Change of the guards… reflection on 100 MP CMOS

[voidshatter]

Hi,

Moore's law is actually a set of laws. The shrinking of component sizes is one of those laws, but it is also about increasing chip sizes. One of the rules says that it is economically feasible to make larger chips.

The other side is that there is a lot of shrinking of pixel sizes for cell phones and other small format technologies like surveillance and machine vision. Jim Kasson has pointed to a moderately technical paper at Chipworks describing some new developments: http://www.chipworks.com/sites/default/files/Fontaine_0.1_The%20State-of-the-Art%20of%20Mainstream%20CMOS%20Image%20Sensors.pdf

Sensor makers seem to address issues like crosstalk with separation walls and shallow pixels, like 1.5 micron deep. That kind of technology may solve a lot of issues with beam angle sensitivity of todays pixel designs.

It seems that APS-C is moving to 36 MP and 4/3 to 24 MP. Scaled up to 24x36 that would correspond to 81 resp 96 MP.

Modern pixels are built with an extra capacitor connected to the photodiode, thus increasing well capacity. That may eliminate/reduce of the disadvantages of small pixels, name reduction of dynamic range.

So, many interesting things going on.

Best regards
Erik

Looking back into history it's quite amazing to see how things have developed over years. I still remember that once upon a time my friend said that a professional photographer recommended the Sony A700. Others laughed that the sensor is an APSC joke and is far behind the Canon fullframe. Now, 8 years later, that Exmor architecture has grown up into the Phase One IQ3 100MP, with every generation beating the dynamic range of the Canon and the CCD (Dalsa/Kodak). It is just as amazing as the Sandy Bridge architecture of Intel CPUs - given so many years, AMD is still unable to catch up!

[voidshatter]

Moore's law don't apply to camera sensors. Miniaturization doesn't get you much beyond what is already possible with current CMOS tech. And the size of the chip determines cost for the most part. This new generation of MF sensors have mature CMOS technology developed elsewhere. I don't see it being updated quite so frequently as smaller sensors. It will not be obsolete any time soon.

Canon has been showing off their 120 MP APSH sensor for many years now. Clearly it's not strictly Moore's Law, but hopefully you see my point.

It is inevitable that at some point the Phase One IQ3 100MP's price will be crushed by advancement of 135 format. This is just electronics.

[Sean H]

Hi Yunli, the hyperlink does not seem to work. However, when I pressed the 'quote' button here, it exposed the URL for me and that works and I will put it here for anyone who is interested:

http://cpn.canon-europe.com/content/news/120_megapixel_apsh_format_cmos_sensor.do

Sean

[ErikKaffehr]

Hi,

I would guess that when 24x36 arrives at 100 MP MFD would have moved on to new designs, too. Essentially, Phase One have jumped on the Sony wagon.

If Phase One can keep their pricing strategy depends on the competition.

Lenses do matter and I would not rule out that Phase One may have a lead in that area.

Best regards
Erik

Canon has been showing off their 120 MP APSH sensor for many years now. Clearly it's not strictly Moore's Law, but hopefully you see my point.

It is inevitable that at some point the Phase One IQ3 100MP's price will be crushed by advancement of 135 format. This is just electronics.

[BernardLanguillier]

Simple keystone correction is fairly easy in software but complex vertical and horizontal lines that would exist in a nodal pan stitch are a nightmare and best shot with movements...... at least that has been my experience.

Were you using PTgui Pro when you did those tests?

IMHO, this is the only stitching software I would trust for accurate planar projection of such images. Both autopano pro and PS generate default results that are very different from what a lens (shifted or not btw) does generate.

I am not saying that stitching is necessarily the best option, there are intrisic problems with stitching in some cases too - not to mention preferences in terms of shooting experience, subject constraints,... - , but I don't believe that the reproduction of vertical/horizontal lines is one of those problems when using PTgui Pro.

Cheers,
Bernard

[narikin]

For pixel-peeping purposes, it might be even better to shoot with the 100MP CMOS unshifted with keystone correction in post-processing, than to shoot with the 60MP Dalsa CCD shifted. I find it difficult to justify the 60MP now, other than cost efficiency.

I wondered the same thing, kind of: does a 20-40 frame pano-stitched and keystoned mirrorless image rival a 4 or 6 way MF stitch on a 80/100 mp digital back? These are 2 ways to 'skin the cat' (delightful saying!) but one pushes and pulls pixels to make buildings vertical & remove angular distortion. A lens-locked MF stitch on a stitching tech cam body, has none of that - it's basically a static 4x5" frame shot in sections, via sliding back, that are simply joined together into 1 larger frame. 

Logically the MF one has no invented pixels, no pushed/pulled fine detail,and should be much better, but maybe in the end there is not such a great difference?

[torger]

I think now at the 100MP level it's starting to become more interesting to test alternatives.

When Hasselblad brings a 100MP back to their system I would not be surprised that their tilt-shift adapter will go obsolete (except for tilting) as you may get the same quality by cropping on the 100MP sensor instead of optically enlarging.

Results in tech cam land I think would show that the traditional method still wins out but it would be interesting to see what the margin is.