View Single Post
  #7  
Old July 31st 04, 10:36 AM
David Littlewood
external usenet poster
 
Posts: n/a
Default Minimum pixel size

In article CTGOc.219271$XM6.177658@attbi_s53, Joseph Schutz
writes


"Alfred Molon" wrote in message
m...
Is there a lower limit to pixel size, i.e. could pixel size in a CCD go
to 1 micrometer and below (if we disregard noise issues for a moment) ?
--


Of course 1 micron is possible, but noise is the key point.
It is also worth mentioning that lens resolution would also need
to be very good, which would cost more.


The limit must be somewhere of the same order as the limit for computer
chips (since the same manufacturing techniques are used). IIRC, these
are now using 90nm (0.09 microns) technology. This involves using far-UV
light and extremely sophisticated process lenses.

However, to take this as a practical proposition overlooks the fact that
there is a hard limit to how much a lens can resolve: diffraction. This
means that any lens, even a perfect one, will produce not a single point
image (of a single point object) but a slightly blurred spot. The size
of this spot depends on the relative aperture of the lens (the
f-number). Any attempt to resolve below this point will just make the
fuzzy dots bigger.

When do we reach this point? Well, figures I saw a couple of weeks ago
suggest we are there about now. I can't remember where I saw them, it
may have been here. However, a quick back-of-the-envelope calculation
suggests that for a lens at f/2 the theoretical limit for resolving
detail is about 750 lp/mm. Ignoring all the sophisticated aspects of
MTF, sampling theory, Nyquist limits and the like, this implies at most
1500 pixels/mm, i.e. 670nm per pixel. Taking the complex sampling issues
into account, this is going to come out to 1 micron (1000 nm) or worse.

And that is at f/2 - at f/4 it will be 2 microns, at f/8 it will be 4
and so on. If someone can refer us to the more sophisticated table I saw
that would of course be in all probability worked out on a more
sophisticated basis, but the above should be "in the ball park".

The solution? Just as for film: bigger sensor pixels, bigger sensors. It
*will* happen, but only when commercial pressure makes it so.

You can assume that the camera companies trade off the cost
of the lens verses the cost of the sensor. I would imagine relatively
small sensors give the lowest total cost for both.

The high end cameras have lower noise because of larger pixels.
This costs more for the system.

Whilst true, this is not what was asked.

--
David Littlewood