Olympus-OM
[Top] [All Lists]

Re: [OM] the end game

Subject: Re: [OM] the end game
From: Kennedy <rkm@xxxxxxxxxxxxxxxxxx>
Date: Mon, 16 Mar 1998 16:25:39 +0000
In article , MR THOMAS N CURLEE <JNVS44B@xxxxxxxxxxx> writes
>Some crude back-of-the-envelope calculations for resolution needed 
>for digital cameras.
>
>Assuming 100 lines/mm needed (about what reasonably good film 
>resolves):
>
>With 25.4 mm/inch, then a 35 mm frame is 25.4 X 100 by 1.5 X 25.4 X 
>100. This is 9.68 million pixels.  With 30 bits per pixel (assume 4 
>bytes) this is 38.7 MB per photo.  Note that many cam corders 
>advertise 300K-400K pixel resolution. Without image compression wee 
>would need a 1.4 GB hard drive to store a 36 exposure 'roll'.
>
>Just a thought about how far we need to go before film will go away.
>
>Tom Curlee
>
Remember the solid state sensor needs 3 times the resolution to match
the film resolution since it needs 3 separate pixels for each point -
one for each colour - this makes your calculations on pixel count out by
around a factor of 9, but the final file size correct, since 30
bit/pixel is actually the coding of a colour triad.

The problem isn't storage capacity or resolution, its the sheer size of
the detector that is the problem.  Current CCD and CMOS devices
operating in the visible band have pixel sizes of the order of 5um or
less, giving the necessary 200pixels/mm required to get close to
reasonable speeds of film, but existing chips are tiny comapared to even
the compact 35mm format - most camcorder and digital camera chips are
around the 8x12mm size or less.  Some devices have been made which are
much larger than this for astronomical applications but these are very
expensive and will remain so for some considerable time.

The cost comes from two factors :
Yield - defect densities in silicon have reached a plateau of the past
15 years so increasing the size of chips increases the probability of a
fatal defect occuring within the die area.  For die sizes of 2 square
cm, the yield can be as low as 1 0.000000or larger dice, perhaps 0.01%.
This equates to around 1 working die per batch of 8inch wafers - and
then all these thousands of devices must be tested to identify the
working one.  Silicon foundries are notoriously coy about quoting yield
figures for generic processes because the figures are relatively easy to
determine and they don't want to get into arguments about decimal
percentages yield advantages.  Big pieces of active silicon are however
expensive - thats why the Pentium II chips have the cache off chip,
rather than on the same silicon die.

Lith technology - current lithographic processes used by the major
silicon houses have a limited field coverage for the step and repeat
projection units.  The largest Canon steppers can support die designs up
to around 28mm diagonals - and this must include the processing
circuitry around the active image area and the bond pads required to
connect the device to the outside world.  Typically, the largest image
sensor which could be processed on such units would be around 2 square
cm.  There are techniques which are available to produce larger devices,
such as direct ion beam writing to silicon or reticle stitching, but
these are not mainline silicon processes, have appallingly low yield and
throughput and are therefore extremely expensive.

Whilst it is likely that technology could be developed to overcome the
second limitation and permit large area chips to be produced by
conventional silicon processes, it is unlikely that the defect density
of silicon will fall in the foreseeable future.  Consequently such chips
will always be far too expensive for the likes of you and I to get hold
of them ($x000 each), but are more likely to be within NASA and other
organisation's budgets (>$x00,000 each).

Some low density chips (say 5um pixels on a 20um pitch) might appear
once fabrication technology develops - the lower density will permit
defects to be present but not on active areas.  These devices will fall
far short of the resolution required to rival silver technology and the
gross undersampling of such low densities will cause aliasing like you
have never seen, limiting their application significantly.
-- 
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's pissed.
Python Philosophers         (replace 'nospam' with 'kennedym' when replying)

############################################################
| This message was delivered via the Olympus Mailing List
| To receive the Digest version mailto:listserv@xxxxxxxxxxxxxxx
|    with "subscribe olympus-digest" in the message body.
| To unsubscribe from the current list mailto:listserv@xxxxxxxxxxxxxxx
|    with "unsubscribe olympus" in the body.
| For questions mailto:owner-olympus@xxxxxxxxxxxxxxx
| Web Page: http://Zuiko.sls.bc.ca/swright/olympuslist.html
############################################################


<Prev in Thread] Current Thread [Next in Thread>
Sponsored by Tako
Impressum | Datenschutz