Image Processing Reference
In-Depth Information
Time
FIGURE 3.7
Pattern diagram of current fluctuation by RTN between two levels.
The image sensors by which RTN is observed are CMOS sensors. These sensors have a
transistor in each pixel for amplification, as will be explained in Section 5.3, and among
them are particular transistors that cause RTN. Because the outputs of specified FETs fluc-
tuate between multiple levels, they appear in images as a white blemish at a particular
position on the pixels.
As a result of recent investigation based on statistical data from numerous MOSFETs
with a wide range of sampling times and systems that can measure very low noise, it
was clarified that RTN exists even at very low levels (amplitudes), and it was suggested
that there is RTN not only in specified groups of FETs, but in every FET. That is, RTN was
always observed where that level was not hidden by the noise of the measuring system. In
addition, RTN was observed from two states to more than four states. This is quite impor-
tant information, which dispels the myth that RTN generation is categorized digitally by
FET size.
The low-noise system made it possible to observe RTN. This fact evokes the idea from
image sensor history that when the origin of the worst noise or white blemish is removed,
the second-worst noise or defect (which is behind the worst one) comes out of hiding and
so on, iteratively.
3.4 Shot Noise
The physical basis of shot noise originates in the random arrival rate of a discrete par-
ticle such as a photon or an electron. The nonuniformity of a photon density distribution
in both the time and space domains causes optical shot noise, because photons obey the
Poisson distribution. The probability distribution of the Poisson distribution
f
(
x
) is shown
as follows:
()
=
fx e
−λ
λ
x
x
!
(3.10)
In this distribution, the expected value is the same as with dispersion and is shown as λ
in Equation 3.10. The expected value is the average value of the photon number for a signal
and is expressed by
S
. The square root of dispersion corresponds to noise
N
. The relation-
ship of both is shown as follows:
NS
=
(3.11)
