those CCDs developed mainly for other market like medical, scientific, industrial and military uses. So I doubt that Phase One or Hasselblad have much influence of the development of CCDs, and they can hardly be blamed for the industry-wide slow-down in or abandonment of CCD development. Note that many other sensor makers like Sony and Panasonic moved away from CCDs to active pixel CMOS sensors, and one cannot blame Phase One and Hasselblad for that abandonment of CCD development.
Instead, it seems that the older CCD approach to image sensors has simply matured to the point that there is little room for further progress. Except that is by adding features like on-sensor amplification -- which is the defining feature of "Active Pixel Sensors", commonly known as CMOS sensors (but known in the trade as APS!).

Secondly, what advantages do you claim that CCDs have over active pixel CMOS sensors that would make it worth trying to improve them rather than going with the industry-wide trend to active pixel sensors?  The single biggest disadvantage of CCDs is their greater read noise, which is related to the fact that the charge from each photosite is moved to the corners of the sensor without amplification, using a succession of thousands of hops of the charge from one photosite to the next, first down a column of photosites to the edge, and then (far faster) along the edge to a corner.  Active pixel CMOS sensors instead move the signal directly from a photosite to the edge of the sensor in a single step via a voltage signal, and this is done with amplification: the charge on a photosite is mapped to a proportionately larger change in a sense capacitor at the bottom of the column.  This amplification (when combined with DCS) is the key to the lower read noise and high dynamic range of active pixel sensors.  Doing ADC at the bottom of each column is a subsequent improvement, but even without that, Canon's older technology CMOS sensors with their off-board ADC win easily on low noise and DR over CCDs.

The only way I can see usefully adding on-chip ADC to a CCD is doing it in column parallel style, but that still has the problems associated with the thousands of hops of the unamplified charge signal down the column.  How would that be better than the active pixel sensor approach?

[eronald]

Thread title - "The Future of CCD Sensors"
Consensus answer ...

 NONE.

A bit puzzling we're not getting major dissent

Edmund  

Firstly, AFAIK, MFD camera and back makers account for only a small fraction of the demand for CCDs from their sensor suppliers, with those CCDs developed mainly for other market like medical, scientific, industrial and military uses. So I doubt that Phase One or Hasselblad have much influence of the development of CCDs, and they can hardly be blamed for the industry-wide slow-down in or abandonment of CCD development. Note that many other sensor makers like Sony and Panasonic moved away from CCDs to active pixel CMOS sensors, and one cannot blame Phase One and Hasselblad for that abandonment of CCD development.
Instead, it seems that the older CCD approach to image sensors has simply matured to the point that there is little room for further progress. Except that is by adding features like on-sensor amplification -- which is the defining feature of "Active Pixel Sensors", commonly known as CMOS sensors (but known in the trade as APS!).

Secondly, what advantages do you claim that CCDs have over active pixel CMOS sensors that would make it worth trying to improve them rather than going with the industry-wide trend to active pixel sensors?  The single biggest disadvantage of CCDs is their greater read noise, which is related to the fact that the charge from each photosite is moved to the corners of the sensor without amplification, using a succession of thousands of hops of the charge from one photosite to the next, first down a column of photosites to the edge, and then (far faster) along the edge to a corner.  Active pixel CMOS sensors instead move the signal directly from a photosite to the edge of the sensor in a single step via a voltage signal, and this is done with amplification: the charge on a photosite is mapped to a proportionately larger change in a sense capacitor at the bottom of the column.  This amplification (when combined with DCS) is the key to the lower read noise and high dynamic range of active pixel sensors.  Doing ADC at the bottom of each column is a subsequent improvement, but even without that, Canon's older technology CMOS sensors with their off-board ADC win easily on low noise and DR over CCDs.

The only way I can see usefully adding on-chip ADC to a CCD is doing it in column parallel style, but that still has the problems associated with the thousands of hops of the unamplified charge signal down the column.  How would that be better than the active pixel sensor approach?

[ErikKaffehr]

Hi Jim,

Thanks for making that point, I have very limited knowledge of electronics. Well, I have limited knowledge about most of things ;-)

The point I tried to make is that ADCs just convert a voltage into numbers. Clearly there is noise and un-linearity, but I would guess that any of that would apply to low signal levels (that is shadows/darks) and the quality of the ADC plays a much smaller role on mid tones where the sampled analogue signal is quiet strong. So I don't think that "MFD magic" comes from better ADCs.

A question: It seems that the Nikon cameras having Sony sensors are "low ISO DR kings", the D3S and D4 have external converters and very high ISO capability but cannot achieve the same DR as the Sony sensors. My impression is that the only explanation is that the sensors give very clean signal, but the ADCs have significant noise. So the clean signal from the CMOS sensor is going to variable gain amplifiers, increasing amplification reduces shadow noise, but pushes some of the signal beyond ACD range. Would the ACD be 14-bit "clean", those sensors would also excel with low ISO DR. Is that correct?

Best regards
Erik

Uh, Erik, I know you were trying to simplify, but, if that was your aim, I think you succeeded altogether too well. As somebody who's put together a lot of analog/digital systems, I can testify that no ADC is perfect. There are static and dynamic errors, and they are often noisy, so that error has both a stochastic and a predictable component. Even a perfect ADC has an error of plus or minus half the least-significant bit, and practical ones, especially in the rarefied atmosphere of 14 bits, have more. When I have modeled cameras, I often use mean ADC errors of 0.5 to 1.5 LSBs, in addition to the worst-case half-LSB theoretical error.

That aside, you're doing a great job of explaining complex things very well in this thread, as usual.

Jim

[JoeKitchen]

Wow, lots of technical stuff I don't understand.

If CMOS can be done better, than sure, lets have it.  Only I want to see the same or better quality of color as we get with a CCD; so far I haven't. 

[ErikKaffehr]

Well, it is technical stuff. The difference between CCD and CMOS is pure physics.

As Michael stated, colour rendition has nothing to do with CCD or CMOS, but very much on CFA (Color Filter Array) and CFA design may depend on who is making that sensor to which specifications. After CFA design is done it is mostly about interpretation of data. Colour conversion matrix and things like that. Once basic image rendition is in place the conversion profiles can be tweaked.

So there are a lot of variables involved, and it is mostly science.

Best regards
Erik

Wow, lots of technical stuff I don't understand.

If CMOS can be done better, than sure, lets have it.  Only I want to see the same or better quality of color as we get with a CCD; so far I haven't. 

[Paul2660]

Hi Jim

A question: It seems that the Nikon cameras having Sony sensors are "low ISO DR kings", the D3S and D4 have external converters and very high ISO capability but cannot achieve the same DR as the Sony sensors. My impression is that the only explanation is that the sensors give very clean signal, but the ADCs have significant noise. So the clean signal from the CMOS sensor is going to variable gain amplifiers, increasing amplification reduces shadow noise, but pushes some of the signal beyond ACD range. Would the ACD be 14-bit "clean", those sensors would also excel with low ISO DR. Is that correct?

It's my understand that all modern Nikon cameras are using a Sony Sensor, as Nikon does not have a fab plant like Canon.  It's my understanding Nikon has never produced a sensor, even back to the D1 in 1999.  I may have this wrong.  

The sensors in the D4, D4s may have tuned more to the high iso than producing the low iso DR.  

Paul

[ErikKaffehr]

Hi,

That is not correct. D3 and D4 sensors are not made by Sony. My guess is that they are built by Aptina.

Best regards
Erik

It's my understand that all modern Nikon cameras are using a Sony Sensor, as Nikon does not have a fab plant like Canon.  It's my understanding Nikon has never produced a sensor, even back to the D1 in 1999.  I may have this wrong.  

The sensors in the D4, D4s may have tuned more to the high iso than producing the low iso DR.  

Paul

[Paul2660]

Eric:

My oversight for sure.  Did some research as I should have before.

 D4 Sensor is the NC81366W fabricated for Nikon by foundry partner Renesas using 0.25 µm design rules.

Not sure who they picked for the D4s.

Paul

[Jim Kasson]

Thanks for making that point, I have very limited knowledge of electronics. Well, I have limited knowledge about most of things ;-)

Well, if that's true, you've had me fooled all along.

The point I tried to make is that ADCs just convert a voltage into numbers. Clearly there is noise and un-linearity, but I would guess that any of that would apply to low signal levels (that is shadows/darks) and the quality of the ADC plays a much smaller role on mid tones where the sampled analogue signal is quiet strong. So I don't think that "MFD magic" comes from better ADCs.

You are right there is an error component that is the same regardless of signal level, and thus that part is thus more important at low signal levels. When using a ramp ADC, the nonlinearlities can be seen at high levels, but my measurements indicate that cameras, whether CCD or CMOS, are in general remarkably linear. So your conclusion that the ADC has nothing to do with CMOS/CCD differences is almost certainly right.

A question: It seems that the Nikon cameras having Sony sensors are "low ISO DR kings", the D3S and D4 have external converters and very high ISO capability but cannot achieve the same DR as the Sony sensors. My impression is that the only explanation is that the sensors give very clean signal, but the ADCs have significant noise. So the clean signal from the CMOS sensor is going to variable gain amplifiers, increasing amplification reduces shadow noise, but pushes some of the signal beyond ACD range. Would the ACD be 14-bit "clean", those sensors would also excel with low ISO DR. Is that correct?

I don't know that the D3s and the D4 use external converters. I had assumed that they did not since they are pretty close to "ISOless". I don't think the ADC itself is the source of most of the read noise, since turning down the ISO -- and thus the gain -- doesn't affect the SNR very much. If we're looking for culprits in CCD cameras, I'd go with two: the repeated shifting that's already been mentioned, and the opportunity for picking up noise in getting the signal to an off-chip ADC. The odds of getting the signal to the ADC cleanly are much better if everything is on the same chip. If there's an off-sensor ADC and a variable gain on-sensor amplifier, there could be a noise win in turning up the gain of the amplifier if you can stand the loss in DR. In cameras with off-sensor ADCs, I don't know where the programmable gain amps are. In a completely ISOless camera, increasing the gain doesn't reduce the shadow noise. In these cameras, and indeed in all cameras I've tested, you'll get the best DR at base ISO.

That's kind of turgid; sorry. Ask away if I'm unclear.

Jim

[BJL]

If A/D is on chip or on board, does that make a difference for the end user? (I don't know)

The primary difference between CCDs and active pixel CMOS sensors is not where the A/D is done: Canon for example still uses off-board ADC, like all CCD cameras do.

Instead most basic difference is that an active pixel CMOS sensor reads out each photosite by transferring the signal directly to the edge of the sensor, and amplifying it in the process, which helps to reduce the effect of subsequent noise in the analog electronic signal path.  I see no good reason why anyone remains nostalgic for the older, slower, noisier, unamplified approach of a CCD.  In particular, Ronald's idea of a CCD with on-chip ADC might be viable, but it would not address the primary disadvantage of CCD's.

On the other hand, most recent active pixel sensor designs [Sony EXMOR, recent Panasonic 4/3" sensors, the Aptina sensors in Nikon One cameras, the CMOSIS sensor of the newest Leica M, etc.] also then do the ADC on the sensor, and in fact do it with an ADC unit at the end of each column of photosites, and this early ADC seems to help further with noise reduction, by avoiding the need for the analog signal to be transferred along the sensor's edge and beyond.


CCDs might be better suited to small volume products: apparently, once a basic CDD photosite is designed, it is relatively easy to lay out sensors of various shapes and sizes using that photosite design, whereas each different shape and size of active pixel CMOS sensor requires more new design work.

[EricWHiss]

Does anyone know if there is any cost advantage to making larger sensors with one flavor or the other - CCD or CMOS?  Perhaps differences in yield or other factor?

[Iluvmycam]

With the introduction of CMOS sensors into the medium-format world, what does the future look like for CCD sensors?

In your view, is there a CCD look or is it something that is vastly overvalued?

Is CMOS in your future or are you one of the photographers who is not so impressed with the latest CMOS medium-format sensor offerings?

What is your view in general about the impressions of other photographers about the CMOS sensors? Is the problem only the price? Do you predict a good future or a serious flop of the new equipment?

I think they have a look to them. But my experience is limited.

[torger]

An interesting aspect of large CCDs is that they are sold in different quality levels depening on how many pixels that are broken -- more broken pixels, cheaper sensor. Yes, your digital back probably has a few, a part of "calibration" is to map out the broken pixels which then are interpolated by neighboring working pixels.

I don't know if CMOS sensors are sold in the same way, and if that limits yield.

[ErikKaffehr]

Hi,

That was discussed in one of the LuLa videos (#14), at that time they had something about 7% rejection rate. That was a few years ago on 22 MP chips.The chips that were accepted had typically about two dead columns and perhaps around 300 dead pixels. These are mapped out in raw processing.

That was some time ago, so I guess statistics may be better.

I don't know if CMOS is different, but a manual I have for my Sony's says that the camera checks the sensor each month and maps out dead pixels.

Best regards
Erik

An interesting aspect of large CCDs is that they are sold in different quality levels depening on how many pixels that are broken -- more broken pixels, cheaper sensor. Yes, your digital back probably has a few, a part of "calibration" is to map out the broken pixels which then are interpolated by neighboring working pixels.

I don't know if CMOS sensors are sold in the same way, and if that limits yield.

[bcooter]

I think they have a look to them. But my experience is limited.

With the CCD cameras I've owned and still use, I see a difference, our crew sees a difference, our retoucher sees a difference, when client's select images from our portfolio about 70% are from CCD images and since more than half of what we show is from cmos cameras, that's interesting.  

I know the assistants see a difference because they all ask to borrow the contax/phase and now the leica and they don't ask unless they really want something.

Though except for certain situations, I'm off the ten billion iso, 600 frames a set squirrel wheel.  

Due to budget and time restraints I got into that but 400 frames of junk is 400 frames of junk.   I'd rather have 20 good options that work.

I'm also off the we'll fix it in post style of working.    That only goes so far and post production should be a part of the beginning creative brief, not a band aid to fix something.

To me CCD cameras work great in post processing.

Whether the look comes from a filter array, the convertor or the sensor I don't really care.   I just know what I see and I see it from the cropped sensor M8 to the p21+, p30+, Aptus 22 and the Leica S2.

In fact I bought the Leica because I knew cmos was coming and I thought I should get in while there was still something left to get.    

A great byproduct of the S2 was how well it handled HMI and even LED lighting.  We do a lot of parallel productions with motion and stills and usually my cmos camera files look washed out and thin with hmi lighting, the ccd has bit and color.

All of the samples I've posted from the S2 were continuous lighting, mostly hmi.



and this is the original crop



This was a one off from the same session with the oly em-1 and it's pretty it worked, but it's a much more fragile file than the Leica and much more difficult to separate colors.


Anyway, we shoot a lot of images, This is 9 months of master raws

and I stopped counting numbers when we crossed the 300 terabyte count.  (which obviously covers a number of years). though I based my opinion ONLY on my experience.  

I don't know or care how other photographers get their results, that's none of my business.

But let me be clear, that doesn't mean I'm right, it just means I'm right for me.   We all work differently, we all have different end agendas and obviously different opinions.

IMO

BC

[ErikKaffehr]

Hi,

This link http://image-sensors-world.blogspot.se/2013/03/cmos-sensors-market-grows-ccd-market.html says:

"CMOS image sensors were [one of the] fastest growing product category in 2012 with sales rising 22% to a new record-high $7.1 billion, blowing past the previous peak of $5.8 billion set in 2011. Since the 2009 downturn year, CMOS image sensor sales have climbed 85% due to the strong growth of embedded cameras used in smartphones and portable computers (including tablets) and the expansion of digital imaging into more systems applications. CMOS designs are now grabbing large chunks of marketshare from CCD image sensors, which are forecast to see revenues decline by a CAGR of 2.4% between 2012 and 2017. Sales of CMOS imaging devices are projected to grow by a CAGR of about 12.0% in the forecast period and account for 85% of the total image sensor market versus 15% for CCDs in 2017. This compares to a 60/40 split in 2009."

So CMOS is growing fast while CCD is in slow decline. CMOS development may driven by smart phones and webcams, meaning small sensors, but development is scaling up well to larger sensors which are much more profitable.

I guess that essentially says that CMOS is where the money goes.

It seems that Sony dominates both CCD and CMOS market: http://image-sensors-world.blogspot.se/2012/05/tsr-market-share-data-for-2h-2011.html

Best regards
Erik

With the introduction of CMOS sensors into the medium-format world, what does the future look like for CCD sensors?

In your view, is there a CCD look or is it something that is vastly overvalued?

Is CMOS in your future or are you one of the photographers who is not so impressed with the latest CMOS medium-format sensor offerings?

What is your view in general about the impressions of other photographers about the CMOS sensors? Is the problem only the price? Do you predict a good future or a serious flop of the new equipment?

[MrSmith]

  I buy about 4 drives a year, but then some shoots only involve 20 frames, I am yet to break/replace a shutter.