Canon 7D Mark II - Finally using Canon's newer fab?

[AcutancePhotography]

I was not aware of http://sensorgen.info/ Thanks for posting the link.

 

[Lee Jay]

That's nothing: we should all be still using Nikon D70s, QE two hundred and something percent !

Pfft. D2Xs is 476%.

http://www.sensorgen.info/NikonD2Xs.html

 

[ajfotofilmagem]

That's nothing: we should all be still using Nikon D70s, QE two hundred and something percent !

Pfft. D2Xs is 476%.
http://www.sensorgen.info/NikonD2Xs.html

If the Nikon D2Xs has quantum efficiency of 476%, there is a spell that multiplies the luminous rays. I call this magic of
"RAW has already been cooked."

 

[Coldhands]

I would like nothing more that for Canon to finally move on from their 500nm process, but I can't help but think that if they went to all the trouble to implement a new fab, that they would use it to create an all-new sensor that provides market separation from the 70D. In other words, why spend so much on R&D for your top-of-the-range APS-C body only for people to think "oh, its just the same as the old sensor with a few tweaks."

Then again my knowledge of CMOS production is elementary, and there could be a perfectly good reason to use a new process on existing architecture. Here's hoping that Chipworks analyses one so we can know for sure.

Intel, the leading CPU manufacturer, uses a TIC TOC method for their CPUs, where one changes the the manufacturing process and the other changes the architecture, but Intel does not change both at same time, too difficult / risky. They prefer to consolidate the manufacturing process first, and only on next iteration change the architecture.
Let's hope the same is happening with Canon, but I still find too good to be true.

Ah, right. I'd heard Intel use the "tic toc" term with respect to development, but didn't know what it actually referred to. Figured it was just a bit of marketing speak. Thanks for the info.

 

[Lee Jay]

That's nothing: we should all be still using Nikon D70s, QE two hundred and something percent !

Pfft. D2Xs is 476%.
http://www.sensorgen.info/NikonD2Xs.html

If the Nikon D2Xs has quantum efficiency of 476%, there is a spell that multiplies the luminous rays. I call this magic of
"RAW has already been cooked."

Yes, I believe I said that just above.

http://www.canonrumors.com/forum/index.php?topic=23616.msg461491#msg461491

 

[jrista]

I don't think sensor QE tells us much. Consider the following Nikon camera models:

Camera QE
D3 90%
D3X 42%
D3s 57%
D4 52%
D4s 52%
D610 51%
D800 56%
D810 47%

Just goes to show, the further the measurement moves away from actual best processed images, the less interesting and relevant the metric becomes. D810 9% less than the D800 and 12% less the the 7D MkII.

That's nothing: we should all be still using Nikon D70s, QE two hundred and something percent !

I'm not convinced on how this quantum efficiency is calculated using SNR. I have both the 5DII ( QE 31 %) and 6D ( 49%) and am struggling to see how this is relevant in practice. Perhaps in low light there is an advantage in over exposing the 5DII slightly compared with the 6D, but it's unclear. I presume that if enough photons have been retained by the pixel to record a 100% accuracy and converted to an appropriate charge, that is all that matters: once it's 100% accurate it doesn't matter what percentage it was of the total available.

I guess that as QE goes up ( in the way it is calculated) the sensor can withstand less exposure, but this doesn't necessarily relate to pushing under exposure. This then also allows more acceptable results at higher ISOs. That also matches the characteristics of the 5DII and 6D.

However I do often look back at images from the Nikon D70 that I had - replaced it with a D200 - and think 'that D70 was a damn fine camera'.

Just a bit of additional information: I believe the 'equivalent QE' of film was in the region of 10%.

I never posted this as a means of bringing up exposure pushing.


All I see is that the quantum efficiency of a Canon sensor has jumped quite a bit from it's predecessors. As I've been watching Canon for some time, and have seen 59% Q.E. on some of their smaller sensors that use the 180nm/Cu interlink fab process, and have never seen more than a 2% jump in Q.E. between any successive APS-C or FF generations in the past, it seems to me as though Canon has started moving to a newer process.


I think the idea that Canon may be borrowing Intel's "TIC TOC" approach is reasonable. Canon is currently moving production to a new fab, and once they have settled in on that new fab, they will ramp up production on a newer sensor design. (My guess is that's actually already happening...a layered sensor that will hopefully be revealed sometime Q1/Q2 2015.)


As for the value of Q.E....more Q.E. means less gain for a given exposure (at any ISO setting.) That's a good thing, regardless of how you slice it. It's better for high ISO, and is probably one of the key reasons the 7D II gained as much as it did at the high ISO end of things. It's also better for low ISO, however the benefit at low ISO for Canon is currently being masked by their read noise (which I am totally convinced now is due to the use of only a handful of high frequency ADC units off die.)


Regarding accuracy, I am not sure exactly how that plays into things. Quantum efficiency is a rate...it determines the rate of photon to electron conversion. At 59% Q.E. then roughly 59% of the photons incident at the photodiode actually push the potential high enough to free an electron. In other words, for every 100 incident photons, you end up with 59 electrons worth of charge. Accuracy isn't a factor here.


Higher Q.E. means more photons are collected in a given time span. So, if for a 1/30th exposure there are 35,000 photons median incident at each photodiode, then a sensor with 40% Q.E. will end up with a charge of 14,000e-, while a sensor with 59% Q.E. will end up with a charge of 20,650e-. If the FWC of both sensors is 30,000e-, then the former will need more gain to produce the right output voltage than the latter (in other words, you would need to use about 2/3rds stop higher ISO, or simply more gain at base ISO), and therefor will be noisier.


As far as I care, it doesn't matter if you prefer high ISO or low ISO or use a mix of the two...higher Q.E. means lower noise, across the board. At low ISO on Canon sensors, thanks to post-sensor read noise, the benefits at low ISO are swamped. However Canon still has relatively low read noise at higher ISO, so the benefits of higher Q.E. should be apparent. In the case of the 7D II...I think they are, at around 2/3rds of a stop improvement or so.

 

[Suri JV]

This is my first post but I always read the discussions in the CR with interest and benefit. This one about QE also made clear to me some new ideas. Can we take it that bigger pixels are also better for that reason? I shall be happy to have a clear answer. Thank you.

 

[DominoDude]

I don't think sensor QE tells us much. Consider the following Nikon camera models:

Camera QE
D3 90%
D3X 42%
D3s 57%
D4 52%
D4s 52%
D610 51%
D800 56%
D810 47%

Just goes to show, the further the measurement moves away from actual best processed images, the less interesting and relevant the metric becomes. D810 9% less than the D800 and 12% less the the 7D MkII.

That's nothing: we should all be still using Nikon D70s, QE two hundred and something percent !

I'm not convinced on how this quantum efficiency is calculated using SNR. I have both the 5DII ( QE 31 %) and 6D ( 49%) and am struggling to see how this is relevant in practice. Perhaps in low light there is an advantage in over exposing the 5DII slightly compared with the 6D, but it's unclear. I presume that if enough photons have been retained by the pixel to record a 100% accuracy and converted to an appropriate charge, that is all that matters: once it's 100% accurate it doesn't matter what percentage it was of the total available.

I guess that as QE goes up ( in the way it is calculated) the sensor can withstand less exposure, but this doesn't necessarily relate to pushing under exposure. This then also allows more acceptable results at higher ISOs. That also matches the characteristics of the 5DII and 6D.

However I do often look back at images from the Nikon D70 that I had - replaced it with a D200 - and think 'that D70 was a damn fine camera'.

Just a bit of additional information: I believe the 'equivalent QE' of film was in the region of 10%.

I never posted this as a means of bringing up exposure pushing.


All I see is that the quantum efficiency of a Canon sensor has jumped quite a bit from it's predecessors. As I've been watching Canon for some time, and have seen 59% Q.E. on some of their smaller sensors that use the 180nm/Cu interlink fab process, and have never seen more than a 2% jump in Q.E. between any successive APS-C or FF generations in the past, it seems to me as though Canon has started moving to a newer process.


I think the idea that Canon may be borrowing Intel's "TIC TOC" approach is reasonable. Canon is currently moving production to a new fab, and once they have settled in on that new fab, they will ramp up production on a newer sensor design. (My guess is that's actually already happening...a layered sensor that will hopefully be revealed sometime Q1/Q2 2015.)


As for the value of Q.E....more Q.E. means less gain for a given exposure (at any ISO setting.) That's a good thing, regardless of how you slice it. It's better for high ISO, and is probably one of the key reasons the 7D II gained as much as it did at the high ISO end of things. It's also better for low ISO, however the benefit at low ISO for Canon is currently being masked by their read noise (which I am totally convinced now is due to the use of only a handful of high frequency ADC units off die.)


Regarding accuracy, I am not sure exactly how that plays into things. Quantum efficiency is a rate...it determines the rate of photon to electron conversion. At 59% Q.E. then roughly 59% of the photons incident at the photodiode actually push the potential high enough to free an electron. In other words, for every 100 incident photons, you end up with 59 electrons worth of charge. Accuracy isn't a factor here.


Higher Q.E. means more photons are collected in a given time span. So, if for a 1/30th exposure there are 35,000 photons median incident at each photodiode, then a sensor with 40% Q.E. will end up with a charge of 14,000e-, while a sensor with 59% Q.E. will end up with a charge of 20,650e-. If the FWC of both sensors is 30,000e-, then the former will need more gain to produce the right output voltage than the latter (in other words, you would need to use about 2/3rds stop higher ISO, or simply more gain at base ISO), and therefor will be noisier.


As far as I care, it doesn't matter if you prefer high ISO or low ISO or use a mix of the two...higher Q.E. means lower noise, across the board. At low ISO on Canon sensors, thanks to post-sensor read noise, the benefits at low ISO are swamped. However Canon still has relatively low read noise at higher ISO, so the benefits of higher Q.E. should be apparent. In the case of the 7D II...I think they are, at around 2/3rds of a stop improvement or so.

Interesting conclusion! Could, in your mind, something besides the design and positioning (in the signal chain) of the A/D converters explain the structure and level of the read noise for example?

 

[Don Haines]

I don't think sensor QE tells us much. Consider the following Nikon camera models:

Camera QE
D3 90%
D3X 42%
D3s 57%
D4 52%
D4s 52%
D610 51%
D800 56%
D810 47%

Just goes to show, the further the measurement moves away from actual best processed images, the less interesting and relevant the metric becomes. D810 9% less than the D800 and 12% less the the 7D MkII.

That's nothing: we should all be still using Nikon D70s, QE two hundred and something percent !

I'm not convinced on how this quantum efficiency is calculated using SNR. I have both the 5DII ( QE 31 %) and 6D ( 49%) and am struggling to see how this is relevant in practice. Perhaps in low light there is an advantage in over exposing the 5DII slightly compared with the 6D, but it's unclear. I presume that if enough photons have been retained by the pixel to record a 100% accuracy and converted to an appropriate charge, that is all that matters: once it's 100% accurate it doesn't matter what percentage it was of the total available.

I guess that as QE goes up ( in the way it is calculated) the sensor can withstand less exposure, but this doesn't necessarily relate to pushing under exposure. This then also allows more acceptable results at higher ISOs. That also matches the characteristics of the 5DII and 6D.

However I do often look back at images from the Nikon D70 that I had - replaced it with a D200 - and think 'that D70 was a damn fine camera'.

Just a bit of additional information: I believe the 'equivalent QE' of film was in the region of 10%.

I never posted this as a means of bringing up exposure pushing.


All I see is that the quantum efficiency of a Canon sensor has jumped quite a bit from it's predecessors. As I've been watching Canon for some time, and have seen 59% Q.E. on some of their smaller sensors that use the 180nm/Cu interlink fab process, and have never seen more than a 2% jump in Q.E. between any successive APS-C or FF generations in the past, it seems to me as though Canon has started moving to a newer process.


I think the idea that Canon may be borrowing Intel's "TIC TOC" approach is reasonable. Canon is currently moving production to a new fab, and once they have settled in on that new fab, they will ramp up production on a newer sensor design. (My guess is that's actually already happening...a layered sensor that will hopefully be revealed sometime Q1/Q2 2015.)


As for the value of Q.E....more Q.E. means less gain for a given exposure (at any ISO setting.) That's a good thing, regardless of how you slice it. It's better for high ISO, and is probably one of the key reasons the 7D II gained as much as it did at the high ISO end of things. It's also better for low ISO, however the benefit at low ISO for Canon is currently being masked by their read noise (which I am totally convinced now is due to the use of only a handful of high frequency ADC units off die.)


Regarding accuracy, I am not sure exactly how that plays into things. Quantum efficiency is a rate...it determines the rate of photon to electron conversion. At 59% Q.E. then roughly 59% of the photons incident at the photodiode actually push the potential high enough to free an electron. In other words, for every 100 incident photons, you end up with 59 electrons worth of charge. Accuracy isn't a factor here.


Higher Q.E. means more photons are collected in a given time span. So, if for a 1/30th exposure there are 35,000 photons median incident at each photodiode, then a sensor with 40% Q.E. will end up with a charge of 14,000e-, while a sensor with 59% Q.E. will end up with a charge of 20,650e-. If the FWC of both sensors is 30,000e-, then the former will need more gain to produce the right output voltage than the latter (in other words, you would need to use about 2/3rds stop higher ISO, or simply more gain at base ISO), and therefor will be noisier.


As far as I care, it doesn't matter if you prefer high ISO or low ISO or use a mix of the two...higher Q.E. means lower noise, across the board. At low ISO on Canon sensors, thanks to post-sensor read noise, the benefits at low ISO are swamped. However Canon still has relatively low read noise at higher ISO, so the benefits of higher Q.E. should be apparent. In the case of the 7D II...I think they are, at around 2/3rds of a stop improvement or so.

exactly!

The more electrons you have, the less gain you need. The problem with gain is that you are amplifying the signal AND you are amplifying the noise... realistically, you get nothing out of gain....

Imagine you have the worlds greatest A/D converter and it can count every single electron... You have a sensor operating at 30% QE and it collects 1,000 electrons. your A/D converter spits out a 10 bit number..... but there is noise in the circuitry and your last 4 bits are toggling at random due to that noise.... this gives you 6 bits of signal to noise ratio... If you amplify the signal by a factor of 8, you now have a 13 bit number, but the noise is also amplified and you have 7 bits of noise and your signal to noise ratio remains as 6 bits.

You have three things you can do to improve this.

The first is to get rid of some of the noise. The closer to the source the pixel is sampled, the less noise there will be. This is one of the big reasons why it is best to sample it on the sensor. The other reason is rthat on the sensor you can have many times more A/D units than on a separate chip (connections to an external A/D are restricted by wiring complexity) and with more converters, they can run at slower speeds, and that reduces noise and improves accuracy. In our above example, this might knock off a bit or two worth of noise and improve your signal to noise ratio.

The second thing you can do is to get a bigger pixel. A bigger pixel gathers more electrons.. A pixel twice as large will now gather 2000 electrons, as opposed to the original 1000 electrons, and your A?D converter will now spit out an 11 bit number instead of a 10 bit number, and you have just gained a bit of S/N ratio. This is the reason why FF cameras work better than crop cameras in poor light.

The third thing you can do is to improve quantum efficiency. If you bumped the sensor from 30% up to 60% efficiency, you get twice as many electrons, and like above, you gain another bit of signal. ... and this is why the bump up in QE has some of us quite excited for the future.

Put all of this together and you have some very significant improvements. You might just end up with the next FF sensor having 2 stops or more improvement over the 5D3.....

 

[Woody]

Higher Q.E. means more photons are collected in a given time span.

Not quite. QE is the number of electrons created per incident photon. If QE > 100%, then more than 1 electron is created per photon.

This begs the question: if the D3 has QE of 90%, then why doesn't Sony/Nikon maintain that number for later models (e.g., D4, D800, D610 etc etc)?

 

[jrista]

Higher Q.E. means more photons are collected in a given time span.

Not quite. QE is the number of electrons created per incident photon. If QE > 100%, then more than 1 electron is created per photon.

This begs the question: if the D3 has QE of 90%, then why doesn't Sony/Nikon maintain that number for later models (e.g., D4, D800, D610 etc etc)?

Sure. When I say "collected", I mean the photons actually released an electron. Photons can reflect, convert to heat, etc. and when they do, they are not actually "collected" or converted.


I have wondered if more than 100% Q.E. is possible with a sensor. I read something a while back that >100% Q.E. has been achieved with new materials for solar panel cells...but I've never heard of an image sensor that had more than 100% Q.E. Getting to 90% is actually quite difficult, and I don't know of many Grade 0 CCD sensors that top 90%.


Regarding Sensorgen.info data, he regenerated it all recently. After the site went down, a lot more data was brought over. I think there is a bug in whatever code brings the data over, because some sensors have wild read noise values and several hundred percent Q.E. I don't believe any of those numbers are valid...I think they are flukes generated by a buggy algorithm. I am not sure that the D3 has 90% Q.E....DPR lists it's Q.E. at 40% on this page:


http://www.dpreview.com/forums/post/34581894


That seems to be more in line with the generation the D3 came from.


Given this, it is entirely possible that the 59% Q.E. value for the 7D II is also wrong.

 

[Woody]

Regarding Sensorgen.info data, he regenerated it all recently. After the site went down, a lot more data was brought over. I think there is a bug in whatever code brings the data over, because some sensors have wild read noise values and several hundred percent Q.E. I don't believe any of those numbers are valid...I think they are flukes generated by a buggy algorithm.

If so, how then can we trust their QE value of 59% for 7D2? Or for that matter, any sensor listed on that site?

 

[jrista]

Regarding Sensorgen.info data, he regenerated it all recently. After the site went down, a lot more data was brought over. I think there is a bug in whatever code brings the data over, because some sensors have wild read noise values and several hundred percent Q.E. I don't believe any of those numbers are valid...I think they are flukes generated by a buggy algorithm.

If so, how then can we trust their QE value of 59% for 7D2? Or for that matter, any sensor listed on that site?

If you would read everything, I already said as much:

Given this, it is entirely possible that the 59% Q.E. value for the 7D II is also wrong.

 

[jrista]

I shouldn't laugh... :

Sorry jrista.

I still think the 7D II numbers are valid. If you look at the data for all the cameras that have "odd" Q.E. numbers, they also all have odd read noise/saturation numbers. The D3 is no exception...oddly, at ISO 6400, the numbers get wonky. I am pretty sure it's those perturbations in the data that are causing crazy Q.E. numbers. The 7D II read noise levels and saturation points look fine, so I am pretty sure the data is accurate.

 

[lintoni]

Well, let's hope so, and hope this is a step by Canon to improve the readout of their sensors.

 

[jrista]

Well, let's hope so, and hope this is a step by Canon to improve the readout of their sensors.

I am really hoping they ditch their current sensor designs entirely and go with something radically new. A layered sensor with on-die readout would be awesome.

 

[Lee Jay]

Regarding Sensorgen.info data, he regenerated it all recently. After the site went down, a lot more data was brought over. I think there is a bug in whatever code brings the data over, because some sensors have wild read noise values and several hundred percent Q.E. I don't believe any of those numbers are valid...I think they are flukes generated by a buggy algorithm.

There's evidence to support this.

http://www.sensorgen.info/NikonD2X.html
https://web.archive.org/web/20140715194150/http://www.sensorgen.info/NikonD2X.html

 

[Aglet]

Maybe it's just from the CFA filter having been made even more color blind.

The low color measurement on DxOmark does seem to indicate they may have picked up a little QE from a looser CFA.

 

[Lee Jay]

Maybe it's just from the CFA filter having been made even more color blind.

The low color measurement on DxOmark does seem to indicate they may have picked up a little QE from a looser CFA.

I don't think one has to do with the other. I've been told in the past that the QE modeling approach makes it effectively ignore the CFA.

 

[jrista]

Maybe it's just from the CFA filter having been made even more color blind.

The low color measurement on DxOmark does seem to indicate they may have picked up a little QE from a looser CFA.

I don't think one has to do with the other. I've been told in the past that the QE modeling approach makes it effectively ignore the CFA.

Agreed. I believe the Q.E. is referring to the actual response of the silicon itself (that's usually how sensor manufacturers report it...as the silicon's response to light). That is the Q.E. of incident photons that make it all the way to the sensor.


So, to be clear, that means that even if you had a 100% response in the silicon itself (at 565nm, green light...the Q.E. is usually for green light), you could still increase overall light gathered by improving your filtering, or by eliminating filtering and going with MCS, or by improving microlens design, or by using ISOCELL in a BSI design, etc. Anything that increases the incident photon count at the photodiode will still increase overall efficiency...for whatever Q.E. you have.