Pages: 1 2 [3] 4 5 6   Go Down

Author Topic: 48mm x 48mm Sensor  (Read 50922 times)

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
48mm x 48mm Sensor
« Reply #40 on: April 02, 2008, 08:25:19 pm »

Talk of a possible 48x48mm sensor from Dalsa has come up in other threads recently, so I thought I would wake this discussion up and ask:


Is there any news or evidence of plans from Dalsa for a 48x48mm color sensor suitable for use in a medium format camera? Or is such talk all purely in the realm of speculation, rumor and wishful thinking?

Since the rumors of such a sensor have been mentioned numerous times, I am only interested in more solid facts, such as statements from Dalsa or in reliable media outlets.

Given that Kodak is currently the principal sensor supplier for both the Hasselblad and Mamiya systems (including Phase One backs), Dalsa is left with the Hy6 as its main DMF platform, and so maybe one possible "unique selling point" for Dalsa sensors and Hy6 is a roughly 48x48mm sensor. This would also reduce the large mismatch between lenses, viewfinders, mirrors and such designed for 56x56mm and sensors no larger than 36x48mm.

I doubt it though: one third more area would make for quite a price difference, and the vast majority of final images would be produced by cropping away the extra to fit the far more common oblong prints shapes.


P. S. I believe that sensors need to be at least about 6mm narrower in each direction than the film format for which a camera is designed in order to be compatible with a film/digital hybrid body like the Hy6, as the whole sensor including "non-imaging parts" around the edge must fit through a gate the size of the film frame. That is probably the reason that the Leica R digital back used a Kodak sensor of about 18x27mm, not the full 24x36mm that Kodak was clearlty capable of. It might also be one reason that the largest current DMF sensors are about 36.7mm high, while the 645 frame is 42.5mm high.

If so, about 50x50mm is the max for 56x56mm (6x6) format bodies.
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
48mm x 48mm Sensor
« Reply #41 on: April 02, 2008, 09:15:53 pm »

Quote
For example, if 1/4 of chips at one size are usable, only 1/16 will be usable at twice the area, giving about 1/8 as many usable chips per wafer. Another example: Kodak's 36x48mm sensors seem to cost about fifty or more times as much as its 4/3" sized ones, while the area is only eight times as great. And there is competitive pressure on those 36x48mm sensors, from Dalsa.
[a href=\"index.php?act=findpost&pid=125422\"][{POST_SNAPBACK}][/a]

BJL,
I'm having trouble with the logic of the above statement. Surely the number of reject chips will be proportional to the area of each chip. Double the area of the chip and you double the number of rejects. However, mathematics is not my strong point. Is there some esoteric principle at work here?

Also, do we actually have any information on actual manufacturing costs? Whilst economy of scale has a great bearing on price as well as amortisation of R&D costs, it might be an eye-opener if we knew what the unit production cost actually is for different size sensors.
Logged

eronald

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6642
    • My gallery on Instagram
48mm x 48mm Sensor
« Reply #42 on: April 02, 2008, 09:57:34 pm »

Quote
Understood, Nik!

As for the beer at PK '08: I already owe one to "eronald", that would definitively be a "though" PK for me!

Thierry
[a href=\"index.php?act=findpost&pid=125993\"][{POST_SNAPBACK}][/a]

Edmund to my friends ...

E.
Logged
If you appreciate my blog posts help me by following on https://instagram.com/edmundronald

eronald

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6642
    • My gallery on Instagram
48mm x 48mm Sensor
« Reply #43 on: April 02, 2008, 10:07:45 pm »

Quote
BJL,
I'm having trouble with the logic of the above statement. Surely the number of reject chips will be proportional to the area of each chip. Double the area of the chip and you double the number of rejects. However, mathematics is not my strong point. Is there some esoteric principle at work here?

Also, do we actually have any information on actual manufacturing costs? Whilst economy of scale has a great bearing on price as well as amortisation of R&D costs, it might be an eye-opener if we knew what the unit production cost actually is for different size sensors.
[a href=\"index.php?act=findpost&pid=186587\"][{POST_SNAPBACK}][/a]


Ray,

Mathematics is something I can do when pushed  BJL is correct. It is a Poisson statistic.

However, in this case going square means upsizing the chip by 1/3, I'd expect it to be doable. Precise yield statistics are closely guarded secrets, but the fact that 36x48 has been economically doable for at least 4 years indicates that 48x48 is now feasible economically.

Canon now seems to have the ability to mask 48x48  in a single step, Sony  can do at least 24x36, I guess many fabs will soon have this ability making stitching a thing of the past, obsoleting the APS size, and improving the quality of the next chip generation and reducing the cost.

Last not least, from what I have been told the centerfold issue was not as stitching issue. As usual I cannot talk about it - in public- but I'll be glad to tell Thierry

Edmund
Edmund
« Last Edit: April 02, 2008, 10:23:22 pm by eronald »
Logged
If you appreciate my blog posts help me by following on https://instagram.com/edmundronald

thsinar

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2066
    • http://www.sinarcameras.com
48mm x 48mm Sensor
« Reply #44 on: April 02, 2008, 10:37:30 pm »

hi Edmund,

Stefan (Brumbaer) had explained the exact reasons why the effect of the centerfold happens (see the corresponding tread/topic about > 1 year ago).

Best regards,
Thierry

Quote
Last not least, from what I have been told the centerfold issue was not as stitching issue. As usual I cannot talk about it - in public- but I'll be glad to tell Thierry

Edmund
Edmund
[a href=\"index.php?act=findpost&pid=186594\"][{POST_SNAPBACK}][/a]
Logged
Thierry Hagenauer
thasia_cn@yahoo.com

thsinar

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2066
    • http://www.sinarcameras.com
48mm x 48mm Sensor
« Reply #45 on: April 02, 2008, 10:46:53 pm »

You don't seriously think that I've forgot it, do you?!

 

Thierry

Quote
Edmund to my friends ...

E.
[a href=\"index.php?act=findpost&pid=186591\"][{POST_SNAPBACK}][/a]
Logged
Thierry Hagenauer
thasia_cn@yahoo.com

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
48mm x 48mm Sensor
« Reply #46 on: April 03, 2008, 12:00:39 am »

Quote
Ray,

Mathematics is something I can do when pushed  BJL is correct. It is a Poisson statistic.


[a href=\"index.php?act=findpost&pid=186594\"][{POST_SNAPBACK}][/a]

Edmund,
Can I infer from that statement that there is an uncertainty principle at work, at the quantum level, which would make it impossible to produce a defect-free chip, just as it's impossible to capture an image which is free of all shot noise?
Logged

eronald

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6642
    • My gallery on Instagram
48mm x 48mm Sensor
« Reply #47 on: April 03, 2008, 12:06:09 am »

Quote
Edmund,
Can I infer from that statement that there is an uncertainty principle at work, at the quantum level, which would make it impossible to produce a defect-free chip, just as it's impossible to capture an image which is free of all shot noise?
[a href=\"index.php?act=findpost&pid=186629\"][{POST_SNAPBACK}][/a]


I don't think we need to invoke a quantum phenomenon. I think we can probably make small defect-free chips at the moment. The larger ones end up beoh sold with defects, but we tolerate defects which just disable isolated pixels ...

Edmund
Logged
If you appreciate my blog posts help me by following on https://instagram.com/edmundronald

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
48mm x 48mm Sensor
« Reply #48 on: April 03, 2008, 06:54:12 am »

Quote
I don't think we need to invoke a quantum phenomenon. I think we can probably make small defect-free chips at the moment. The larger ones end up beoh sold with defects, but we tolerate defects which just disable isolated pixels ...

Edmund
[a href=\"index.php?act=findpost&pid=186631\"][{POST_SNAPBACK}][/a]

But you've invoked Monsieur Poisson to explain this. I'm curious as to why a Canon G9 sensor of approximately 72 sq.mm, containing 12mp can be so cheap ($100 would you say?) and a P45+ sensor of 24x the area, but much lower pixel density, is so expensive.

Isn't the likelihood of defects also proportional to the number of pixels?

I'm here to learn. Teach me!  
Logged

Sean Reginald Knight

  • Awaiting Authorisation
  • Jr. Member
  • **
  • Offline Offline
  • Posts: 66
48mm x 48mm Sensor
« Reply #49 on: April 03, 2008, 10:06:14 am »

Quote
But you've invoked Monsieur Poisson to explain this. I'm curious as to why a Canon G9 sensor of approximately 72 sq.mm, containing 12mp can be so cheap ($100 would you say?) and a P45+ sensor of 24x the area, but much lower pixel density, is so expensive.

Isn't the likelihood of defects also proportional to the number of pixels?

I'm here to learn. Teach me! 
[a href=\"index.php?act=findpost&pid=186700\"][{POST_SNAPBACK}][/a]

Oooh! Quantum levels and the uncertainty principle. Let's hope it involves di-lithium and tachyon warp drives too.  

Dr. Ronald, pray tell. What has the Poisson distribution got to say about wafer yields?  

P.S. I wanna play "I'm smarter than a 5th grader". Can I play? Can I play?
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
48mm x 48mm Sensor
« Reply #50 on: April 03, 2008, 11:17:15 am »

Quote
Oooh! Quantum levels and the uncertainty principle. Let's hope it involves di-lithium and tachyon warp drives too.  

Dr. Ronald, pray tell. What has the Poisson distribution got to say about wafer yields?  

P.S. I wanna play "I'm smarter than a 5th grader". Can I play? Can I play?
[a href=\"index.php?act=findpost&pid=186722\"][{POST_SNAPBACK}][/a]

No, you cannot play. You've just frightened off Edmund. I was after serious information.
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
48mm x 48mm Sensor
« Reply #51 on: April 03, 2008, 11:19:51 am »

Here is a rough explanation (requiring neither Poisson nor quantum mechanics.)

Suppose that for a sensor of a certain area (say 24x36mm) only one in four will be free of fatal flaws.
Next consider a sensor of twice the area (say 48x36mm) and consider its left and right halves separately, each the same size as the smaller chip.
One quarter of the left halves will work, and of these sensors, one quarter will also have a working right half. So only 1/4 of 1/4 or 1 in 16 will have both halves free of fatal flaws, so the yield is reduced from 1/4 to 1/16.

As to Poisson, I believe that his result is that as area in increased by a factor S, the probability of something (like a fatal flaw) not happening anywhere is raised to the power S, so my example is for S=2, squaring the yield factor.

Of course this is an all things equal comparison for the case where a single flaw is fatal. Maybe pixel count is more relevant than area for the probability of some types of flaw. And some acceptability criteria for sensors are based on a low enough number of defects, rather than being defect free: Kodak's sensor spec. sheets state some such criteria for "point" and "cluster" defects.
« Last Edit: April 03, 2008, 06:41:35 pm by BJL »
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
48mm x 48mm Sensor
« Reply #52 on: April 03, 2008, 11:42:21 am »

Quote
Here is a rough explanation (requiring neither Poisson for quantum mechanics.)

Suppose that for a sensor of a certain area (say 24x36mm) only one in four will be free of fatal flaws.
Next consider a sensor of twice the area (say 48x36mm) and consider its left and right halves separately, each the same size as the smaller chip.
One quarter of the left halves will work, and of these sensors, one quarter will also have a working right half. So only 1/4 of 1/4 or 1 in 16 will have both halves free of fatal flaws, so the yield is reduced from 1/4 to 1/16.

As to Poisson, I believe that his result is that as area in increased by a factor S, the probability of something (like a fatal flaw) not happening anywhere is raised to the power S, so my example is for S=2, squaring the yield factor.

Of course this is an all things equal comparison for the case where a single flaw is fatal. Maybe pixel count is more relevant than area for the probability of some types of flaw. And some acceptability criteria for sensors are based on a low enough number of defects, rather than being defect free: Kodak's sensor spec. sheets state some such criteria for "point" and "cluster" defects.
[a href=\"index.php?act=findpost&pid=186747\"][{POST_SNAPBACK}][/a]

Maybe I should sober up before attempting a response, but there's also the factor of CCD costs as opposed to CMOS costs. Your explanation is a bit difficult in my bleary haze. Tomorrow it may be clearer.  
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
48mm x 48mm Sensor
« Reply #53 on: April 03, 2008, 11:43:33 am »

Quote
Ray,
Canon now seems to have the ability to mask 48x48  in a single step, Sony  can do at least 24x36
[a href=\"index.php?act=findpost&pid=186594\"][{POST_SNAPBACK}][/a]
Canon has for a long time made a stepper with field size big enough for 48x48mm, but does not use it to make its 24x36mm sensors. That jumbo stepper is an old design with a rather large minimum feature size of 500nm, so perhaps it is not suitable for DSLR sensors. (I believe it is useful for devices like LCDs, and the first roughing out of other devices.) Canon has referred twice in white papers to 26x33mm as the largest field size of steppers suitable for DSLR sensor fab.

Nikon used to have a stepper with field size large enough for 24x36mm in its online catalog, but has apparently discontinued it. Nikon and the largest stepper maker AMSL now offer no stepper with field size larger than 26x33mm. That 26x33mm is a de facto industry standard, being the size of many steppers from each of AMSL, Nikon and Canon, with only that single old low resolution Canon model going larger.

Stitching (fabbing a sensor with multiple exposures of different parts of the chip) seems the dominant approach for the small fraction of devices larger than 26x33mm: AMSL offers 2D stitching as a feature on some stepper models, so it is not a trade secret of Canon or Dalsa, at least not anymore. Sony has also mentioned that 26x33mm limit, and both Canon and Sony said that they use stitching to produce their 24x36mm sensors.

So Sony and Canon can do 24x36mm, but only in the way that anyone can by buying (or outsourcing to) a suitable stepper, such as some AMSL models: by stitching. The same is probably true for Dalsa, Kodak, Tower Semiconductor in Israel (which fabbed the original sensor for the Kodak 14/n), the English foundry that fabbed the sensors for the Kodak SLR/N and SLR/C, and whoever fabs the sensor for the Nikon D3. Matsushita also used stitching to make the roughly 18x27mm CCD for the original Canon 1D.

In summary, there are many foundries capable of stitching and thus fabbing a 24x36mm or larger sensor, including third party foundries, so that a DSLR maker does not need to do it in house, but apparently no one can do it without stitching, which keeps costs high.
Logged

free1000

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 465
    • http://www.foliobook.mobi
48mm x 48mm Sensor
« Reply #54 on: April 03, 2008, 12:01:27 pm »

Quote
The trick of the large Dalsa chips is to stitch several smaller ones together to one big surface. This way they can decrease the high outtake rate considerably. A modern CCD as it appears in the Leaf and Sinar backs is built up of six smaller pieces.

Three points.

1) This is what causes the horrible centre fold and the even worse problem of non-uniformity causing magenta and green striping.  I keep hearing they will overcome this, but it seems to be an achilles heel for really high quality imaging for photography.

2) Any sensor sizes and formats created won't be for MF photographers... but for other markets. We just get the spin offs.

3) Is it really such a good idea to have a square sensor? From my POV I don't get a wider angle of view... plus the flange/focal length is now longer making it harder to create good wide angle lenses at affordable prices for MF SLR's. We've seen the cost of a Mamiya 28mm. Will it be practical to even create a 28mm at an affordable price for a 48x48 sensor? Would I prefer such a lens or a new car ;-)
Logged
@foliobook
Foliobook professional photo

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
48mm x 48mm Sensor
« Reply #55 on: April 03, 2008, 12:04:45 pm »

Quote
In summary, there are many foundries capable of stitching and thus fabbing a 24x36mm or larger sensor, including third party foundries, so that a DSLR maker does not need to do it in house, but apparently no one can do it without stitching, which keeps costs high.
[a href=\"index.php?act=findpost&pid=186758\"][{POST_SNAPBACK}][/a]
 

That's very surprising, that no-one can produce a 24x36mm sensor without stitching. Why is this? Are we waiting for future developments to come into play? Is this just a temporary set-back?
Logged

Sean Reginald Knight

  • Awaiting Authorisation
  • Jr. Member
  • **
  • Offline Offline
  • Posts: 66
48mm x 48mm Sensor
« Reply #56 on: April 03, 2008, 12:43:44 pm »

Quote
No, you cannot play. You've just frightened off Edmund. I was after serious information.
[{POST_SNAPBACK}][/a]

I found Edmund (Dr. Ronald to you) rolling around the floor in Paris, with stitches in his side.

Naw, you underestimate him. Edmund won't be frightened off by the Poisson distribution. It is grade school stuff. Me poor head still throbs from sitting my A levels on that. No, I don't know what it means either. He will certainly be frightened off by those who try to appear as smart as he is.

Say, Ray, do you know what 'Plethora' means?

[a href=\"http://wordwise.typepad.com/blog/2007/03/plethora_puhlee.html]http://wordwise.typepad.com/blog/2007/03/p...ora_puhlee.html[/url]

Kind of up there with your 'quantum levels' and 'uncertainty principle'.   Not forgetting invoking Monsieur Poisson.
« Last Edit: April 03, 2008, 03:10:49 pm by Sean Reginald Knight »
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
48mm x 48mm Sensor
« Reply #57 on: April 03, 2008, 05:25:23 pm »

Quote
That's very surprising, that no-one can produce a 24x36mm sensor without stitching. Why is this?
[{POST_SNAPBACK}][/a]
The current reason is that all new steppers introduced in the last five years or more have a maximum field size of 26x33mm or smaller, and possibly all are exactly 26x33mm. (There are some current steppers with smaller maximum field size, but they might be older models.) Of the several stepper models once made with larger maximum field size, only one remains available. Its minimum feature size of 500nm should be compared to the current range of mostly 90nm down to 45nm, with 35nm coming soon.

As to why, I can only speculate, but here I go:


1) The vast majority of all IC devices have die sizes no larger than 26x33mm, and in fact mostly far smaller than that. DSLR's as a whole are a tiny fraction of all IC's, and so do not drive design decisions for new steppers, but rely on using steppers designed primarily for other main markets, like CPU's, memory chips and the smaller sensors for mobile phones, compact digicams, video cameras and such.

Memory chips do not need to be large, as it is easy to wire together multiples, and all the recent mainstream CPU's I know of (from Intel in particular) are comfortably under 26x33mm, even the recent dual core ones.

The biggest microprocessors that I know of are the Intel Itanium processors, and the dimensions I can find are die size of 27.72 x 21.5 mm = 596mm^2 for recent dual core models made using 90nm process:
[a href=\"http://www.chiplist.com/Intel_Itanium_2_9000_series_processor_Montecito/tree3f-subsection--2242-/]http://www.chiplist.com/Intel_Itanium_2_90...section--2242-/[/url]
http://en.wikipedia.org/wiki/Montecito_(processor)
(This is about as wide as a 1D sensor, but slightly higher.)

The other Intel processors families like Xeon and the mainstream CoreDuo and such have far smaller dies sizes, 143mm^2 or less. (That is far smaller than even 4/3" sensor size of 225MM^2).

The smaller sizes of these other processors is largely due to using new, smaller feature size fabrication technology, like 65nm and 45nm instead of Itanium's 90nm, which for one thing allows higher clock speeds. That diminishing feature size seems to be keeping die size the same or smaller in new CPU models, balancing out increases in transistor counts, even with dual and quad core designs.


2) The remaining larger devices are small volume items that can be most cost effectively handled by stitching rather than having  a rarely needed stepper of larger field size: a larger stepper would reduce unit costs for such IC's, but demand for such a stepper would be so low that the cost of designing and building the stepper itself would be too high.


Unless the demand for sensors larger than 26x33mm increases a thousand fold or more, DSLR sensors will be constrained by the size needs of other far larger parts of the IC market.
« Last Edit: April 03, 2008, 06:43:42 pm by BJL »
Logged

free1000

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 465
    • http://www.foliobook.mobi
48mm x 48mm Sensor
« Reply #58 on: April 04, 2008, 07:32:28 am »

Quote
No, Dalsa specifically makes large sensors for photographical use. When you buy their sensors you can choose between three different qualities. A top quality A-grade sensor can be used for reference purposes (and theoretical state-of-the-art backs). A B-grade is fine for general use in DB's, and for those companies that wish to cut costs there is the C-grade sensor quality. C-grade is near-outtake and I suspect these are the sensors used by Mamiya for their ZD back.

Dalsa's R&D department and production facility in the Netherlands (where all these large sensors are being made) is very small. Like a lab. The engineers who work there often have guest jobs as professors in tech universities. Some have written scientific papers about CCD technology. This small scale gives a lot of production flexibility (but also the disadavantage of small scale economics). You can even buy your owns sensors from them and start building your own back. Start with their evaluation kit:

[attachment=5912:attachment]
[a href=\"index.php?act=findpost&pid=186897\"][{POST_SNAPBACK}][/a]

Thanks for the correction... I assumed that although they produced these for photographic purposes it would be for more industrial/military imaging applications.

In that case they need to get their act together and 'like quit with the centrefold man'

I think I'll pass on building my own back for the moment though :-)
Logged
@foliobook
Foliobook professional photo

narikin

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1372
48mm x 48mm Sensor
« Reply #59 on: April 04, 2008, 12:32:37 pm »

Quote
No, Dalsa specifically makes large sensors for photographical use. When you buy their sensors you can choose between three different qualities. {snip} C-grade is near-outtake and I suspect these are the sensors used by Mamiya for their ZD back.
[a href=\"index.php?act=findpost&pid=186897\"][{POST_SNAPBACK}][/a]

wow - thats quite an accusation - I'd suggest checking it for being 100% accurate before you thrown it in there and tarnish Mamiya!
Logged
Pages: 1 2 [3] 4 5 6   Go Up