Anyone know where I can buy CCD's at a reasonable price? Perhaps ones with instructions about how to wire 'em up? Voltages and etc.
I'm thinking there's a reason I don't see view cameras converted to LF digital. Technology too obtrusive? Expense issues? Both?
Both. The cost of a CCD chip goes up dramatically as its size increases. This is due to the way chips are made, and a concept called "yield". Basically, chips are made in groups on large "wafers", which are round disks of silicon. The process is not perfect, and flaws on the wafer are almost always present, ruining the chip that contains the flaw. It is always the case that there are flaws, but with small chips, there are plenty of good ones on the wafer to offset the cost of the few that don't work. The larger the chip, the more likely it will contain a flaw. So the larger the chips on a wafer, the fewer there are, and the lower the yield from the wafer. Once they get large enough, the yield drops to the point that it is not economically feasible to even attempt it.Originally Posted by davidwrogers
A large format chip would basically be one chip per wafer, pretty much impossible as chips are currently made. This is why it took a long time to get to full frame DSLR sensors, and why 645 backs like the IQ180 are so expensive. Size drives cost.
So that drives you to scanning backs, where there is not a single chip covering the area, but a smaller CCD that moves over the image area. It works for stationary objects.
Some people are creating backs for dslrs to mount on large format cameras. That's a cheap way to do it I suppose. But obviously the sensor doesn't cover the whole image area. It takes "cropped sensor" to a whole new level!
I too have searched high and low for DIY-able CCDs, not necessarily large format. I would love to have digicam-sized CCD to make a digital back for my Olympus OM cameras. Or a linear CCD to make a roller-transport scanner for roll film. Sadly, even when you can find the bare chips, they require some rather heavy-duty processing power to read the data out. It's a shame nobody sells them on breakout boards with hardware to read the image data to RAW or JPG with an I2C or a micro-SD card slot.
Science is what we understand well enough to explain to a computer. Art is everything else we do.
--A=B by Petkovšek et. al.
If you can deal with tiny (1/3"), there are plenty of cheap camera-modules you can buy that have serial jpeg or raw output. Yield is definitely a problem but as you go to (much) larger process feature-sizes, the defect rate drops accordingly. If you wanted a low-res (8000x10000) CCD for 4x5", that's an 11um pixel-spacing, which is gargantuan for a modern process. You could churn those out with very high yield even from an old fab but there's still the unavoidable silicon and electronic interfacing costs. And you're not doing any better in your results than 6x7 film or a 2012-era commercial 645 digital back.
For larger (non-integrated) sensors, you're well into "you must be an electrical engineer to do this" because of the need to perform high-speed low-noise ADC then handling the significant quantities of data produced. The reason the DSLR world moved to CMOS was because the (small) loss in visual fidelity allows them to pack a bunch of conversion electronics into the same chip, between the pixels. That means the data leaving the chip is generally digital and you don't need crazy-fast ADCs sitting next to the chip.
I don't have the link on me now, but I vaguely recall (from pictures posted on photo.net or something) that one nutter had a matching pair of LF CCDs manufactured and mounted in neat aluminium backs. They put about $250k or something into the project, which they expected to pay off in a few years compared to paying for film at the rate they were burning it for commercial use.
I believe digital x-ray systems use large-format CCDs with a phosphor layer on the front to give xray sensitivity. They're extremely low-resolution though.
ex-Pic-A-Day (slowed after 2 years)
on flickr
Analogue Photo and Film FAQ (for APUG)
Open Source F/Stop Timer
I think it is safe to say that at this scale, film is pretty much the highest tech around - best bang for the buck in image capture. There is no cost effective CCD that's going to get you where large format film already is in terms of resolution, at least in 2012. And the world will end in 10 days, so don't expect anything better! :-P
Since there are a few engineers here, a curiosity question: Rather than making one very expensive and large CCD for a view camera, couldn't someone network a number of smaller CCDs? If I understand correctly, supercomputer equivalents are made today by networking lots of PCs, couldn't something similar be done to avoid the difficulties of manufacturing large CCDs? (Since it hasn't been done, there are obviously problems with this approach, I just wondered if there was a short explanation.) What triggered my question was a very nice photo posted in the "Haven for Small Formats" thread, which consisted of 5 digital photos stitched together, and 5 x (1x1.5) = 5x7.5, or in better English, stitching 5 35mm full frame sensors would give you a 5x7 sensor.
Um, Peter, the 35 mm frame is 24 x 36 mm. The aspect ratio isn't quite the same as 5x7's. To cover 5x7 with 24x36 chips needs roughly 9x6 = 54 chips, which will cover 216 x 216 mm.
Can we compromise? 24x36mm is pretty close to 1 x 1.5 inches (actually .93 x 1.42 inches), but where I really messed up was in the number of sensors. If we lay out a grid of 5 across and 5 down, the long side would be 5 x 1.42 inches, or almost exactly 7". The short side would be 5 x .93 inches, or 4.65". So we would need 25 sensors to get a 4.65 x 7 back. I went back to check if I had really made a mistake in the mm to inch conversion of 35mm negative dimensions, which isn't far off, and to realize that you were absolutely right about my math, it wasn't 5 sensors, but 5 sensors along each side, a 5x5 array. Nonetheless, my original curiosity, why a manufacturer couldn't link an array to avoid the manufacturing problems of one 5x7 (or 4x5) chip still holds.
Peter, the "networking" you refer to is called "tiling" in the packaging industry. It has been done on a limited scale for otherwise intractable problems when imaging arrays are too small for the purposes that are needed.
I've been involved with such projects (unfortunately), but the scale and precision of the equipment needed is significant.
For example, we can usually make the individual chips in an array on a wafer with a layout that provides for precision dicing, by sawing or laser scribing. But the precision needed for the sawing needs to approach the pitch of the pixels so they can be tiled seamlessly without noticeable spacial error ( within a few µm). Then the exact placement and fixing into alignment with each other is formidible (special machine needs to be used).
Next the real problems begin. Row and column readout is very challenging because the mounted chips are still discrete entities so must somehow be electrically interconnected without obscuring the sensor. It is possible to do this using silicon thru hole vias which bring the sensor wiring to the back of the chip to an interconnect carrier substrate. This kind of technology exists but not in any kind of sensor manufacturing environment. The plethora of pixels around the edge of each chip will each need a thru via so clearly that will use chip real estate.
Theoretically all these problems can be surmounted but (not having done a cost analysis) I think that in the end the cost would be equal to the cost of making one large sensor. After all you still need enough chips to cover 4X5 etc. (you need 4X5 inches of silicon), Then you need to engineer seamless tiling.
Actually the Xray panels mention above are quite interesting. These would be immediately available for B&W large format work. They come in sizes suitable for chest Xrays in the 100 to 200µm pixel range. Higher resolution plates are available at down to 30 to 50 µm pixel size in smaller format. Generally these plug into a reader which delivers the image to a computer. Still not cheap to buy but reportedly an 8X10 panel may be on the order of $1000 to $2000. To buy a reader I don't know - check with your local hospital.
Nate Potter, Austin TX.
There may be an ability to make image sensors using the same thin-film-on-glass techniques used to create LCD displays. The infrastructure for this is in place for LCD displays. Now that mobile devices are coming out with these 'retina' displays of 300+ ppi, resolution is approaching usability for an image sensor. However, 300PPI is still only good enough for 'contact printing'...an 8x10 would be only 7Mp.
Science is what we understand well enough to explain to a computer. Art is everything else we do.
--A=B by Petkovšek et. al.
Posting Permissions
Reply With Quote
Bookmarks