Image Processing Reference
In-Depth Information
10
-16
1/f noise
10
-17
10
-18
ermal noise
10
-19
10
-20
10
-21
10
-1
10
0
10
1
10
2
10
3
10
4
10
5
Frequency (kHz)
FIGURE 3.6
Measured example of a MOSFET noise power spectrum.
spectrum in a frequency space is also called white noise, meaning that it contains equal
amounts of all frequency component. The mechanism of thermal noise originates from the
thermal random motion of an electric charge as well as that of kTC noise. Although electric
charges flow from the source to the drain through an FET channel as a whole, each elec-
tric charge has components of random motion moving in various directions by Brownian
motion. Therefore, the electric charge distribution in the channel is inhomogeneous. This
means that the density distribution of electric charges that arrive at the drain as current is
not homogeneous and varies with time. Accordingly, this inhomogeneous arrival becomes
current fluctuations, which are detected as voltage fluctuation via the transconductance
of FET, that is, noise. From this mechanism, it is clear that thermal noise is universal for
currents flowing through areas of material having nonzero resistance. Thus, although the
indications of kTC noise and thermal noise are different, the origin of both is physically the
same, that is, the random motion of electrons by thermal energy.
3.3.3 Random Telegraph Noise
RTN is noise in which the channel potential of FETs fluctuates between the quantized states.
Figure 3.7 shows a schematic diagram of current fluctuation caused by RTN. Although in
many cases of a measured example, the quantized state number is two, there are examples
where it is three or four to six.
2
Due to shrinkage of the FET area size, RTN is becoming a
serious issue not only for image sensors but also for flash memories. Concerning its mech-
anism, it is thought that RTN is caused by trapping and releasing an electric charge at the
interface state in an Si-SiO
2
interface. The indication of RTN is the channel potential fluc-
tuation itself caused by trapping and releasing an electric charge. And it is thought that the
two levels correspond to trapped and released states, respectively. The number of interface
states that trap an electric charge decreases along with the shrinkage of the gate area size.
And it is suspected that the quantized level number is two in the case of the trapping state
number being one, and three or four quantized levels correspond to two trapping states.
On the other hand, there is a report that the interface states, which are different from those
causing 1/f noise, bring about RTN.
3
