Largesense: Large format digital backs

[BernardLanguillier]

Has hassy already announced a MS version if the 100mp back?

This should be very special!

Cheers,
Bernard

[BJL]

This one is of 78μm size... it nearly matches the characteristics for X-Rays... There is no IR filter either and there is no mention whether it is a Cmos or a CCD sensor... I guess if it can do HD video at that size, then it may well be Cmos...

It is almost certainly neither CMOS nor CCD, but uses a TFT type panel as in LCD displays. The largest CMOS or CCD sensors are full wafer ones made on 200mm wafers, while this 11x9" sensor is too big even for a 300mm wafer. That very different technology probably explains the low well capacity and high minimum usable ISO speed.

[torger]

Torger,
2400DPI is only needed for dithering, to get many shades of gray from black ink, and fine final gradations from dots of ink of just a few colors. Nowhere near that is needed in a TFT display than can vary the brightness of individual pixels.  This is the old PPI vs DPI distinction/confusion.

I am referring to typography, black and white no grayscale. Images in books are rasterized, a very different technique resolution measured in lpi and at a much lower resolution than the typography. Open a high quality book and look at the type. It's sharper than the sharpest mobile phone screen.

[BJL]

I am referring to typography, black and white no grayscale. Images in books are rasterized, a very different technique resolution measured in lpi and at a much lower resolution than the typography. Open a high quality book and look at the type. It's sharper than the sharpest mobile phone screen.

For a long time, camera-ready copy for high quality book printing was often required to be at 300DPI (actually I think it was 12 per mm, so 305 per inch). There was no information beyond that level, so any upscaling beyond that was related to optical/mechanical reproduction issues, not resolution needs, and I doubt that those ink-on-paper issues are relevant to the very different process of displaying glowing pixels on a screen.  So I find it hard to believe that monochrome text printing resolution needs have jumped eight-fold from 300DPI to 2400DPI.

Maybe the situation is that modern book typesetting equipment accommodates illustrations, and so the printers are 2400DPI for them, and those same machines are then also used even with "pure text" books.

Human visual resolution is usually stated as somewhere around 1/3000 to 1/4000 of the viewing distance, so even 50 micron pitch, 504PPI, is only relevant with rather close viewing distances of about six to eight inches.  To put it another way, if an image on screen needs 2400PPI (not DPI) to look its best, then a 24MP (6000x4000) image would start to suffer if displayed larger than 2.5" by 1.7".

[torger]

Color images needs much less, and printing technology is much more limited there. For black text on white paper the ability for the eye to detect quality differences is much higher.

However I too find it unlikely that screens will be pushed to say 2400 PPI anytime soon, especially since antialiasing techniques can to a large extent mask the lack of resolution. Only when extremely high resolution can be had without much additional cost, like in the typography printing case, that may happen.

I would guess that 800 - 1000 ppi may be a suitable target in the foreseeable future. Some overkill in resolution is desired as it makes scaled and transformed graphics (where pixels in image and screen mismatch) look better/sharper. Just like with medium format's 100 megapixel sensors there's diminishing returns, but when the technology is there it can still appear because it's possible and it provides some advantages in some situations for some people.

But I guess huge sensors with huge resolution really needs some sort of new technology than we have today. That $100k pricing is not nice, and I assume a standard sensor made as large as possible on a 300mm wafer (that many seems to be able to do today, CMOSIS being one of them), which would be for example 7x10" (pretty close to 8x10) would just be astronomically pricey. It would be interesting to know what is technically possible with current technology though. I guess with a CMOS sensor that large you would get all sorts of timing and bandwidth issues for readouts if you have 6um pixels which would yield about 1200 megapixels. With 9um pixels you get about 560 megapixels which probably is more suitable for the format.

[BJL]

Color images needs much less, and printing technology is much more limited there. For black text on white paper the ability for the eye to detect quality differences is much higher.

. . . I assume a standard sensor made as large as possible on a 300mm wafer (that many seems to be able to do today, CMOSIS being one of them), which would be for example 7x10" (pretty close to 8x10) would just be astronomically pricey. It would be interesting to know what is technically possible with current technology though. I guess with a CMOS sensor that large you would get all sorts of timing and bandwidth issues for readouts if you have 6um pixels which would yield about 1200 megapixels. With 9um pixels you get about 560 megapixels which probably is more suitable for the format.

You ignore the higher DPI needs of dithering in ink-jet style printing for tonal gradations in images (be they monochrome of color) as opposed to pure black on white text printing, so I will end for now my discussion of whether there has actually been a recent eight-fold increase in what the printing industry deems adequate for black on white text.  I remain skeptical, but am open to any evidence or citations that you can provide.

As to wafer-scale CMOS (or CCD) sensors: they exist with Teledyne-Dalsa for example offering them on 200mm wafers (about 8" diagonal) and I have seen research papers referring to 300mm versions in development (about 12" diagonal).  However in everything I have read, the sole target market is X-rays, where 50-70 micron pixel pitch seems to be enough.  For that pixel size, the best suited fab equipment is steppers like the Canon FPA-5510iV with a big field size (54x34mm) but relatively low resolution (1.5 micron minimum feature size), as opposed to the higher resolution (<=0.5 micron and usually <= 0.35 micron) steppers and scanners used to make modern camera sensors, none of which goes beyond an industry-standard 33x26mm field size.

By the way, these X-ray panels are often designed for butting along three edges, so that they can be merged in to 2x2 or 2x3 devices, with the join lines in the images not a problem for X-rays but unacceptable for large format artistic photography. That probably sets an upper sensor size limit in the medical market of where a 2x3 butted sensor array is big enough for a chest X-ray (which is the biggest medical X-ray format, from what I have read.)

So I predict two options for such large format sensors:
- Ones with pixel pitch limited to the needs of large X-rays, about 50 microns and up, made with large field low res. steppers, at rather high prices; possibly only useful for "digital Polaroids": test shots in preparation for film photography.
- Ones with the smaller pixel pitches needed to make an expensive 10"x8" LF camera for making final digital images, made as wafer-scale CMOS devices with on-wafer stitching of many small fields using 33x26mm field size steppers, and so not sharing design or production costs with the market for X-ray sensors, and at far, far higher prices – probably "custom made for vain dilettante billionaire" pricing.

Seriously, a 5"x4" sensor, doable on a 200mm wafer, makes more sense as a goal for large format digital view cameras.

[torger]

Inkjets are not used to make newspapers, books, stamps, bank notes. You use platesetters and then a printing press. For newspapers a bit lower resolution platesetters are used, which do about 1200dpi. For books 2400 dpi, and for special applications like stamps you can go higher to 5000+ dpi, but that's rarer. Google Kodak Trendsetter for example if you want to find actual products.

These don't dither, it just produces plates which then can be used for offset printing with CMYK or more colors. For images they do raster masks and resolution is then measured in lpi, you get say 450 lpi from a 2400 dpi platesetter, which requires 1.5x=675 ppi original image to avoid aliasing.

The original question was however if the printing industry use this high resolution only to make rasters, or if the resolution is there to make sharper black and white typography and other vector graphics. Well, finding a "reference" for that is not easy, you can just conclude that even black-and-white books without images are printed at these resolutions, and platesetters dedictated for newspapers are limited to 1200dpi while those designed for making books have 2400 ppi, and those designed for printing stamps etc double that. Stamps and bank notes are looked at with loupes so I guess those don't count I suppose.

When arguing about resolution and print quality the same type of argument as within photography arises -- you know the discussion "you only need 4 megapixels from the camera because the eye cannot resolve more", and indeed in certain viewing conditions that is true. However exceptional quality is not about printing pixels smaller than we can count, it's also about the side effects like rendering a sharp edge on a letter. Even the highest ppi screens on the mobile phones today indeed render smooth fonts, but they are just that smooth, they're not sharp. Antialiasing techniques make the edges soft. To make really sharp edges you should have a transition from 100% black to 100% white, no grayscale inbetween. And to make that look really good you need very high resolution.

The reason newspapers only have 1200 dpi limit is that the quality of the paper is not good enough to handle a 2400 dpi resolution.

The platesetting technology has been capable of very high resolution for a long time. I think that the main reason why they have these high resolution is because we can, the technology makes it "easy" to make high resolutions. Platesetting technology of course has nothing to do with sensors, it's just an illustration that if a technology appears that easily can achieve high resolution then it's likely it will be used. As you say we're probably not there yet with CMOS technology, as the steppers that can do large scale sensors can't make small featuers.

I've heard about people that make prints from 8x10" negatives without upsizing, achieving a special look from the ultra-high resolution. Going overkill can have subtle effects.

For really large sensors maybe we need a breakthrough in chemistry in addition to electronics. Imagine a film sheet that could be exposed over and over again, and scanned + reset in place...

[torger]

For those interested, here's paper about the sensor used in Mitch Feinberg's back (that is not largesense):

http://www.imagesensors.org/Past%20Workshops/2011%20Workshop/2011%20Papers/R46_Loijens_596cm2.pdf

That one is indeed a modified X-ray sensor, 9x10". It's not a single chip though, but four separate chips that have been butted together. The gap between the sensors are 80um, which is about a single pixel in this case that is possible to interpolate.

[BJL]

For those interested, here's paper about the sensor used in Mitch Feinberg's back (that is not largesense):

http://www.imagesensors.org/Past%20Workshops/2011%20Workshop/2011%20Papers/R46_Loijens_596cm2.pdf

That one is indeed a modified X-ray sensor, 9x10". It's not a single chip though, but four separate chips that have been butted together. The gap between the sensors are 80um, which is about a single pixel in this case that is possible to interpolate.

Thanks for that about Mitch Feinberg's backs using four butted Teledyne-Dalsa CMOS panels: I had been misinformed about the technology of his sensors.  It still leaves the issue that the availability of such big sensors is driven by the needs of large X-ray detectors, and I see little sign that such X-ray usage will ever push the pixel size down to a level of much interest for actual large format digital capture, as opposed to Feinberg's test shot usage.  Even if the pixels get smaller on each of the panels, that roughly 80 micron gap between the panels would probably not go down much, so maybe the best hope is 300mm wafer-scale X-ray panels.

[BJL]

Inkjets are not used to make newspapers, books, stamps, bank notes. You use platesetters and then a printing press. For newspapers a bit lower resolution platesetters are used, which do about 1200dpi. For books 2400 dpi, and for special applications like stamps you can go higher to 5000+ dpi, but that's rarer. Google Kodak Trendsetter for example if you want to find actual products.

These don't dither, it just produces plates which then can be used for offset printing with CMYK or more colors.

My point is that in printing, getting the tonal gradations for a photograph or gray-scale image from a handful of inks requires controlling the ink density, which is done with variations on the strategy laying down dots of ink far smaller than the pixels or details being represented: the DPI needed is far higher than the LPI of image resolution.  For example, the specs for the Kodak Trendsetter Q400/Q800 Platesetter include resolution of up to 2400 DPI, along with screening at a maximum of 450 LPI.

Sensors and displays are different in allowing fine gradations of the signal level detected at a photosite or output at a display pixel, so I would not extrapolate from the DPI needs of printing to the resolution needs of sensor of displays.  Maybe the LPI specs of high-end publication printing equipment is more indicative.


Aside: It is good to see Kodak still doing something well!  (I know, it does in fact still make film, having only sold the distribution side.)