How big? The answer is 24 (or 42 backwards...)
(You really can ignore this if you have no interest in digital photography...)
I went to a lecture last night on digital camera sensors. As you probably know, each pixel in a digital camera is composed of at least one photosite (sometimes as many as 4, but ignore that for this discussion). Each photosite is basically a switchable photoelectric capacitor: light comes in, some of it is converted to electrons by the photoelectric effect, and the resulting electrons are stored.
Now, here's the kicker. Each photosite can only store tens of thousands of electrons. How many exactly depends on the size, but it's a number like 50,000.
This is really a problem. Why? Well, for two reasons. First the resolution of the camera depends on the signal to noise ratio of the photosite. Quantum effects (in particular shot noise) limit the resolution - and short of increasing the size of the photosite, there is absolutely nothing you can do about this. Second, to read the data out, you need to be able to move and then measure those 50,000 electrons accurately. And that isn't easy either.
Now there are various smart things you can do, including putting in on board noise reduction in the sensor and doing your A2D conversion as near to the photosite as possible. But at the end of the day, noise reduction just amounts to guessing what the signal should have been. The fundamental camera resolution is limited by the number of photosites you can cram on the chip, and the more you cram on, the smaller they get, the fewer electrons they can hold, and the worse your noise problems get.
What's the limit? Well, it turns out that for really high performance sensors, 35 square micrometers is about as small as you want a photosite to be, or roughly 6 micrometers on a side. 35mm is 24mm x 36mm, which translates to 4000x6000 or 24 megapixels. This is more or less exactly where the best full frame sensor DSLRs are already.
In other words, the resolution wars are over. New digital cameras will either cap out around 24 megapixels, have high noise and/or very aggressive and intrusive noise reduction, need bigger than full frame sensors, or use a wholly different light sensing technology from the one we have been using for the last 20 years. High noise is intolerable, bigger than full frame sensors are very expensive because they require huge pieces of silicon, and a new photosite technology does not appear to be around the corner.
Now, 24 Mp is enough for very nearly all applications, and most of the people who need more will move to medium format digital. But still it is sobering to think that a technology that we have been used to delivering seemingly effortless performance increases every two or three years is close to the quantum limit already.
OK, that's enough of that. Back to the economics.
I went to a lecture last night on digital camera sensors. As you probably know, each pixel in a digital camera is composed of at least one photosite (sometimes as many as 4, but ignore that for this discussion). Each photosite is basically a switchable photoelectric capacitor: light comes in, some of it is converted to electrons by the photoelectric effect, and the resulting electrons are stored.
Now, here's the kicker. Each photosite can only store tens of thousands of electrons. How many exactly depends on the size, but it's a number like 50,000.
This is really a problem. Why? Well, for two reasons. First the resolution of the camera depends on the signal to noise ratio of the photosite. Quantum effects (in particular shot noise) limit the resolution - and short of increasing the size of the photosite, there is absolutely nothing you can do about this. Second, to read the data out, you need to be able to move and then measure those 50,000 electrons accurately. And that isn't easy either.
Now there are various smart things you can do, including putting in on board noise reduction in the sensor and doing your A2D conversion as near to the photosite as possible. But at the end of the day, noise reduction just amounts to guessing what the signal should have been. The fundamental camera resolution is limited by the number of photosites you can cram on the chip, and the more you cram on, the smaller they get, the fewer electrons they can hold, and the worse your noise problems get.
What's the limit? Well, it turns out that for really high performance sensors, 35 square micrometers is about as small as you want a photosite to be, or roughly 6 micrometers on a side. 35mm is 24mm x 36mm, which translates to 4000x6000 or 24 megapixels. This is more or less exactly where the best full frame sensor DSLRs are already.
In other words, the resolution wars are over. New digital cameras will either cap out around 24 megapixels, have high noise and/or very aggressive and intrusive noise reduction, need bigger than full frame sensors, or use a wholly different light sensing technology from the one we have been using for the last 20 years. High noise is intolerable, bigger than full frame sensors are very expensive because they require huge pieces of silicon, and a new photosite technology does not appear to be around the corner.
Now, 24 Mp is enough for very nearly all applications, and most of the people who need more will move to medium format digital. But still it is sobering to think that a technology that we have been used to delivering seemingly effortless performance increases every two or three years is close to the quantum limit already.
OK, that's enough of that. Back to the economics.
Labels: Photography
posted by David Murphy at 05:45
2 Comments:
Hmmm... given the limits, the only option seems to be a redesign of the basic photosite. There *must* be some physicists lurking arounnd here... if the fundamental limit is on photosite volume rather than area, is there no way to boost the electrons/ change the medium to make their travel distance longer, and make the photosite wells deeper? If you *could* do that, then where is the new limit from area, i.e how far can you go with a microlens? Or am I being a bit soft here and confusing my basic measurements again?
If you make them deeper, the frequency response gets worse because the transmission function of the CMOS or CCD is a strong function of wavelength. That might be something you could compensate for to some extent in clever A2D, but it would be a challenge. I suspect other non-shot aspects of the noise get worse too, but I'd need to think about the details.
What you really want to do is boost not just the depth but also the efficiency of the photoelectric effect. That would be the smart question: can you preserve the charge shifting properties _and_ make the photoelectric conversion more efficient?
Post a Comment
<< Home