Size is everything - or, the advantages of a large digital back

[Acheron Photography]

Digital MF is intriguing, but I would suggest that there is a lot about it that makes one pause for thought. Cost, for one thing. But also digital MF requires really fine tolerances to achieve high resolution. Focussing is a pig. (See for instance here for a detailed discussion: http://www.josephholmes.com/news-sharpmediumformat.html ). Moreover, since the tolerances are in microns, manufacture is a pig too. Getting a good back, an in-alignment camera, and a good digital lens is not straightforward, especially given that dealers are not in the habit of admitting the multiplicity of issues.

So, what would make this nasty situation a little nicer?

It seems to me that the issue is the high resolution of the back compared to its size. If the photosites occupied 5x4, rather than 6x7 (or less), then they would be a lot bigger, focussing would be easier, tolerances would be a lot slacker, and one could plausibly use older equippment. What I'd like is not a 65M medium format back like the phase, but rather a 50M 5 x 3.5 inch back. Lower resolution, yes, but with the photosites occuping 15 microns on a side rather than 8, a lot of old technology becomes usable again.
(15 microns, by the way, is very roughly equivalent to 66 line pairs per mm, or roughly where a good LF lens resolves. If you want a digital back you can use with your current lenses, there is some argument that you don't want the photosites to be smaller than that, as they assuredly are in all the DMF backs.)

So, size does matter. The counterargument is that producing digital backs with wafers that size is expensive. Really, really expensive. But it won't be that way forever. Rather than buying digital MF which is _always_ going to be horrid to use and is _never_ going to be able to support your existing collection of lenses, why not clamour for (or at least wait for) a nice big digital back which will slot into your 5x4 camera like any other film holder? That would really rock.

[Bruce Watson]

You've hit on some of the reasons I continue to shoot 5x4 film. I'm open to digital capture, but it's got a long way to go to catch up to 5x4 film, much less anything larger.

[harrykauf]

I think the only reason to create equipment to produce bigger chips would be for
photography. But the top end market is not big enough.
Every other area goes towards smaller structures.
Apparently only Canon has equipment to produce a full 35mm sensor in one go.
http://www.chipworks.com/blogs.aspx?id=4626&blogid=86
Everybody else uses some form of stitching. I am not sure but I think the cost per
area is not really going down in chip manufacture but the structures are getting smaller.
Maybe a different type of sensor will be the solution. If you could manufacture it like a
LCD panel you could go much bigger.

something like this:
http://www.newscientist.com/article/...ont-of-it.html

[JRFrench]

If you read that rather interesting article, it will tell you they are lucky to get a 50% yield from make full frame 35mm chips, imagine what the chances would be like for a 4x5 sensor! Also a piece of silicon that big costs many thousands of dollars, at this stage, while probably possible, it would need a large amount of custom equipment and would be massively expensive.

Making it like an LCD, or composed of discreet imaging sensors, would probably be possible, but as said above the market probably isn't there.

[Mike Anderson]

If you're going to have a big digital sensor, you might as well have an attached LCD (or LED) display, just as big, as well, and view, frame and focus looking at the display. That way you could view stopped down, you could even zoom in on corners for precise focusing. You wouldn't be changing backs, etc. the back would always stay on and could be more rigidly and precisely attached. You could even view right-side-up, if you prefer. My cheap DSLR doesn't have live view, but the people with live view DSLRs say focusing looking at the display is the most precise way to focus.

...Mike

[Nathan Potter]

As the article says full 35 mm frame and smaller sensors are generally made using 0.5 um lithography which necessitates a reduction stepper. The i line 50 X 50 mm field Canon machine is ideal, especially if it can be run in a class 1 fab. Running 12 inch wafers is ideal but I bet they (Canon) runs with 8 inch, at least until recently.

The thing with large format sensors though is that the pixel size (thus wiring interconnects) can be larger due to a somewhat poorer resolution of image that might be acceptable. I suspect a nominally 4X5 sensor could be imaged with 1.5 um lithography, which would bring fabrication into the realm of the Canon MPA 600 mirror 1:1 machine. That machine could be outfitted with a stepping stage with a deep UV light source and with smallest aperture would handle the job at three 4X5 sensors per 12 inch wafer. To be financially practicable the large sensors would need to be run in and existing CMOS sensor line using only 12 inch wafers and absolutely be at class 1 or better defect level. ROI could not be recovered by building a new dedicated fab (est. at $2 billion or more). The larger pixel size yields an advantage in dynamic range which will be quite significant. Larger sensor size implies higher capacitance row and column feed interconnects hence slower response times, so motion performance would presumably be reduced.

This is an evolutionary thing along the model of size scaleup of the LCD flat panel industry over the past 20 years from 13 inch to 60 inch.

Nate Potter, Austin TX.

[papah]

Medical imaging uses very large monochrome sensors for digital x-ray imaging:
http://dalsa.com/sensors/products/cm...?countrytabs=0

The sensor shown at the right would more than cover 8x10. Pixels as small as 20 µm are possible in some of these sensors.

[Daniel Stone]

but what's the quality-control like with those x-ray sensors vs. photo-designed sensors? I had to return my first and only DSLR twice(yes, 2 separate bodies, 5d's) because of too many hot pixels.

who knows though. my dentist still uses film, but also has 1 room with the digital x-ray machine in there. really nifty though. shoot and review, good tethering capability :P

[polyglot]

I wouldn't mind one of those 10x12" Dalsa sensors to stick in the back of a view-camera, even if it is monochrome. Hate to think what they cost though - dentists have a *lot* of money to splash around on equipment when they can charge you a couple hundred bucks for dental xrays and do it 10 times a day. 12 bits isn't too great on the dynamic range; it's about par for digital sensors but I would have expected much better given the huge (20-200um vs 6um) pixels.

The classic yield equations are based on a finite probability of error per unit area and for high res 35mm sensors, I believe the yield is somewhere around 50%, which is why you don't get much change from $1k buying them wholesale. However, that's with pretty fine pixel densities - as you reduce the critical feature size on a given process, the error probability goes down - I wouldn't be surprised if moving from 6um pixels to 20um pixels means the fault rate is basically zero, so you might get 90% yield.

Say you're making 4x5 sensors; at 13x the area of 135 and with a slightly higher yield, in quantity they're still going to cost $10k+ and that's without any support electronics, company overheads, profits, etc, etc. If they were produced in numbers like entry-level DSLRs are produced, you're still talking a net cost of the back somewhere north of $50k; in small/realistic quantities maybe twice that.

I'm gonna go ask my VLSI-nerd friends now.

[Acheron Photography]

... However, that's with pretty fine pixel densities - as you reduce the critical feature size on a given process, the error probability goes down - I wouldn't be surprised if moving from 6um pixels to 20um pixels means the fault rate is basically zero, so you might get 90% yield.

Say you're making 4x5 sensors; at 13x the area of 135 and with a slightly higher yield, in quantity they're still going to cost $10k+ and that's without any support electronics, company overheads, profits, etc, etc. If they were produced in numbers like entry-level DSLRs are produced, you're still talking a net cost of the back somewhere north of $50k; in small/realistic quantities maybe twice that.

I'm gonna go ask my VLSI-nerd friends now.

Thank you (and to all the other posters): I would be really interested in hearing your friend's answer. At $50K you are not going to sell much, but at $5K you might get quite a bit more interest. A 90% yield would rock - surely by that point, especially using old school 90nm technology - the proposition would look a little more economic. Even if it doesn't now one might hope that it is only a matter of time.