How many would be happy with the 6D, only if it had the Exmor sensor

Started Oct 3, 2012 | Discussions thread
dosdan Regular Member • Posts: 499
Re: EXMOR loses advantage as ISO goes up?

Exmor technology mainly helps with read noise & FPN. At high ISO you are boosting the sensor signal a lot. This helps with overwhelming a noisy ADC. If the sensor read noise is low, then the "high ISO" noise figure will appear good. (High ISO RN is mainly related to QE.)  But anytime you hear the term  "good high ISO DR" (usually in Canon circles), substitute "high ADC noise, probably low sensor noise".  Ideally, it would be nice to have both low sensor read noise and low ADC noise.

That's what you see with Exmor sensors. Look at the input-referenced RN vs ISO tables & curves at  Compare a Exmor sensor (e.g. K-5) with a Canon sensor e.g. (5D III). Compare the low-ISO region of the curve to the mid-ISO region.  With a low-RN sensor, you will see a low ratio of the total RN in the change from the low- to the mid-ISO  region e.g:

K-5:  3.3 e- @ ISO80 vs 1.9 e- @ ISO1600: The ratio here is 1.73:1.  1.84:1 if you use ISO100 instead.

5D III:  33.1 e- @ ISO100 vs  3.9 e- @ ISO1600: a ratio of  8.49.  13.8:1 if you use ISO12800.

Now the size of the sensels is different in these two cameras: 4.75μ vs 6.25μ, so ignore the absolute RN values and look at the ratios of the drop in total RN from low to mid ISO. A high ratio is a sign of high RN after the PGA i.e. RN from the ADC. The total RN is made up of the quadrature addition of the the two main read noise sources: ADC RN  and Sensor & PGA RN.

Clarifying this:

Read noise (RN) that originates from the ADC is at constant level, but “input referencing” is used to consider all noises levels as if they came as signals from the sensor. Since input-referencing sees ADC noise as if it was coming from the other side of the PGA gain stage, the total RN value can vary. Say that for ISO100 unity gain was used in the PGA. So, with ISO200, 2x PGA gain is used. Say the ADC noise level is 8 “electrons” (using a sensel term here: 8 electrons of charge that have been liberated by photons and converted into a sensel output voltage). These input-referenced ADC RN values, expressed as electrons, would be:

ISO100 8 e- (electrons)
ISO200 4 e-
ISO400 2 e-

Input-referenced RN curves eventually flatten out at mid-ISO due to sensor & PGA read noise, which really are affected by PGA gain, becoming the dominant noise sources in the total RN figure.

E.g. RN from the sensor is 2 e-

RNtotal = Sq-root( RNsensor^2 + RNadc^2)

ISO100:  Sq-root(2^2 +  8^2) = 8.2 e-
ISO200:  Sq-root(2^2 + 4^2) = 4.5 e-
ISO400: Sq-root(2^2 + 2^2) =  2.8 e-
ISO800: Sq-root(2^2 + 1^2) =  2.2 e-
ISO1600: Sq-root(2^2 + 0.5^2) = 2.1 e-

The trouble with ADC RN & FPN is in boosting the rendered image brightness, particularly in the deep shadows. This can be a problem even at base ISO. FPN is a correlated noise, and the human brain easily picks up small amounts of pattens in the chaos. But standard noise RN values don't really indicate how obvious & objectionable the noise components actually are.

PN & FPN are "bad" noises. Shot noise can be more subjectively pleasing, particularly with a large number of MP (fine grained "film-like" noise).  FPN, if it is "fixed" in its position, makes it able to be subtracted, if you go to the bother of creating a reference frame. But some PN is random or semi-random and can't easily be removed.

Some links:

A interesting engineering presentation from Sony Pro Video about CMOS vs CCD,circa2009. While CMOS has now surpassed CCD, the video is interesting as an insight into the Sony Exmor technology. The links shown on the video page are not working, but you can get to the relevant section by:

1. Going to

2. Click on the "CCD & CMOS" thumbnail.

3. Click on the video navigation bar to get to 22:00 and watch from there to about 29:47 (the rest is interesting to).

I believe that the sensor in the Canon 1Dx is using 8 ADCs. The Exmor sensors use 1 slower-speed ADC per sensel column, situated on-chip at the end ofeach column. So, for example, I think this means that the Sony IMX071 sensor (K-5, D7000, A580 etc.) has at least 3264 ADCs. This results in low EM hash (due to the slow ADC clocking speed and very short analogue signal paths) and very low vert. FPN & read noise. This is discussed here:;wap2

Column Parallel ADC technology (used in Exmor) is included in slides 27-28 here (circa2001):

An article about in-camera NR vs software NR that contains good info on both noise sources and how Exmor combats them is:

While most sensor geeks will have seen this, for completeness here is Sony's explanation of Exmor:

Sony's announcement of stacked BSI. Also includes:

HDR Movie (combination of 2 differently exposed images per frame in-camera on-the-fly)


 dosdan's gear list:dosdan's gear list
Pentax K100D Super Pentax K20D Pentax K-5 Pentax K-01 Pentax K-3 +1 more
Post (hide subjects) Posted by
(unknown member)
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow