PDA

View Full Version : Interesting detective work about Canon image sensors



arash_hazeghi
10-27-2012, 04:07 PM
http://www.chipworks.com/blog/technologyblog/2012/10/24/full-frame-dslr-cameras-canon-stays-the-course/

This basically shows why Canon sensor technology has not scaled. A bit outside the scope of this forum but for people who are interested...

Indranil Sircar
10-27-2012, 04:22 PM
Good stuff.. thanks for sharing.

Kerry Perkins
10-27-2012, 04:35 PM
Very interesting Arash. Thanks for posting.

John Chardine
10-27-2012, 06:40 PM
This thread should be posted in the Science of Photography forum. Having said this, the three parts are very poor examples of good science writing.

arash_hazeghi
10-27-2012, 07:21 PM
This thread should be posted in the Science of Photography forum. Having said this, the three parts are very poor examples of good science writing.

I don't think so. I think it is an excellent analysis (for those familiar with image sensor technology which is a matter of electrical engineering), it is not a theoretical science article.

Roger Clark
10-27-2012, 08:31 PM
I found the information of interest. Thanks Arash for posting. For most photographers, it will likely not mean much. However, the information given, I think, adds weight to rumours that Canon is about to introduce a new sensor line with much higher megapixel count. To do that they'll likely need to move off of the 0.5-micron process like Sony.

I agree this thread should be moved to the Science of Photography forum.

Roger

arash_hazeghi
10-27-2012, 08:47 PM
I think the cost of a new fab is prohibitive, they will need to put in a lot of resources plus risk having lower yields initially and thus higher prices than competition-which they already suffer from. Of course I am not in a position to calculate what is best for Canon but from the evidence it seems like the easiest way out is to just buy from Sony like Nikon. This way they also gain access to the latest technology that is patented by and exclusive to Sony including column-parallel ADC/ back-side illumination etc.

The 0.5um process is the same process used for EOS20D back in 2004 and has not scaled.

Looking at the SEMs it is striking to see the difference between the fill factor in Sony vs. Canon sensors. Canon 5T logic is large occupying over 50% of the pixel area! this means the active diode area for 1DX is actually pretty close to that of D800 despite the latter having a smaller pixel pitch. The effective pixel size is very close between the two cameras.


If the forum is not appropriate please move this thread.

Jon Rista
10-28-2012, 12:47 AM
@Arash: Wouldn't the use of lightpipe technology in a 0.18um sensor largely negate the need for a BSI design, and avoid the problems inherent in BSI manufacture? I hadn't heard of a lightpipe before, but it sounds like the gains, when paired with an appropriately optimized microlens, are very similar to BSI. As for column-parallel, in one of Canon's press releases about their prototype 120mp APS-H they seemed to indicate there was some kind of on-die hyperparallel "image processing" (as they called it) to support the 9.5fps readout rate...sounded a lot like CP-ADC to me.

I wonder if Canon really needs to break their "all done in-house" rules and buy Sony if they already have similar technologies...

BTW, Lightpipe High Refractive Index Filling for Q.E. gains: http://www.silecs.com/download/CMOS_image_sensor_with_high_refractive_index_light pipe.pdf

arash_hazeghi
10-28-2012, 01:07 PM
Hi Jon,

As I understand most manufacturers already use a nitride layer on top of the cell (evident from Chipworks SEMs). The layer denoted by "optical pad" is nitride. Canon also put a lot of effort into MLA in the past, they have patented gapless MLA etc. which is supposed to improve absorption by a few percent, however they never published the data on this. Keep in mind the benefit from index of refraction matching/optimization or even placement of waveguides on pixels (it has been done) is selective in wavelength vs. BSI improves absorption across the entire spectrum so it is more effective in general for RGB sensors.


It is true that BSI is more effective for sensors with smaller pixels (around 1-2um) where M1 lines start to reduce FF, but I think the large-pixel sensor can also benefit by putting the metal lines on the back. The only factor here is cost, if it is reduced then all sensors will go BSI.

With all the recent trends, I no longer have confidence in Canon's press release etc. unless objective data is published. For e.g. They claim many things like increased DR, lower noise etc. for the case of 5D3 which we saw was not true. Sony's CP-ADC technology is robust clever method and it is now incorporated in all of their products with proven benefits.

John Chardine
10-28-2012, 07:06 PM
Reading between the lines, I assume that "scaling" has something to do with the phrase "process technology" or "process generation" in the article. However, the connection is nowhere made explicit in this thread or in the article. To be fair the article does not talk about scaling per se but it does use the "process" word several times, e.g.:

"On the process side, the 1D X is remarkable in that Canon continues to stay with the 0.5 µm process generation it has used for every APS-C and FF device analyzed."

Unfortunately, the context of the word "process" is not given and so the whole article falls apart. Good science writing defines and explains terms so that any intelligent person can pick up the article and get something out of it, without reading between the lines.

arash_hazeghi
10-28-2012, 07:56 PM
Reading between the lines, I assume that "scaling" has something to do with the phrase "process technology" or "process generation" in the article. However, the connection is nowhere made explicit in this thread or in the article. To be fair the article does not talk about scaling per se but it does use the "process" word several times, e.g.:

"On the process side, the 1D X is remarkable in that Canon continues to stay with the 0.5 µm process generation it has used for every APS-C and FF device analyzed."

Unfortunately, the context of the word "process" is not given and so the whole article falls apart. Good science writing defines and explains terms so that any intelligent person can pick up the article and get something out of it, without reading between the lines.

John,

No, Scaling is a rather well defined topic in semiconductor technology, it is not just process or printing a smaller circuit because things will not work, to learn more about scaling you can read the following: http://en.wikipedia.org/wiki/Moore%27s_law it is not a trivial topics, and it takes a PhD thesis to study only one small aspect of the multi-dimensional problem.

Of course Moore's law scaling applies to logic and memory but image sensors have similar trends, the technology that is used for image sensor fabrication is a about 10-12 years behind that of state-of-the art logic because the critical feature size is an order of magnitude larger.

The article is not written for people with no background in this topic to learn about image sensor technology. It is an engineering analysis like a medical report written by and for doctors, an average person will not understand it even if they are intelligent.

I think you are still confusing this engineering analysis with an article written in popular science or American scientific magazine for an average audience...it is not that :S3:

ChrisBeveridge
10-28-2012, 11:49 PM
John,

No, Scaling is a rather well defined topic in semiconductor technology, it is not just process or printing a smaller circuit because things will not work, to learn more about scaling you can read the following: http://en.wikipedia.org/wiki/Moore%27s_law it is not a trivial topics, and it takes a PhD thesis to study only one small aspect of the multi-dimensional problem.

Of course Moore's law scaling applies to logic and memory but image sensors have similar trends, the technology that is used for image sensor fabrication is a about 10-12 years behind that of state-of-the art logic because the critical feature size is an order of magnitude larger.

The article is not written for people with no background in this topic to learn about image sensor technology. It is an engineering analysis like a medical report written by and for doctors, an average person will not understand it even if they are intelligent.

I think you are still confusing this engineering analysis with an article written in popular science or American scientific magazine for an average audience...it is not that :S3:

I agree, this to me was a very well written article in terms of the audience (Engineers or those with some engineering understanding). Yes it may be posted to a blog on the internet but I know of several geological artistic that can be found the same way, and unless you have any kind of background in geology you will not understand what is being said (I am a geologist thus my use of that example).

I am curious about something tho...

Canon unveiled a 50MP sensor back in 2007 (I think) and then recently a year or so ago unveiled a 120MP sensor. Now granted neither of these sensors were designed for DSLR use, but for the medical industry and imaging technology they use there. Would this not demonstrate that Canon is already capable to breaking away from the .5um, or were these designed using the same technology as current DSLR sensors are using? I am not an eng. by any means but to me it seems that Canon has the technology and ability to produce state of the art imaging sensors, however, the production cost to (DSLR cost) might not seem profitable right now. A MRI machine costs upwards of 100K's of dollars where I believe these sensors are being utilized and would make it a profitable production but in a 6.8k dollar SLR I don't see that. Again I am not an Eng. but am interested in what y'all think. I know Sony is breaking the molds of the past and producing outstanding sensors that are preforming very well, but hasn't Canon already been doing this? To me it seems Canon is just not putting them into their DSLR's where Sony is now.

I could be totally wrong about everything and if that's so then please correct me. I like learning :)

arash_hazeghi
10-31-2012, 02:42 AM
Hi Chris,

Those were prototypes, yes you can print the pixels smaller but they won't work unless you have scaled your technology successfully, plus you must bring up the yield for manufacturing-not trivial. As for the prototypes, Canon never published any meaningful data from those so my guess they were just dummies (i.e. pixels are printed but they are not active) for some R&D or just for PR.

MRI systems do not use a CMOS image sensors, they are based on magnetic imagining.


Best

Arash

ChrisBeveridge
10-31-2012, 05:22 AM
Hi Chris,

Those were prototypes, yes you can print the pixels smaller but they won't work unless you have scaled your technology successfully, plus you must bring up the yield for manufacturing-not trivial. As for the prototypes, Canon never published any meaningful data from those so my guess they were just dummies (i.e. pixels are printed but they are not active) for some R&D or just for PR.

MRI systems do not use a CMOS image sensors, they are based on magnetic imagining.


Best

Arash

Arash,

Thanks for the clarification. I just remember seeing them but never heard anything more about them so I totally see what you were saying. Haha guess it worked on me in terms of Canon showing something big :P.

As for the MRI machine... I totally should have known that. Stupid me! haha. Thanks again for the correction I am that much smarter :D (I mean that sincerely).

Cheers,
Chris

Joćo Morgado
10-31-2012, 11:49 AM
Canon 52Mp APS-H CMOS Prototype, was created with 180nm fabrication process*... so, they have the technology since 2007, and they explored the 500nm technology to the extreme... so the question is... how long till we see SLR sensors made with 180nm technology?

* Page 3 http://www.imagesensors.org/Past%20Workshops/2007%20Workshop/2007%20Papers/076%20Iwane%20et%20al.pdf

Some say that the s100 sensor is already made with 180nm technology...

Best Regards,
Joćo Morgado

arash_hazeghi
10-31-2012, 12:18 PM
Thanks for posting. Do you have a link to the final publication ? (this looks like a draft with some figure data missing)

It doesn't however establish that Canon has 0.18um production technology as of today, you can run an R&D fab with much lower yield but you cannot manufacture it in large volume.

for small cameras I think Canon buy their sensors from Sony already.

Anyways, time will tell, they can't sit at 0.5um forever.

Joćo Morgado
10-31-2012, 12:30 PM
Sorry, this was the only one I saw...

Best Regards,
Joćo Morgado

Jon Rista
10-31-2012, 01:21 PM
Hi Jon,

As I understand most manufacturers already use a nitride layer on top of the cell (evident from Chipworks SEMs). The layer denoted by "optical pad" is nitride. Canon also put a lot of effort into MLA in the past, they have patented gapless MLA etc. which is supposed to improve absorption by a few percent, however they never published the data on this. Keep in mind the benefit from index of refraction matching/optimization or even placement of waveguides on pixels (it has been done) is selective in wavelength vs. BSI improves absorption across the entire spectrum so it is more effective in general for RGB sensors.


It is true that BSI is more effective for sensors with smaller pixels (around 1-2um) where M1 lines start to reduce FF, but I think the large-pixel sensor can also benefit by putting the metal lines on the back. The only factor here is cost, if it is reduced then all sensors will go BSI.


I understand that BSI is better overall...directly exposing the photodiode and burying all the lines behind would really be the way to go long-term, regardless of pixel size. I wonder if lightpipe is a viable, cheaper stepping stone on the way top ubiquitous BSI, though. Something else I noticed in all of the electron microscope scans is that the color filters seem to use rather "rough" material...it always seems rather granular compared to the rest of the material in the die. Would that not cause a fair amount of scattering of the remaining light that passes through them, making it that much more difficult to implement viable lightpipe technology?



With all the recent trends, I no longer have confidence in Canon's press release etc. unless objective data is published. For e.g. They claim many things like increased DR, lower noise etc. for the case of 5D3 which we saw was not true. Sony's CP-ADC technology is robust clever method and it is now incorporated in all of their products with proven benefits.

Well, the 5D III did realize some small DR gains at higher ISO, thanks to a higher S/N. I don't think Canon ever claimed improvements to low ISO DR, but then again, their improvements in higher ISO DR were minimal at best (and marginally better than Nikon DR at the same ISO's.) Sony Exmor DR owes thanks to more than just CP-ADC, I believe. They use a double layer of microlensing to improve Q.E., they already moved to a 180nm process which also helps improve Q.E by shrinking wiring relative to pixel area. They use a newer generation of technology for their correlated double sampling than Canon (I think Canon's patent is from the 1990's, Sony's is from only a few years ago.) I also believe Sony Exmor implements additional logic to normalize non-uniform pixel response by tuning each transistor relative to its neighbors (although that my only be implemented in the smaller size sensors), which helps reduce electronic noise even further. Sony has been innovating and implementing their innovations for years, which is a great thing. I wish Canon would do the same.

I do generally agree though, that Canon can't really be trusted in their press releases. They still proclaim to be the best at sensor manufacture, when they clearly are not. They have serious, and apparently growing (as to exhibit more in the 5D III than 5D II), problems with electronic noise (fixed pattern, banding) that can have a visible and detrimental effect on image quality right up into the midtones. There is no good way to get rid of it, either. If it exhibits, your just kind of screwed. They seem to have lost control of the negative perception a growing percentage of their customers have of the company, which really does not bode well for their future customer loyalty, future sales, and eventually future R&D budgets, especially if they refuse to show some competitive willpower and visibly start innovating better technologies that can go head-to-head with their largest competitors. I've invested a fair amount of money into my Canon kit, and I still believe their glass is the best. I desperately need to buy one of their supertelephoto lenses (500mm or 600mm f/4 L II), but I can't seem to actually pull the trigger and do it...I don't really know if Canon will actually be able to deliver better sensors with better IQ and progressive value overall in the future, or whether they will be totally eclipsed by Sony Exmor and the Sony/Nikon alliance. I have held off on buying anything for my Canon kit for about a year now, as I have this itch in my mind that there might be a need to sell my kit and move to Nikon at some point in the relatively near future...even though I greatly prefer Canon ergonomics and lenses.

I really hope they do something commercially useful with their 180nm technology, even if it is minor and low volume. At the very least, that would indicate they are still actually in the game, and that we can still hope for better value from the company in the future.

arash_hazeghi
11-02-2012, 01:16 PM
the CFA is made of polymer so it looks disordered. Typical CFA absorbs 1/2 of the light, it is a major factor in killing QE.

Unlike Canon, Sony EXMOR sensors do not need analog CDS unit as the reset noise is subtracted digitally for each pixel, this was the key innovation and the basis of their advantage. The details are explained nicely here http://www.sony.net/Products/SC-HP/cx_news/vol47/pdf/featuring47.pdf


From an avian photographer's point of view none of this technical detail is material however. A competent photographer can make excellent photographs with Canon cameras despite some technical short comings, as proven by the excellent images posted in this avian forum over and over again.

I do wish Canon had a high resolution FF camera like the D800, it would be a pleasure to use. Hopefully the next generation of Canon FF cameras in a couple of years will address these short comings.
The 5D3 prices have fallen to under $3K in less than 6 months which D800/D800E continue to remain out of stock, so Canon have learned a lesson here.

Jon Rista
11-02-2012, 01:50 PM
the CFA is made of polymer so it looks disordered. Typical CFA absorbs 1/2 of the light, it is a major factor in killing QE.

Yikes!




Unlike Canon, Sony EXMOR sensors do not need analog CDS unit as the reset noise is subtracted digitally for each pixel, this was the key innovation and the basis of their advantage. The details are explained nicely here http://www.sony.net/Products/SC-HP/cx_news/vol47/pdf/featuring47.pdf


Thanks for the article. I did not realize Sony's CDS was digital, although I see how they can take care of non-uniform response and eliminate the reset signal all digitally. Rather cool.




From an avian photographer's point of view none of this technical detail is material however. A competent photographer can make excellent photographs with Canon cameras despite some technical short comings, as proven by the excellent images posted in this avian forum over and over again.


I wouldn't call it immaterial. I think Randy's photo of the loon (http://www.birdphotographers.net/forums/showthread.php/103297-Higher-dynamic-ranges-impact-on-my-shooting-common-loon-in-bright-light?highlight=) demonstrates the value wider dynamic range can offer bird photographers. The deep blacks in that photo have amazing, fully usable detail. I don't think one could lift shadow detail in such a high contrast situation on a Canon camera without experiencing at least some banding, and probably a fair amount of pattern noise as well...which then necessitates stronger NR, thus reducing detail, etc. Just a couple of days ago I was photographing some American Coot's a little before and just after sunset. While I can make them barely acceptable for web size (900px long side usually), there is far too much noise in general, a lot of FPN, and a lot of that blotchy red color noise that really kills detail once you start lifting shadows.

I think better DR has implications even for our field, particularly with darker birds or birds with a lot of darker feathers (cormorants, grebes, loons, etc.), or photographing birds in less than ideal light. I am not really a fan of Nikon bodies (I much prefer Canon ergonomics), but two additional stops of dynamic range is a powerful lure. I too would love a high resolution camera from Canon, but I am not sure I'd really buy one if Canon doesn't step up their game technologically speaking (namely better DR and better shadow noise quality)...especially if they price a high MP body higher than the D800...that would just make the Nikon/Exmor lure all that much more powerful. Will Canon ever step up their game? When they do, will Sony have already pushed Exmor into 16-bit territory with another two stops of DR, higher frame rates with more megapixels than the competition, still offering better IQ, etc? The Exmor PDF you linked has me pretty amazed, that Sony was able to concurrently reduce the complexity of their sensor by adding CP-ADC and moving to digital noise removal processes while improving IQ as much as they have. The more information that comes out on that front, the harder it is for me to really seriously consider putting more money into Canon gear, despite the fact that I really love their lenses.

arash_hazeghi
11-02-2012, 03:43 PM
IMO all Canon pro cameras (1D/5D) can handle avian subjects with high dynamic range nicely, if exposed properly and processed efficiently. The consumer models do have some difficulty here.

here are some examples, these were challenging as there is feather texture/details in BOTH whites and blacks.

Jon Rista
11-02-2012, 08:26 PM
Well, I guess I'll have to give a 5D III a try. I can rent one for a while and see how it is before I make any decision. Those photos look great. I just noticed your professional guide to noise reduction link in your signature. I'll be buying that soon and having a gander at your wisdom, Arash. ;)

arash_hazeghi
11-02-2012, 09:50 PM
I am glad you liked the images and I am sure you will like the 5D3, I feel like now that the prices have dropped to about $2800 it is a competitive camera :)

Roger Clark
11-02-2012, 11:10 PM
Jon,
I agree with Arash, but note Arash's images were made in great light (looks like phase angles of 20 to 30 degrees and low sun) and many DSLRs can handle that dynamic range pretty well. At higher phase angles and harsher light the better cameras will pull detail out of the shadows better, but it would still be less ideal light.

The 5DIII still has fixed pattern noise similar to the 5DII (maybe just a little better). I'll have a sensor analysis complete in a few weeks. However, the AF speed of the 5DIII is significantly better than the 7D. If you want low fixed pattern noise in the canon series, from the data I have personally analyzed, the 1D Mark IV and 1DX come out on top. As you are interested in birds, also consider a used 1DIV. But if you have limited funds, I would put the money into lenses first and the 7D is a quite capable camera.

Roger

arash_hazeghi
11-03-2012, 11:56 AM
I agree a super-telephoto lens has a MUCH bigger impact on overall productivity and output quality than a camera body, plus it is a better investment too.
I also agree if your primary subject is birds a used 1D4 is a better choice than 5D3. They are getting harder and harder to come by as it is discontinued.

Gary Kinard
11-03-2012, 10:37 PM
5X to 10X behind Sony if Sony is current. That is wild. I didn't know they were that far behind.