Sensor Fabrication

[jaayres20]

Is it possible for someone to explain to me, in somewhat simple language, what exactly Canon has been doing in the past vs what Sony is doing now in terms of sensor fabrication? What in terms of the physical makeup of the sensor does Sony have going for it, and what was, or is preventing Canon from doing the same thing? And what are the hopes in technological advancements that we want to see in these new 50mp ff Canon sensors? That is if they are from Canon. I know everyone wants more DR and better low ISO IQ, but what physically needs to happen to the sensor for Canon to once and for all shut the mouths of the Nikon/Sony/I want more DR so I can push my files 5+ stops in post crowd?

 

[Mt Spokane Photography]

Patents and Ego's.

1. EGO - Canon believes their sensors are fine, and sales show that.

2. - Patents of the method Sony uses to place the A-D circuitry on the sensor is probably a minor issue, since a different approach might yield the same results. Canon might even use a Sony sensor on the latest DSLR's, if its cheaper to buy than develop a new one. Canon is out to make a profit, and will go wherever they can do that.

While there are a lot of those who like the Sony sensors, any benefits only show up for a very few users, since you must manipulate a raw image in post processing to get the DR advantage. Out of camera jpegs are the same, and most camera users would not fool with raw, or even know what it is.

CR has mostly camera enthusiasts posting here, and probably only 100-200 frequent posters out of which about 10 post a lot about DR. At the bank, where it counts, DR has yet to become a issue, since a overall system is more critical for most.

Everyone has more features they want, but we won't get them all.

 

[Mitch.Conner]

Patents and Ego's.

1. EGO - Canon believes their sensors are fine, and sales show that.

2. - Patents of the method Sony uses to place the A-D circuitry on the sensor is probably a minor issue, since a different approach might yield the same results. Canon might even use a Sony sensor on the latest DSLR's, if its cheaper to buy than develop a new one. Canon is out to make a profit, and will go wherever they can do that.

While there are a lot of those who like the Sony sensors, any benefits only show up for a very few users, since you must manipulate a raw image in post processing to get the DR advantage. Out of camera jpegs are the same, and most camera users would not fool with raw, or even know what it is.

CR has mostly camera enthusiasts posting here, and probably only 100-200 frequent posters out of which about 10 post a lot about DR. At the bank, where it counts, DR has yet to become a issue, since a overall system is more critical for most.

Everyone has more features they want, but we won't get them all.

There may also be trade secrets as well - stuff that Sony doesn't intend to make known via a patent (which also starts the clock ticking down for exclusivity). This would probably be more in the fabrication methods rather than the design as they can limit the number of people who know certain secret aspects of the fabrication process, but they can't protect against reverse engineering.

I do have to say though that I personally am curious as to how exactly Sony achieves the additional available DR. I would have thought that the 5D3 would have more available DR than higher MP sensors due to the larger pixels. I don't quite understand how Sony achieved the extra stops despite the smaller pixels. Is it somehow simulated?

 

[PureClassA]

Since the AD is ON the sensor, there is zero signal path inbetween, therefore no additional noise, therefore no loss of 2-3 stops. As I understand it

 

[jeffa4444]

Where do you begin! Sony is operating three fab plants in Japan all of which have had massive new investment. Sony R&D into CMOS sensors dwarfs Canon as Sony supply 60% of the global CMOS sensor market. The big growth has been cell phones but automotive is catching up fast and industrial is huge.
Sony is developing a curved sensor, layered sensors, front & back-side illuminated sensors, organic sensors the list goes on, by contract Canon main focus has been cameras but they too see growth in security & industrial to diversify risk and create opportunity.
Panasonic is developing sensors now with Fuji (organic) and sold off its controlling interest in its fab plants to TowerJazz an Israeli company (they fab Leica sensors in the M240 designed by CMOSIS).
Its an extremely expensive business and relies on big volumes to make any money and the volumes in photography have been falling.

 

[Coldhands]

As Spokane and PureClass have mentioned, the Sony sensor uses on-chip analogue-to-digital conversion, which results in very low read-noise. This enables a significant shadow push during raw conversion before the noise overwhelms the detail, which is what sets the DR limit at the low end.

From what I have read, Canon has been using a 500nm CMOS fabrication process for pretty much every DSLR sensor for the past ten years. (For reference Sony Exmor uses 180nm) I suspect this has been a significant barrier to development of new CMOS architecture and, as chipworks has suggested, limits Canon's full-frame sensors to MP numbers in the low twenties. Therefore it would follow that in order to increase the resolution beyond this, they must move to a new process.

Since the rumoured 5Ds' 50 MP resolution is pretty much a dead certainty, we must assume that Canon has either started using a new fab process or is buying the sensor from another company. Interestingly, there was speculation on this forum a little while ago that Canon had used a smaller process for the 7DII's sensor as it appeared to display a higher QE (quantum efficiency - the ratio of captured electrons to incoming photons) than the 70D's sensor. Not sure if there was ever an authoritative answer on this.

 

[Marsu42]

EGO - Canon believes their sensors are fine, and sales show that.

Interesting enough, a top Canon exec recently mentioned that they'll use the "best available" sensors or something like that - so they're even stating the obvious difference to Sonikon in public.

The ego probably is more on the enhusiasts' side that won't have any non-Canon nonsense near their cameras, while Canon themselves are just out for profit and would probably accept a dent in their brand marketing by licensing/buying Sonikon tech.

Since the rumoured 5Ds' 50 MP resolution is pretty much a dead certainty, we must assume that Canon has either started using a new fab process or is buying the sensor from another company.

Not necessarily, they manage to achieve higher pixel density on crop sensors for a long time and the new 50mp version could be just an "upscaled" crop. The reason that is rumored to have prevented them from doing so before is that the full frame sensor yield is low, i.e. upscaling the current tech would result in too expensive cameras.

 

[Machaon]

Since the rumoured 5Ds' 50 MP resolution is pretty much a dead certainty, we must assume that Canon has either started using a new fab process or is buying the sensor from another company. Interestingly, there was speculation on this forum a little while ago that Canon had used a smaller process for the 7DII's sensor as it appeared to display a higher QE (quantum efficiency - the ratio of captured electrons to incoming photons) than the 70D's sensor. Not sure if there was ever an authoritative answer on this.

I would hazard a guess that the 7DII and 5Ds sensors are made at the same fab, whosever that might be. The pixel pitches are much the same, so any fab that can manufacture a 7DII sensor will be able to make the 5Ds sensor.

The main difference would be that the 5Ds sensors would offer lower yields per wafer, making them more expensive and less economical.

My understanding is that Canon is the manufacturer of the 7DII sensor, from which would follow they also make - or at least have the capability to make - the 5Ds'.

 

[rrcphoto]

Since the AD is ON the sensor, there is zero signal path inbetween, therefore no additional noise, therefore no loss of 2-3 stops. As I understand it

yes / no. sony has much different correlated double sampling than canon does. canon doesn't double sample away from the pixel.

 

[Coldhands]

...

Since the rumoured 5Ds' 50 MP resolution is pretty much a dead certainty, we must assume that Canon has either started using a new fab process or is buying the sensor from another company.

Not necessarily, they manage to achieve higher pixel density on crop sensors for a long time and the new 50mp version could be just an "upscaled" crop. The reason that is rumored to have prevented them from doing so before is that the full frame sensor yield is low, i.e. upscaling the current tech would result in too expensive cameras.

I'm basing my assumption off a chipworks article on the 1DX sensor that states :

Given the geometric constraints of 0.5 µm design rules, Canon seems content to hang around the 21 Mp resolution for recent FF sensors through the use of shared pixels. Jumping to a higher resolution generally requires more advanced design rules and pixel sharing architecture.

Link: http://www.chipworks.com/en/technical-competitive-analysis/resources/blog/full-frame-dslr-cameras-canon-stays-the-course/?lang=en&Itemid=815

Granted, that article is from 2012 so things might have changed, and unfortunately the author doesn't quantify their statement, so I suppose it could be that some other development has allowed them to break the low-twenties MP barrier.

Since the rumoured 5Ds' 50 MP resolution is pretty much a dead certainty, we must assume that Canon has either started using a new fab process or is buying the sensor from another company. Interestingly, there was speculation on this forum a little while ago that Canon had used a smaller process for the 7DII's sensor as it appeared to display a higher QE (quantum efficiency - the ratio of captured electrons to incoming photons) than the 70D's sensor. Not sure if there was ever an authoritative answer on this.

I would hazard a guess that the 7DII and 5Ds sensors are made at the same fab, whosever that might be. The pixel pitches are much the same, so any fab that can manufacture a 7DII sensor will be able to make the 5Ds sensor.

The main difference would be that the 5Ds sensors would offer lower yields per wafer, making them more expensive and less economical.

My understanding is that Canon is the manufacturer of the 7DII sensor, from which would follow they also make - or at least have the capability to make - the 5Ds'.

That's essentially what I was getting at. I consider it conceivable that Canon tested a new fab process on an existing sensor architecture (70D -> 7D2) i.e. a die-shrink as in Intel's "Tick", to be followed up by using that fab on a new sensor design later (the corresponding "Tock"). Tick/Tock

Or maybe it's just a Sony sensor and everything I've just said is total rubbish.

 

[NancyP]

Tick-tock is how I would think a camera sensor fab plan ought to work. After all, it has worked for Intel. Current tick is a little behind schedule, though, so even Intel runs into some problems. Tock may be behind as well. I have been wondering if I should buy the current Haswell MacBookPro rather than hang on for the late 2015/early 2016 Skylake MBP. (decision is based on willingness of current 2010 MBP to hang on another few months).

 

[rrcphoto]

Tick-tock is how I would think a camera sensor fab plan ought to work. After all, it has worked for Intel. Current tick is a little behind schedule, though, so even Intel runs into some problems. Tock may be behind as well. I have been wondering if I should buy the current Haswell MacBookPro rather than hang on for the late 2015/early 2016 Skylake MBP. (decision is based on willingness of current 2010 MBP to hang on another few months).

it's more tick .. wait a minute .. tock. no sensor for DSLR's uses design rules from the last decade.

 

[x-vision]

What in terms of the physical makeup of the sensor does Sony have going for it, and what was, or is preventing Canon from doing the same thing?

Others have provided good answers but here is my answer as well:

Basically, Canon is using an outdated manufacturing technology and an outdated sensor design.

Sony achieves high DR (dynamic range) by using on-sensor ADCs (analog-to-digital converters).
Canon is still using external ADCs - and the primary reason for that appears to be that Canon is still
using an outdated CMOS manufacturing process vs the competition.
Other than that, Canon holds multiple patents for on-sensor ADCs, so they could in theory manufacture
high DR sensors.

I'm personally hoping that the 5Ds sensor is a new generation sensor and will have good DR.
It would be very disappointing otherwise.

 

[K-amps]

I doubt the article that states that Canon cannot go much higher over 21mp because of the 500nm fab, I seem to recall they did produce the aps-H sensor of 120mp around 2012 with that fab. And scaled up to FF... it will be higher. The reason for staying close to 21 mp are many some of which are, file sizes, high ISO noise, market need, evolutionary designs, data buses etc.

 

[Coldhands]

I doubt the article that states that Canon cannot go much higher over 21mp because of the 500nm fab, I seem to recall they did produce the aps-H sensor of 120mp around 2012 with that fab. And scaled up to FF... it will be higher. The reason for staying close to 21 mp are many some of which are, file sizes, high ISO noise, market need, evolutionary designs, data buses etc.

I have been unable to find any information relating to the process used to produce that sensor. Perhaps you can provide a source confirming it used the 500nm fab.

Chipworks specialize in reverse-engineering and analysis of semiconductor designs and charge thousands of dollars for their reports. So until I see compelling evidence to indicate otherwise, I am going to consider their information more valid than some speculation about a one-off tech demo.

 

[jeffa4444]

http://www.eetimes.com/document.asp?doc_id=1278518

Read this article it shows how ADCs will become integrated into FPGA designs.

We have been working with FPGAs for a low volume application using the LVDS pin pairs for ADC function in an effort to reduce power consumption and mitigate heat output overall. FPGAs are at a point where low volume or high volume are both practical.

 

[PureClassA]

Sony HAS the capacity to easily pump out 5DS sensors beacuse of the volume they produce, whereas it's been said, Canon perhaps would not be as cost effective because their fab process is not as efficient or would be too expensive to retool in time for so comparably limited an audience as a 5D series would have.

This all goes back to the "Who made the sensor?" debate. It's been pointed out that Sony's "Column Parallel AD Conversion" technology has been in play since 2007 and it has matured into the high DR sensors we see today in their line and Nikon's. It's also been pointed out that Canon itself has such patents with similar "on the die AD processors" If it's not financially viable for Canon to build these themselves, then good business logic would dictate you investigate handing Sony your design if the technology is sound and the costs of having them manufacture these devices is measurably lower, AND you're not on the hook for all the capital investment it would otherwise take in your own factories if you cant sell enough to recoup your costs...

Now enter the overall slumping market for DSLRs in general. iPhones and such have killed demand for the casual user, but not for the pros. But the pros are the ones who demand the products that get largely financially supported by the former buying habits of the casual user...the mass market rebel users for example... a mass market that has shrunk a good bit in recent years.

We have entered an era where Canon and Sony and Nikon and Pentax and Minolta Samsung and so on and so forth just can't ALL remain viable dumping so much capital into their own fabrication lines. The big players in Chip building like Sony and Samsung have the capacity and volume to undercut most everyone.

If I am Canon or anyone else, and I plan to stay A) Financially healthy and B) Competitive in the marketplace, then I have to be realistic and look at someone else to build my designs. I have little doubt that this new sensor wont at least be a Canon design. But it won't surprise me one bit if it was built at Sony. It only makes all the financial sense in the world IF IF IF they want to compete. Canon obvious has the expertise and technology to DESIGN a sensor that can equal or beat a Sony... but do they have the ABILITY to build it themselves?

That's the question. Right now, given the info we have, I'd say likely not. So since Canon has on Sensor AD designs themselves, and Sony already has the capacity to build them, why wouldn't they farm it out? I would!!

 

[PureClassA]

Of course they have the knowledge. It's readily apparent by the mere fact of all the amazing patents they hold. Although we can only speculate, the one thing we DO know is that their current fabrication process is yet incapable, so far as we know, of keeping up with their engineers. And with a shrinking DSLR marketplace, it will become increasingly financially difficult for them to do so with sensors. Hence the move most other companies are making to buy them from HUGE outfits like Sony. Canon can keep it one step closer to home because they still have the technical ability to design them first.

 

[PureClassA]

It's been discussed on this forum many times. Even by far wiser folks than I like jrista, who often laments the fact that Canon is such an amazing innovator that isn't employing near what they have designed. Like this on sensor ADC for but one small thing:

http://image-sensors-world.blogspot.com/2013/08/canon-files-for-dual-range-column.html

 

[PureClassA]

Or how about this?
http://petapixel.com/2014/06/27/canon-patents-multi-layer-image-sensor-five-layers-instead-three/