Olympus-OM
[Top] [All Lists]

[OM] Re: OT: Digi, Large 22M CCD's

Subject: [OM] Re: OT: Digi, Large 22M CCD's
From: Tim Hughes <timhughes@xxxxxxxx>
Date: Sat, 17 Apr 2004 23:20:57 -0700 (PDT)
Cc: om4t@xxxxxxxx
Wayne S. wrote discussing noise in CCD's pointing out that all process issues 
being equally good,the main determinant of noise, is photodiode sensor area.

>>"There are only so many photons to catch, bigger pixels catch more photons."<<

I agree almost entirely with Wayne , but there are some second order effects 
that **make modest differences** to noise performance, or if badly executed can 
degrade inherent performance. 

In general Wayne's comments apply, because the photo-diodes used inherently 
have similar quantum efficiency. 

That is, the the number of electrons generated per photon of light does not 
vary a huge amount for silicon photo-detectors and thus photo-current scales 
with area. 

Also at very low light levels, light has inherently significant "quantum 
noise", because there are only so many photons arriving at any time.

There is a modest quantum efficiecy change with light wavelength that depends 
on depth of diffusion and voltage bias on photo-diode. 

The intrinsict purity of the silicon used, does have some effect on dark 
current and hence low light level noise.

The photodiode dark current (low light performance) is affected by chip 
temperature, so power hungry chips are more noisey at low light levels. 

This generally favours CMOS chips over CCD. 

If the system noise is sensor limited (ideal good design), a 10C temperature 
rise approximately doubles dark current leading to a 40% increase in noise 
power. 

Also increasing reverse bias on the photo-diode improves the quantum 
efficiency, but it increases the dark current. 

So biasing can make modest performance trade-offs of low-light performance and 
speed vs bright performance and blue sensitivity . 

The more wafer processing steps, the more chances for intoducing noise sources 
from trapped contaminants in oxide etc. 

So chip designs with more layers /processs steps will generally have more noise.

Noise of the readout amplifiers following the sensor should ideally be lower 
than the sensor noise, 

but as is illustrated by the success of niche companies like Nucore, who make 
low noise post sensor amplifier chips, this is often not the case.

The switch + clocking scheme to shift charge from sensor capacitor to read 
amplifier, adds noise in a CCD, this is due to capacitor mismatches and switch 
charge injection. 

Placing an amplifier transistor at each pixel can reduce clock noise at cost of 
silicon area and linearity, which then can affect color mapping.

In general, both of these issues is likely to scale badly with high pixel count 
and/or reduced sensor areas.

Performance improvements tend toward an assymtotic limit set by the physics of 
photo-diodes, which is probably not too far distant from where we currently 
are. 

In theory the Foveon sensors might ultimately have a technology edge because 
they use larger area pixels for a given chip size and don't throw away 66% of 
the light in the "filters". 

In reality the fact the Foveon color sensor seperation, results in broad 
overlap, that then requires removal by subtraction, may make this less of an 
advantage in practise.

Regards,

Tim Hughes



The olympus mailinglist olympus@xxxxxxxxxx
To unsubscribe: mailto:olympus-request@xxxxxxxxxx?subject=unsubscribe

To contact the list admins: mailto:olympusadmin@xxxxxxxxxx?subject="Olympus 
List Problem"
        

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