Fill Capacity of Canon Power Shot G1x

[Peter Pershan]

I am trying to estimate the well capacity of the MOS in the Cannon PowerShot G1X.

I have read to expect that the fill capacity of an MOS is of the order of 1000 to 2000 x the area in Sq microns.

I am guessing that for the G1X sensor of the order of 200 sq mm and ~14 x10^6 pixels the area of
one MOS is of the order of 14 or 15 sq microns . If the well capacity is 1000 to 2000 times the MOS area this would yield a well capacity for the MOS of the order of between 15000 and 30000 , which corresponds to between 14 and 15 bits.

Furthermore the system gain (electrons/Analogue to Digital Units) would reduce the number of stored bits
by 10 or more..

Since raw images from the G1x of 16 bits my estimation of the well capacity seems much too small

Can anyone help me to understand what is wrong with my thinking?

[herbert]

Hi Peter,

Your calculation seems about right.

The following website collates that information using the raw capture data from DxO Mark.

http://www.sensorgen.info/

They list the G1x as 22719 electrons. They also have information on how it is calculated.

Hope this helps your homework.

Alex

[Peter Pershan]

Alex

Thanks for the reference but this leads to another question.

To be explicit I am considering just one camera, the Canon Power Shot G1X;
however, the same question arises with any other camera that I looked at.
Sensor: 14 MP (18.7 mm x 14 mm)
Area per Pixel: 18.7 μm2
Gain: (I guess) 25 (elect)/ADU
Full Well Capacity: ~ 28,000 electrons (Guess ~ 1500 electrons/μm2)
~1000 ADU ≈210ADU or 10 bits per pixel
RAW Shooting: 14 bit
Assume that the 14 bits are divided into three color channels (i.e. use 4 for simplicity) → Approximately only 12 bits per color channel.

My Question
The difference between the ~12 bits in the RAW data and the 10bits that are needed to record the full-well-capacity would seem to imply that the RAW image had about 2 bits more than was required for the image.

What is the purpose of these 2 extra bits?

[xpatUSA]

Sensor: 14 MP (18.7 mm x 14 mm)
Area per Pixel: 18.7 μm2

Did you account for the true (not effective) fill factor? Well capacity is a function of the photo-sensitive area, not pixel pitch squared.

Gain: (I guess) 25 (elect)/ADU
Full Well Capacity: ~ 28,000 electrons (Guess ~ 1500 electrons/μm2)

Somewhere I read that there is a correlation between conversion gain and well capacity. Obvious, if you think about capacitors and their response to coulombs). I'll see if I can find it . .

RAW Shooting: 14 bit
Assume that the 14 bits are divided into three color channels (i.e. use 4 for simplicity) → Approximately only 12 bits per color channel.

My Question
The difference between the ~12 bits in the RAW data and the 10bits that are needed to record the full-well-capacity would seem to imply that the RAW image had about 2 bits more than was required for the image.

What is the purpose of these 2 extra bits?

The question itself is moot until your estimated full-well capacity is verified.

However, a one-to-one correspondence between the various hardware or software stages of the "pipeline" is not necessary and sometimes quite undesirable. An example is perhaps a digital signal processor. Most likely it does 16-bit arithmetic for a fast one. Which means that it's digital input(s), serial or parallel, are expecting 16-bit data. In the case of a parallel input receiving data from a 14-bit device the upper (most significant) bits are simply not connected, usually grounded if high = 1.

[Peter Pershan]

The Full Well Capacity that i used is the number of electrons that saturate the detector. This does not involve the fill factor.

The question is simply why the raw file seems to contain more bits than is need to account for the maximum number of ADU that is in the sensor pixel.

[Peter Pershan]

I just realized that the number of bits that are needed when the well is filled to capacity increases by about 2 when the gain is increased from 25 electrons/ADU to 5 electrons/ADU.

The mystery is solved.

Thanks to everyone.

[xpatUSA]

This does not involve the fill factor.

Is that so?

You said 14MP and 18.7mm x 14mm. But then, by saying that the pixel area is 18.7 um^2, you have obviously calculated the area of the pixel assuming that the entire pixel is a photodiode - with no border and no CMOS circuitry! Normally a CMOS sensor pixel would have only approx. 60% of it's area sensitive to light. Even the so-called back-lit sensors have some unused pixel area. Therefore, the fill factor is most certainly involved in the estimation of well capacity and it would be quite difficult to prove otherwise, IMHO.

So the sensitive area of your quoted pixel is much more likely to be 11 um^2 giving an estimated well capacity of only 16,500e- at your estimated 1500e-/um, not 28,000e-.