How many Analog-to-digital converters on Mysterium?

Keith Walters · February 5, 2008

I've just downloaded the RED operating manual, which certainly makes interesting reading. (I've spent quite a bit of time spent wading through the discussions on various forums looking for factual information about RED battery connections and so on, and nobody seemed to have mentioned that the RED.com site has a manual with all that in it, that can be downloaded by anybody, for free)!

I have to admit doing a bit of a double-take at this sentence in the "Theory of Operation" section on page 10:

"Mysterium includes an integrated 12bit resolution analog to digital converter for each pixel,

and is capable of delivering up to 66dB dynamic range (11 stops) when operating at the default

Exposure Index of 320 ASA."

Is that a misprint? Each one of the 12 million pixels has its own ADC? If so, that's an unprecedented level of integration for what is essentially an analog device!

But then the manual is dated 17 Feb 2008, so I suppose it's ahead of its time!

Phil Rhodes · February 5, 2008

It won't be per pixel, it'll be per bin - that is, a chunk of the display of a certain size which is read out independently. This sort of integration is part of the reason that CMOS sensors are attractive to manufacturers - they're easy to build, and you can put more than just the photosites on the same chip.

The Red sensor presumably has a comparatively large number of bins and ADCs because it is a high resolution device which (allegedly) runs at fairly high frame rates. This represents an engineering compromise, the specific caveats of which include quantising noise from the shading correction that must be applied to match all those ADCs, and the fact that the noise floor and highlight response of the sensor as a whole can only ever be as good as the least good bin.

Phil

John Sprung · February 5, 2008

Given the way the rolling shutter works, are the bins likely to be lines, parts of lines, or groups of lines?

-- J.S.

Phil Rhodes · February 5, 2008

I would expect a bin to be a rectilinear area but you could do more or less anything you liked - you just don't want to have to have lots of long connecting traces (for want of the correct term) running around on the substrate. How they read out is just a factor of how they're clocked.

Keith Walters · February 5, 2008

It won't be per pixel, it'll be per bin - that is, a chunk of the display of a certain size which is read out independently. This sort of integration is part of the reason that CMOS sensors are attractive to manufacturers - they're easy to build, and you can put more than just the photosites on the same chip.
Phil

That sounds more like it. As far as I am aware, the D-20 only has 32 ADCs, and that was hard enough to manufacture.

Even Sony's IMX017CQE sensor only(!) has 2,921 ADCs. !2 million on one chip seems a bit of an ask, although admittedly they would only have to process no more than 60 samples per second.

The Sony chip appears to use a successive approximation register, so for a 12-bit device, each pixel would need 12 flip-flops, a 12-bit analog-to-digital converter, a precision comparator, and a lot of driver logic. Multiply that by 12 million, and it would be an awfully complex piece of silicon. In any case, if the photosites are about 5 x 3 microns each, I can't see that there would be enough room for all those parts as well as the phototcell, even with 0.05 micron design rules!

On the other hand, if such a thing is possible they would only have to add a couple more transistors and the ADC output would be able to selectively bleed charge off its associated photosite. That would give each individual photosite a much greater exposure latitude, and film would at last have to start putting its affairs in order!

Maybe sometime down the track there will be a sensor upgrade that does this.