End of Megapixel wars and EF-S buying decisions

Started Feb 23, 2006 | Discussions thread
Arjen van Andel Contributing Member • Posts: 594
Re: New technology will emerge

David Martin wrote:

Arjen van Andel wrote:

David Martin wrote:

The whole point of the post though, I should perhaps point out, is
that we can't perhaps be quite so confident as we used to be that
there will be new sensor with greatly improved technology cooming
along soon as we once were!

We might not benefit from more pixels per area unit since those
large diameter lens elements probably cannot deliver such
resolution (at least not the existing glass), but by raising the
per pixel quality and dynamic range there's much room for

There are a number of ways to improve sensors. I will name two:

A. By rain metering the photons instead of collecting them in a
large well and then digitalizing the resulting voltage, dynamic
range can go basically through the roof. By rain metering I mean
using a very small well size that even at f/22 in a dark room
immediately overflows. By counting the number of overflows for each
subpixel in a 12, 16 or even more bits counter, the achievable
dynamic range really has no limit and I suspect noise can be
efficiently suppressed in such a system.

Is this in any was similar to what Fuji have done on the S3? How
much of the silicon real estate is it likely to use and what are
the trade-offs?Have you got any links to any articles explaining
the concept in more depth?

AFAIK the SuperCCD in the S3 has two diffent sensors per pixel site; it doesn't count photon 'raindrops' with any of them in the way that I described; the smaller cell still is relatively insensative and won't overflow quickly.

I have no links because I got this from sucking my tumb.

About real estate. A sensor has two sides. If you create the detectors deep into the silicon in reversed order and place all the wiring and electronics on top, then after production you only have to shave off a bit of the other side of the chip to excavate the detectors. The full real estate of that side is avalable for light gathering and the now rear side's fill real estate is available for switches, amplifiers, counters and wiring. This, I believe, is common practice with photographic sensors and is known as a 'back illuminated' sensor.

B. The current mainstream sensors all utilize the very clever, but
not very precise, Bayer technology. Most pixels in such a sensor
are 'invented', 'guessed', which basically results in
redistribution of detail which leads to soft blurry edges of most
everything in the frame. By using the light filtering properties of
silicon one can stack the RGB sensors, instead of placing them
side- by-side, and that increases the individual pixel site by a
factor of 3 to 4, which improves sensitivity and signal to noise
ratio, and also gives a 3-4 times larger subpixel pitch. Al this
results in much more precisely recorded cleaner pixels with much
higher diffraction-numbers. Such a device will not introduce moiré
and thus has no need for a image-blurring anti-alias filter, which
on it's own will improve sharpness a lot. Such a sensor already
exists, it's made by Foveon and is used in the Sigma SD-series
DSLR's, a Polaroid and a number of specialty camera's. Foveon does
not make a 6 or 8Mp chip (yet) but the results of their 3.3Mp chip
are promising. It has only 3.3 million distinct photosites on the
surface but the result is a 3.3Mp image with extraordinary clarity
which rivals a 6Mp image from a Bayer type sensor. I would like to
see what that technology on APS-C and with 8-12-16Mp can deliver.


Canon, we believe are working on their version of this technology,
but are some way off of being able to use it in practise.
Here is the latest on progress from Paul Pope:
And here is how Canon's version of this works:

Sounds to me as though the limiting case for this technology is
noise - possibly more severely than from current Bayer sensors, as
exposure for each colour is only 1/3rd as long - should be great at
ISO100, but difficult to get good resuts form at higher ISOs

The noise levels might be similar because the detector for the particular color is also three times larger compared to a bayer sensor. But the sensitivity gained by the three times larger sensor is lost by only allowing it to detect 1/3 of the time.

The rain meter principle I mentioned combined with a full color detector would IMO be a better way to get both high dynamic range and true spatial color detection. The Foveon technology isn't matured yet and they also need to put more pixels on the chip before the market wants them. I'd imagine a 6Mp Foveon-like sensor to blow away any 8-12Mp Bayer type.


Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow