Chipworks low level sensor reviews: Canon vs Nikon vs Sony

[NormanBates]

I haven't still read this in detail, but it looks incredibly interesting:
http://www.chipworks.com/blog/technologyblog/2012/10/23/full-frame-dslr-cameras-part-1-nikon-vs-sony/
http://www.chipworks.com/blog/technologyblog/2012/10/24/full-frame-dslr-cameras-canon-stays-the-course/

A few things I've learnt here already:
* Nikon still designs some of its own sensor. The one in the D800 is a Sony model, but the on on the D4 is a Nikon design manufactured by Renesas.
* Sony uses the most advanced manufacturing process by far: Sony=180nm, Renesas = 250nm, Canon=500nm
(I think it was in 2000 that a 500nm process was state-of-the-art for manufacturing CMOS image sensors)

[NormanBates]

RE: D800: "The shared pixel layout and narrow metal lines afforded by a 0.18 µm process are critical to maintaining the fill factor needed to achieve a high performance 36 Mp FF pixel array"

RE: Canon: "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"

It seems like for Canon to make a high-Mpix camera, they'd need both a redesign in their pixel layout (in the 1DX they use "stand-alone (unshared) pixel architecture") and more modern manufacturing capabilities (and making a few prototypes -probably on an extenal fab- does not mean they can get them cheaply to mass production)

Then again, shared pixel architectures are not new for them at all (the 5D2 and 5D3 both use shared-pixel architectures) and "Canon does have a 0.18 µm generation CIS wafer fab process, featuring a specialized Cu back end of line (BEOL) including light pipes" (but it seems to be using that only for compacts)

[traveller]

Sorry,

I think that I may have posted on the same topic before I realised that you had already! 

It's good to actually read some analysis based upon research and facts in all the rumour, hype and speculation.  For me, the Chipworks articles show that Canon has been eeking out the last performance from an older generation (or two!!) of technology.  The important thing is that Canon has reached the point where they need to change their fab process and modify their sensor designs to support the new technology. 

It seems to me that Canon aren't in a great position with regards to sensors.  They have been buying in Sony designs for their compact cameras and so lack the design experience as pixel size shrinks, as well as the economies of scale for their R&D departments.  Canon also maintain their own capital intensive fabs solely for larger sensors (APS-C & full frame), which could be what is tying them into older fab technology (i.e. lower production volumes means it takes them longer to amortise the fab assets). 

[gmrza]

Sorry,

I think that I may have posted on the same topic before I realised that you had already! 

It's good to actually read some analysis based upon research and facts in all the rumour, hype and speculation.  For me, the Chipworks articles show that Canon has been eeking out the last performance from an older generation (or two!!) of technology.  The important thing is that Canon has reached the point where they need to change their fab process and modify their sensor designs to support the new technology. 

It seems to me that Canon aren't in a great position with regards to sensors.  They have been buying in Sony designs for their compact cameras and so lack the design experience as pixel size shrinks, as well as the economies of scale for their R&D departments.  Canon also maintain their own capital intensive fabs solely for larger sensors (APS-C & full frame), which could be what is tying them into older fab technology (i.e. lower production volumes means it takes them longer to amortise the fab assets).

While I think it is almost certain that Canon will need to move to (at least) a 0.18µm process in its next generation of sensors, that doesn't mean that Canon isn't equipped to do that.  Without intimate knowledge of Canon's capabilities, we are all speculating.
Canon potentially benefits a lot commercially from skipping 0.35µm and 0.25µm processes for its larger sensors.

On the compact side, Canon has for a number of year made its own sensors - notably in the G series.  Although these do not have the highest resolution, they do pack a high pixel density.  I think we may be speculating hugely to say Canon does not have the experience of designing to higher pixel densities.

An alternate view could also be that, given what Canon has pulled out of a 0.5µm process, it bodes well for what Canon can do with a 0.18µm process.  Canon will almost certainly release a new generation of APS-C sensors next year, so it will be interesting to see how well such a new design performs, and whether there is a significant performance improvement.

[traveller]

Sorry,

I think that I may have posted on the same topic before I realised that you had already! 

It's good to actually read some analysis based upon research and facts in all the rumour, hype and speculation.  For me, the Chipworks articles show that Canon has been eeking out the last performance from an older generation (or two!!) of technology.  The important thing is that Canon has reached the point where they need to change their fab process and modify their sensor designs to support the new technology. 

It seems to me that Canon aren't in a great position with regards to sensors.  They have been buying in Sony designs for their compact cameras and so lack the design experience as pixel size shrinks, as well as the economies of scale for their R&D departments.  Canon also maintain their own capital intensive fabs solely for larger sensors (APS-C & full frame), which could be what is tying them into older fab technology (i.e. lower production volumes means it takes them longer to amortise the fab assets).

While I think it is almost certain that Canon will need to move to (at least) a 0.18µm process in its next generation of sensors, that doesn't mean that Canon isn't equipped to do that.  Without intimate knowledge of Canon's capabilities, we are all speculating.
Canon potentially benefits a lot commercially from skipping 0.35µm and 0.25µm processes for its larger sensors.

On the compact side, Canon has for a number of year made its own sensors - notably in the G series.  Although these do not have the highest resolution, they do pack a high pixel density.  I think we may be speculating hugely to say Canon does not have the experience of designing to higher pixel densities.

An alternate view could also be that, given what Canon has pulled out of a 0.5µm process, it bodes well for what Canon can do with a 0.18µm process.  Canon will almost certainly release a new generation of APS-C sensors next year, so it will be interesting to see how well such a new design performs, and whether there is a significant performance improvement.

You're correct with the G-series, but that isn't where the volume is: think mobile 'phone camera chips. 

I'm sure that Canon are very good at the 0.5µm design level, but that won't necessarily help at 0.18µm.  My main issue is with Canon's approach of fabbing their own sensors; I think they could get a lot more flexibility by subcontracting the actual manufacturing out and concentrating on R&D.  If they are determined to keep fab in-house, I think that they need to try and expand production by fabbing sensors for other manufacturers. 

[bbasiaga]

An alternate view could also be that, given what Canon has pulled out of a 0.5µm process, it bodes well for what Canon can do with a 0.18µm process. 

No.  Sorry.  The only allowable alternative view on the internet is 'sell all your gear and switch to Nikon'.  Actual sensor performance in the real world need not apply. 

Seriously though, this is a good point.  You see it with computer processors all the time.  The new, smaller achitecture comes out, but for a time there is the older, larger achritecture in the market place with it.  And the real world performance is very small, if noticeable (in processors, its ususally power efficiency that is the major difference, not speed). 

Canon's latest sensors have been of very high quality, and have upped the bar on high ISO performance compared to previous generations.  Has the cow been milked for the last time?  I dont know, and neither does anyone but the Canon engineers.  The technology will reach a limit, and they will have to go to the next genearation, eventually.  We've been conditioned to think that something that was 'state of the art' in technology a few years ago MUST be obsoloete by now (whenever 'now' is).  But that is not always the case.  There are many layers that go in to performance of systems like this, and just because one layer has not changed in its base architecture doesn't mean advancements havent been made in the execution of that architecture, or that advancements in other areas of the system have unlocked more performance from system as a whole.

-Brian

[MarkII]

In conventional silicon design, even 0.18µm is practically an antique process, let alone 0.5µm. I don't think that design expertise is the issue - it is more probably about production expertise and yield issues (i.e. manufacturing cost).

Canon is quite capable of designing sensors that match Sony in their compact range. They use their own sensors in the S1x0 series, having displaced Sony from the S95. If Canon could apply the same technology to their DSLR sensors it would be revolutionary, but I suspect that it would result in a sensor that was just prohibitively expensive.

Canon's ultra-high-res sensor demos from a few years back may be an example of this: you may be able to make a one-off demo, but mass producing the sensor at a viable price point is a very different problem.

[ScottyP]

Maybe the smaller fabs and higher MP will hit a wall or at least fall out of fashion due to diminishing returns on cost, or IQ issues.

How do we kick off a consumer fad for "Highest acceptable ISO rating!!!"  Maybe "New & improved ISO performance"!!!  Have them emblazon that on the camera boxes and the cameras themselves, instead of "XX MP".

[EYEONE]

That is actually a very interesting story. Thanks for posting that.

[Mr Simpleton]

What we heard in the grapewine is that Canon surely do know how to design very high quality and high megapixel FF sensors. But apparently lack the fab to produce them... As building one or two new lines capable to mass produce FF sensors with high MP is such a big investment it will show in business related press and so far no one has spotted any indication that Canon has started to invest in such lines.

So it seems Canon are staying with the old process for a wee bit longer than we want But I sure hope I'm wrong!

[gmrza]

What we heard in the grapewine is that Canon surely do know how to design very high quality and high megapixel FF sensors. But apparently lack the fab to produce them... As building one or two new lines capable to mass produce FF sensors with high MP is such a big investment it will show in business related press and so far no one has spotted any indication that Canon has started to invest in such lines.

So it seems Canon are staying with the old process for a wee bit longer than we want But I sure hope I'm wrong!

It would make commercial sense for Canon to use a new fab with its smaller sensors, as the yields are higher.  Canon has probably chosen the last possible moment to still release full frame sensors based on an older process - the current crop of full frame bodies could have a life of up to 4 years, assuming past technology life cycles can be relied on as a guide (not always a good one).

That means that, assuming Canon launches APS-C (and possibly also compact) sensors based on a new process next year, it has up to 3 years from there to get the production costs using a newer process down to a level where a full-frame sensor can be produced cost effectively.

All speculation, but it seems to indicate that the next 12 months will reveal some interesting pointers to where Canon's sensor production may head.  It will be very interesting to see an analysis of the next APS-C sensor that Canon puts onto the market.