Canon to start using 0.18um (180nm) process for FF?

[AdamJ]

Much as I'm hesitant to support (albeit partially) Mikael's contention about the Digic 5+ chip, chipworks believe it to be designed jointly by Canon and TI, and fabricated by UMC.

https://chipworks.secure.force.com/catalog/ProductDetails?sku=CAN-CH4-6501&viewState=DetailView&cartID=&g=&parentCategory=&navigationStr=CatalogSearchInc&searchText=digic%205+

But that doesn't matter. This debate is ultimately about the sensor, not the processor.

OK, I've got myself more popcorn so, please carry on.

Edit: Sorry, I meant ljunglinus, not Mikael.

 

[Canon-F1]

Much as I'm hesitant to support (albeit partially) Mikael's contention about the Digic 5+ chip, chipworks believe it to be designed jointly by Canon and TI, and fabricated by UMC.

believe.. i thought that belongs in church?

so it´s not common knowledge... it´s common believe?

 

[win nut]

The digic 5 and 5+ were designed jointly withTI and Canon. Prototyped and qualified by TI and then farmed out to UMC for mass production.

 

[Canon-F1]

Yes many belives a lot here and I keep to facts , digit is not a canon processor

so canon is lying and the digic(!) development team is just for show?

http://www.canon.com/technology/interview/digic4/digic4_p1.html

Am I right in saying that Canon has always had a policy of developing its image processors in-house, right from when the first digital cameras started being developed? Given that, today, most other camera makers customize processors provided by specialist LSI chip suppliers, what is Canon's stance regarding in-house development now?

Ikeda : I think it's more than that. There's also the pride that we feel at being the world's leading camera manufacturer. Canon has built up a great deal of knowledge with respect to image capture. However, with traditional cameras, camera makers didn't really need to have much to do with camera film or developing. With digital cameras, on the other hand, we are in a position to handle everything ourselves, right up to the "developing" stage. There is no way we are going to just abandon all of the possibilities that this opens up by entrusting it to someone outside the company

 

[marekjoz]

Yes many belives a lot here and I keep to facts , digit is not a canon processor

so canon is lying and the digic(!) development team is just for show?

You think, that inscription design on the chip comes for free?

 

[win nut]

I would not say that Canon is lying, they got a little help. No the Digic 5 is not a Divinci derivative, I can't say what it, but I can say what it not.

 

[Stu_bert]

The main problem as I see it is that Canon don't really have any real development momentum (or budget!) on higher resolution processes. They outsource almost everything except for the larger format sensors.

The biggest difference between the others and Canon is that all the other manufacturers are all dominated by their small-sensor image sensors sales, that already now are manufactured at 90 and 110-130nm metal processes on 300mm wafers. Panasonic and TSMC will start volume shipping of sensors made on 65/45nm rules in Q1 2013. Lower mask resolutions than 130nm are not enough to land you any sales any more. Most cellphone and compact camera sensors are manufactured at those levels now, and have been for the last few years. Also consider the fact that some of the others have very large yearly revenues from logic CMOS processes at 45, 32 and even 22nm levels. All of those markets are areas where Canon totally lack any type of experience. Canon outsource all more advanced fabs on their camera bill-of-materials.

The Digic sensors are made by UMC http://www.umc.com/english/class_300/index.asp, and were designed by Texas Instruments http://www.ti.com/lsds/ti/apps/videovision/end_equipment.page
The memory is most often made by Samsung http://www.samsung.com/global/business/semiconductor/product/consumer-dram/overview
The memory/Digic package-on-package mount has to be outsourced, since Canon cannot do it themselves, and it is believed that UMC does the mounting too.
Other peripheral control and logic chips are TI, Mitsubishi, AD and Fairchild.
.........

And is that different from Nikon how? Canon design their sensors and currently co-design the Digic 5, but to what extent they do it and what extent TI do it is not clear. What is clear is that TI don't sell that part or even mention it on their web site, so it is exclusive to Canon. Does not infer of course who own the most significant IP.

Going from a 500nm process to 180nm is like going from a 10MP FF camera to a 70MP camera in one generation. This means some really noticeable strains on the process, especially since the CMOS manufacturing process isn't as forgiving as just "taking pictures". You cant "scale to web size" and sharpen in post when you're making CIS wafers.

What you're basically asking from your equipment is to all of a sudden provide compact camera type linear resolution in a FF lens projection coverage - something almost unthinkable in the normal photographic world. The process has to be precise, to a degree where every single contrast and item on the new 70MP image is equal to or better than the 10MP camera - per pixel. In normal photographic resolution usage, we just want the final output to be good enough, which means that we downsample most images - we seldom deliver full-res images to the customers, and we seldom use full-res images in our own output.
Going from a 350nm mask to 250 and then 180nm and 12" wafers was a BIG step for most CMOS manufacturers, and most other manufacturers are a LOT bigger than Canon in this area.
............

So it's not that Canon COULDN'T do it. Even really small (in the imaging field) firms like STMicro can do it by stitching, and thereby tripling the unit prices. For Leica this isn't really a problem since the total BoM on a M series camera is most certainly lower than 2k USD. This gives a healthy margin up to the projected 7k USD end price point.
This isn't what Canon does. They live on volume, not on extreme margins.

Hold on, your making multiple assumptions here

1) Canon needs to make the leap, in volume, on FF at once otherwise they'll what exactly... go bankrupt?
2) Canon needs to sink billions (immediately) in order to compete with Sony's sensor manufacturing capability or indeed other (sensor) manufacturers to stay competitive on their own products
3) A 180nm sensor is the only way Canon can provide competitive products to Nikon / Sony
4) Canon cannot survive if it manufacturers its' own sensors

oh and presumably, Canon has been ignoring all this, has changed culture and wants to do it all together?

They presumably know what investment is possible, and what is required for their next iteration of sensors. I'm also assuming that this information is not something recent, and that they have been developing capability. Not to the level to suddenly compete with Sony, but to a level that would be sufficient to remain profitable. Whether the next FF or APS-C will be 180um and whether they will design or design / manufacturer frankly is high conjecture. Anyone who wants to make a decision on their future camera purchases based on this conjecture is someone with more disposable income than I!

In another thread, it was mentioned that the latest Sxx sensor was 180um. And clearly that's an ocean away from FF manufacturing @ 180um, but maybe that is the first "volume" process for Canon and that elsewhere they are refining a low-yield FF capability. I'm happy to be wrong, but Sony & Nikon have not been manufacturing FF sensor for more than 5 years (less?), and that's not exactly a high volume capability either. Agreed it is more than Canon have been doing, and that may explain why Canon has taken so long to switch (problems) or just the bean counters (sweat assets / more risk adversed in a shrinking market)

But Sony doesn't have too many other lines of business making money for them. Finding their position in sensor manufacturing has been a great success. Canon is not looking to manufacturer sensor for smartphones and other areas, either because they don't have the capability or don't want to be there.

And presumably, that investment Sony has made is based on the diversity of sensors they produce and indeed the volume. Again, Canon is not in that market, and surely that means the investment - should they decide it will be viable over the next 10 years, is considerably less than Sony needs to.

Finally, are we sure a 320um or 250um with the other sensor tech that has been outlined would not provide a significant step-change for Canon sensors, and since it's been done to death that a photographer / photography is more than his sensor, then if Canon's next APS-C or FF sensor is not 180um does that definitively mean it won't be much different from the current sensors???

Think I'll go back to err, taking pictures on my dinosauric kit...

 

[Stu_bert]

I would not say that Canon is lying, they got a little help. No the Digic 5 is not a Divinci derivative, I can't say what it, but I can say what it not.

The Digic V is the first joint-design is it not? But not one which TI mention one iota on their site. Not even a press-release, "we helped design Canon's next gen Digic processor in their dSLR"?

TI definitely provided some know-how, but perhaps more around the "manufacturing design" aspects and not in the critical imaging side based on what Canon has developed before. Sorry, don't consider the TI angle to be anything significant....

 

[jrista]

Well by some of us it is knowledge , digit Is davinchi processor from TI

You need to back that claim up with some actual facts, man. I've been searching the web for an hour now, and I have yet to find any actual evidence of that. The ONLY direct link between DIGIC 5 and Texas Instruments DaVinci specifically is a hypertext link to the Texas Instruments DaVinci wiki page from the DIGIC wiki page's References. The only references to ARM on that wiki page are DIGIC II and III, so it stands to reason that if DaVinci ARM-only processors are related, that they would be related to those two versions of DIGIC. The only other source of information is someone on DPReview claiming that the ARM subprocessor embedded within DIGIC 5 "was probably a Texas Instruments DaVinci series of application processors". According to Tech-On's dissection of a DIGIC 4, it does appear to be a package-on-package, however it appears there is only a DIGIC 4 and the Samsung memory...if there is an ARM processor embedded in the DIGIC 4 die, they don't say.

As for ARM, that is simply the general purpose sub-processor contained within some DIGIC packages, however there is also a DSP for high speed image processing. TI may have made or designed the ARM processor (I still have yet to find any actual evidence of that, but it stands to reason Canon would outsource such design to a third party that has more experience with GENERAL logic processing units), but historically Canon designs and manufactures their own DSP/image processors according to all of the actual evidence I've been able to find thus far. The ARM processor is also only a confirmed part of DIGIC II and III, I have yet to find any direct evidence that DIGIC 4 or 5 also contain an ARM processor or if so whether it was designed by TI.

Much as I'm hesitant to support (albeit partially) Mikael's contention about the Digic 5+ chip, chipworks believe it to be designed jointly by Canon and TI, and fabricated by UMC.

believe.. i thought that belongs in church?

so it´s not common knowledge... it´s common believe?

Yes many belives a lot here and I keep to facts , digit is not a canon processor

Here's the problem...what you just posted there, and all of your previous posts? Those are not actually facts...they are what we call an anecdote. You are making anecdotal proclamations in the total absence of fact. If you wish to make a claim, and intend for that claim to be accepted "as fact", you need to back that claim up with some external references that allow others to VERIFY your claims as fact. Simply stating that what you say is plain and simply fact does not actually make it a fact.

Based on the information I've been able to find, information which I have also linked within my posts, The Suede is at the very least being very misleading about DIGIC being designed by a third party. He makes it sound as though Canon had no hand in the design, and that it was entirely designed and manufactured by third parties. I've found a single reference from Chipworks that the DIGIC 5 (and quite specifically JUST the DIGIC 5, not the DIGIC 5+ or other variants) was jointly designed by Canon and TI. I can't figure out exactly why they make that claim, and if the DIGIC 5 includes an ARM processor like the DIGIC II and III, then it is entirely possible that TI is only responsible for the ARM processor embedded within the rest of DIGIC 5's logic gates. There does not seem to be any information at all that indicates DIGIC 5 of any variant is specifically a DaVinci DSP either. As far as I can tell, the reason Wikipedia linked the TI DaVinci page was the same reason they linked the Nikon EXPEED page...DaVinci, EXPEED, and DIGIC are all digital camera DSP's, some of which have a general purpose ARM processor embedded.

I have yet to find any concrete evidence that it was manufactured by a third party, however I wouldn't state it as plainly impossible at this point (Chipworks seems to think there are some design markers that indicate UMC, but that is far from definitive enough to claim as undeniable fact...UMC devices usually have their logo, "UMC" encapsulated in an angular oval or the little wireframe globe symbol). All of the actual evidence indicates that Canon does indeed design and manufacture their own image processors in general. There are also only two verified instances (that I have yet to find) in the past where Canon has included an ARM processor embedded in the same die or package as the DIGIC processor...in the DIGIC II and III. Here's that same link, in case you want to verify.

 

[jrista]

Edit: Sorry, I meant ljunglinus, not Mikael.

Are we sure they are not the same person? Mikael and possibly ankorwat have been banned...and ljunglinus only signed up on the 28th. Given how much ljunglinus sounds like Mikael and ankorwat, I think the real question at hand is...is this Mikael Incarnation #3?!?

Much as I'm hesitant to support (albeit partially) Mikael's contention about the Digic 5+ chip, chipworks believe it to be designed jointly by Canon and TI, and fabricated by UMC.

https://chipworks.secure.force.com/catalog/ProductDetails?sku=CAN-CH4-6501&viewState=DetailView&cartID=&g=&parentCategory=&navigationStr=CatalogSearchInc&searchText=digic%205+

But that doesn't matter. This debate is ultimately about the sensor, not the processor.

OK, I've got myself more popcorn so, please carry on.

Well, because of personalities like Mikael and TheSuede, the question they are posing is "Can Canon keep doing what they do?", with the implication that "Canon is utterly doomed and completely incapable of competing at all, on any level, for any reason, and they can do nothing to change that 'FACT' as we proclaim it." I dispute that "fact". I don't think Canon is implicitly incapable of competing. I think Canon's bean counters have a stronger hold over the company than say Nikon or Sony, both of which are in deepening financial doo-doo. It may be Canon's saving grace that they have employees to ensure they remain a profitable company despite tough and toughening economic conditions. Is Canon simply incapable of competing? Or is Canon trying to weather a sour market full of penny-pinching consumers such that they actually survive at all and come out the other side?

Evidentially, it does not appear that Canon is incapable of manufacturing CMOS devices down to as small as 65nm, there seems to be evidence they have been prototyping advanced 180nm parts that use advanced technologies for at least a while (couple years...since the 120mp APS-H?), and have 46.1mp DSLR prototypes out in the field. Cameras that supposedly sport a sensor that is sufficiently advanced from their current line of 500nm sensors that would (according to the specs...high MP, active cooling and "lowest read noise on the market", 16-bit ADC) put them at the front of the back, even against the likes of Sony. Now, whether that is indeed the case or not is a case where only time will tell, but there is no question that there is mounting evidence Canon is still competitive and NOT the end-of-life dud that some among us seem to want us to believe.

 

[x-vision]

And Sony is only the third largest manufacturer (at 18% total market revenue), Both Samsung and Omnivision are larger than Sony.

Hmm, you are quite incorrect, actually.

Sony is the industry leader with $2.6B in revenue for 2011.
Somewhat surprisingly, second is ... Canon, with $874M in revenue for 2011 (3x smaller than Sony).

Source: Image Sensors World.

Also, as I said already, Canon has had a 180nm process in house since at least 2007.
See the White Paper on their 52mp prototype, which they developed in 2007.
Scroll to page 3 of the White Paper, where the sensor technology is listed as 180nm.

Finally, as I've also said already, the S100/S100 sensor is made on 180nm.
This sensor has been in production for a year now.

Overall, Canon has been working on a 180nm process for at least five years now and has had in production for at least a year.

We can only speculate why this process has not been used for DSLR sensors ... yet.
Probably they are working on the yields - or they are just milking their 500nm process as much as possible.

I feel that the time for change has come, though.
Canon has not had a new APS-C sensor for three years now (since the current 18mp sensor was introduced back in 2009).
For comparison, between 2004 and 2009, they've had a new APS-C sensor basically every year.

This 3-year 'silence' indicates to me that a major change is coming.
Likely the transition from 500nm to 180nm; we'll know for sure next year.

Something else to consider:
Canon entered the high-end video market just last year with the Cxxx series.
Sony, RED, and Arri are the incumbents in this market.

If Canon thinks that they can compete with a 500nm process in the brewing 4K/8K video resolution wars, they are complete idiots.
Or, more likely, they know very well what's coming - and feel that they can compete.
Certainly not with a prehistoric 500nm process, of course.

Check out the lasted video cams from Sony, btw. This is getting quite serious.

 

[win nut]

The divinici line is mainly for automotive, nav systems, dvd etc....

 

[jrista]

And Sony is only the third largest manufacturer (at 18% total market revenue), Both Samsung and Omnivision are larger than Sony.

Hmm, you are quite incorrect, actually.

Sony is the industry leader with $2.6B in revenue for 2011.
Somewhat surprisingly, second is ... Canon, with $874M in revenue for 2011 (3x smaller than Sony).

Source: Image Sensors World.

Also, as I said already, Canon has had a 180nm process in house since at least 2007.
See the White Paper on their 52mp prototype, which they developed in 2007.
Scroll to page 3 of the White Paper, where the sensor technology is listed as 180nm.

Finally, as I've also said already, the S100/S100 sensor is made on 180nm.
This sensor has been in production for a year now.

Overall, Canon has been working on a 180nm process for at least five years now and has had in production for at least a year.

We can only speculate why this process has not been used for DSLR sensors ... yet.
Probably they are working on the yields - or they are just milking their 500nm process as much as possible.

I feel that the time for change has come, though.
Canon has not had a new APS-C sensor for three years now (since the current 18mp sensor was introduced back in 2009).
For comparison, between 2004 and 2009, they've had a new APS-C sensor basically every year.

This 3-year 'silence' indicates to me that a major change is coming.
Likely the transition from 500nm to 180nm; we'll know for sure next year.

Something else to consider:
Canon entered the high-end video market just last year with the Cxxx series.
Sony, RED, and Arri are the incumbents in this market.

If Canon thinks that they can compete with a 500nm process in the brewing 4K/8K video resolution wars, they are complete idiots.
Or, more likely, they know very well what's coming - and feel that they can compete.
Certainly not with a prehistoric 500nm process, of course.

Check out the lasted video cams from Sony, btw. This is getting quite serious.

Thanks a lot for the references. The 52mp prototype paper is an interesting read. Any chance you have a link to a paper like that for the 120mp APS-H? If they were already using 180nm process for the 52mp, I wonder if they were experimenting with an even smaller process (90nm?) for the 120mp version. I think Canon has more up their sleeve than just a 180nm process change...the 120mp sensor had a 9.5fps readout rate, and some kind of on-die processing (CP-ADC, or some kind of on-die parallel ADC). That speaks well to high framerate video on a 180nm process, although who knows yet if it will be able to compete head-to-head with the incumbents.

As for Sony, it sounds like their investment in CP-ADC and high-speed readout is really starting to pay off. A 4k camera with global shutter and 180fps is a pretty astounding feat!!

 

[jrista]

You are bending words, digit 5 is not a canon processor , se chipworks
Im waiting for a answer from my source regarding davinchi
But like Fuji claims it is a fuji sensor inside X1 (sony with different color filter) canon present a TI as theirs device
And the sensor you are describing has the traditionell read out, no column adc etc
We all know that canon can make a sensor with small cells, but nothing like sony or panasonic as two examples and columnwise adc

Secret sources make not a verifiable fact...

Evidence, Man! Links! Pages! Text! Images! EVIDENCE, MAN!! EVIDENCE!!! That we can all view, read, and evaluate ourselves....

 

[win nut]

jrista, pm in your inbox.

 

[AdamJ]

Edit: Sorry, I meant ljunglinus, not Mikael.

Are we sure they are not the same person? Mikael and possibly ankorwat have been banned...and ljunglinus only signed up on the 28th. Given how much ljunglinus sounds like Mikael and ankorwat, I think the real question at hand is...is this Mikael Incarnation #3?!?

Mikael is banned? I didn't know that. I was going to challenge him about the posterisation I'm seeing in Nikon cameras. It can be seen in these comparison photos of an elderly gentleman, if you look closely.

 

[jrista]

Edit: Sorry, I meant ljunglinus, not Mikael.

Are we sure they are not the same person? Mikael and possibly ankorwat have been banned...and ljunglinus only signed up on the 28th. Given how much ljunglinus sounds like Mikael and ankorwat, I think the real question at hand is...is this Mikael Incarnation #3?!?

Mikael is banned? I didn't know that. I was going to challenge him about the posterisation I'm seeing in Nikon cameras. It can be seen in these comparison photos of an elderly gentleman, if you look closely.

Man, you know...I have to scrutinize, but yea...I see a LOT more posterization in the D800. It IS in color...basic color...but color nonetheless. I wonder when Canon will finally enter the modern age and make color sensors like Nikon and Sony....I just hope they skip the posterization.

 

[AdamJ]

The Digic sensors are made by UMC http://www.umc.com/english/class_300/index.asp, and were designed by Texas Instruments http://www.ti.com/lsds/ti/apps/videovision/end_equipment.page
The Suede

Finally, I'd like to know where you get your information. How do you know TI designed the DIGIC sensor, or that UMC manufactured it? According to Canon, Canon themselves designed the DIGIC sensor, and as it stands
From Jrista

You twist and turn things Jrista , you have got the answer many times through Chipworks

Good morning, Mikael. I'll just get some coffee brewing and then make myself comfortable for Round 2. ;D

 

[jrista]

The Digic sensors are made by UMC http://www.umc.com/english/class_300/index.asp, and were designed by Texas Instruments http://www.ti.com/lsds/ti/apps/videovision/end_equipment.page
The Suede

Finally, I'd like to know where you get your information. How do you know TI designed the DIGIC sensor, or that UMC manufactured it? According to Canon, Canon themselves designed the DIGIC sensor, and as it stands
From Jrista

You twist and turn things Jrista , you have got the answer many times through Chipworks

Ah, there is that signature bold and the shoddy quoting of Mikael. The "twist and turn" comment also gives you away. You just keep on creating new accounts as each one is banned, don't you? You have a very distinct online "signature presence", so you might as well just call your next account IncarnationOfMikaelNo4. ;P



Chipworks merely SUSPECTED that the device may have been fabricated by UMC, due to some die markers that might indicate it's them. They did not include any visual evidence of those die markers. The only two markers that would definitively indicate the chip was indeed fabricated by UMC would be its signature logo, a little wireframe globe, or their name "UMC" set inside an angular oval. If the chip was manufactured by a third-party, there are a number of better fabs that Canon could have chosen.

I did find an image that had TI's logo in one of the corners of the die, however the die packaging is clearly Canon packaging, with Canon's signature etching, Canon's signature part numbers, etc. Ironically, the plastic package over the chip is labeled with Elpida, a (once major) memory manufacturer that went bankrupt. Perhaps that indicates the DIGIC 5 is a PoP that combines the DSP with Elpida memory (or would that be Micron memory now)? I still don't see any clear, factual evidence that UMC is the fabricator...just a SINGLE statement from Chipworks that they think it might be. The DIGIC 4 was a 65nm chip, which would have been at the limits of Canon's in-house lithography capabilities. If the DIGIC 5 uses a smaller process than 65nm, then it is logical that they would outsource to a third-party fab like UMC, although TSMC seems to be a more reputable option. I'd still like concrete information about DIGIC's process and fabrication, though. Everyone claiming that there is no question it "is" UMC is just presumption, and that doesn't benefit anyone. And if it is still a 65nm part, then I see no reason it couldn't be manufactured in-house.

I did receive a tip today from someone who stated they work for TI (for over 12 years). I was asked to keep the name confidential and as such this is by no means verifiable fact, so, take this as you will. They indicated that TI was solely responsible for designing a few optimizations to make the DIGIC 5 easier/more efficient to manufacture, but that the fundamental and functional design of the chip was entirely Canon. I am willing to accept that Canon my have collaborated with TI in order to gain their expertise in designing a complex chip on a small die that could be efficiently laid out such as to maximize yield and improve manufacturability, especially if the process was smaller than Canon could fabricate in-house. They also stated quite matter-of-factly that DIGIC was most certainly not DaVinci...DaVinci is more of a general purpose DSP+CPU intended to support a very broad and diverse field where DSPs can be used, where as DIGIC is a family of highly specialized DSPs specifically designed (by Canon) to function optimally for the kind of image processing Canon cameras do.

Cmos
http://www.i-micronews.com/upload/Rapports/Yole_CMOS_Image_Sensors_October_2012_Report_Sample.pdf

There is a surprising amount of redaction in that document...suspicious...! Why can't you just provide complete, unadulterated source information to back up your claims? Why is there ALWAYS something suspicious about your links, assuming you provide one at all?

 

[PeterJ]

This thread made me wonder about something... red has a wavelength of around 750nm and after some Googling I see a 7D is has a pixel density of 4300nm. To avoid aliasing I'm guessing below 1500nm pixel pitch would be useless, does the 7D sensor use a 500nm process or smaller? Just thinking if 500nm wouldn't a move to 180nm just about be at about the edge of what's actually useful for maximum pixel density?

Anyway would be interested to hear from anyone that knows more about sensor fab. I'm not sure about how the manufacturing process scales down especially with photodiodes in the mix and of course the main advantage might be more sophisticated amplifers and gain control near the photosites.