Does any1 know what it would take to improve Foveon high ISO performance?

Started Aug 31, 2012 | Discussions thread
Aku Ankka
Aku Ankka Contributing Member • Posts: 591
Re: One main question

xpatUSA wrote:

jl123 wrote:

Just what is the issue? Was my thought potentially correct that by its nature this sensor doesn't get as much light to each layer of its 3 part sensor?

Not really correct. Many people think that the blue and green layers, for example, "block off" some light from the red layer. But the physics of light reveal that the blue and green layers are, to a great extent, transparent to red light.

You're right. The job of the photodiodes is to collect light and turn it to carriers. If one layer "blocks" a photon from another, it does it by using it itself in carrier generation, thus the photon is just used by this layer and not another.

However, there is also the point that the extra transistors and wiring do block light, thus the QE is lower than it would be without the wiring (13 transistors per pixel for the old Foveon, 6 for the current - conventional pixels use 4,but often share the transitors (like the new Foveon), to make it 2 or so at least in the small sensors). On the other hand there is no color filter array on top of the silicon blocking half the light.

More to the point is a ) the low quantum efficiency ( QE = % conversion of photons to electrons ) and b ) the insanely wide bandwidth of each layer:

So the sensor itself starts off hampered by the need for more light than other sensors (for it's 30% per layer QE) and the complex processing needed to weed out the color information from those overlapping curves. Of course, Foveon cheats by somehow adding the R, G and B quantum efficiencies together and calling it "total quantum efficiency" (the black line above). 70%? . . . don't be fooled.

This is not cheating. It is what the QE for each pixel is. The individual layers each get a part of it which the curve is also shown for completeness. The color processing needed has nothing to do with the sensors QE, any more than for exaple jpg engines of any cameras do. It is just a different (and very noisy) problem.

The QE figures for the Bayer CFA equipped sensors are after the filteration.

One should also understand that direct comparison is not really easy between there two different architectures because of the CFA influence.

Don't get me wrong, I love the layered principle and own a couple of SD10's, but improvements in "ISO" performance, i.e. SNR, lie in the direction of larger pixels and less resolution in a APS-C, APS-H or even FF sensor, I would have thought. Think Nikon D700, e.g.

Pixel size is almost irrelevant for low light signal quality. Not only the non-correlating noises adds up in quadrature, but also a smaller pixel tends to have slightly lower read noise as well. This adds up to trivial differnces.

When it comes to QE, for pixels of the size that are used in DSLRs, the pixel pitch is also quite irrelevant due to the relative small size of the transistors and the existance of microlenses.

The way to get good SNR is by using more light - this means having a bigger sensor and/or higher quantum efficiency, and/or by lowering the sensor generated noise (which would be the key for Foveon).

(Also, in the case of Foveon there will always be the color separation issue which will make color imaging noisy.)

-- hide signature --


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