Characteristics of random telegraph signal noise in time delay integration CMOS
image sensor
Han Liqiang, Yao Suying, Xu Jiangtao
⇑
, Xu Chao
School of Electronic Information Engineering, Tianjin University, Tianjin 300072, China
article info
Article history:
Received 11 May 2012
Received in revised form 21 September 2012
Accepted 5 October 2012
Available online 25 October 2012
abstract
A statistical model based on large sample simulation is established to study the relationship between
random telegraph signal (RTS) noise and the number of time delay integration (TDI) stages in TDI CMOS
image sensor (CIS). Matlab simulation results show that the mean value of RTS noise increases by a factor
greater than M
0.5
when the number of TDI stages is M, and the factor approximates to M
0.5
with larger TDI
stages. In noise histogram, RTS noise exhibits Gaussian distribution when the number of TDI stages is
more than a special value. These results serve as a guideline for the design of TDI stages and the analysis
of noise.
Ó 2012 Elsevier Ltd. All rights reserved.
1. Introduction
As CMOS technology scales down, random telegraph signal (RTS)
noise has become the major source of read noise in traditional
CMOS image sensor (CIS), especially under low light conditions
[1]. In most cases, the source of RTS noise is the in-pixel source
follower [2], and the noise is attributed to trapping–detrapping
phenomenon caused by an individual trap near the Si/SiO
2
interface
with the times in the high or low current state corresponding to
carrier capture or emission [3]. The special mechanism makes RTS
noise be a non-Gaussian distribution noise and exhibit a special tail
in the noise histogram [1,4], which means that some pixels in tradi-
tional CIS exhibit a high noise level [5]. Several methods have been
reported to reduce RTS noise in traditional CIS [6,7].
Time delay integration (TDI) is a particular architecture applied
in Charge Coupled Device (CCD) and CIS. It is widely used in
machine vision, document scanning and imaging for Earth obser-
vation from satellites or aircrafts for its excellent performance
under extremely low light conditions [8,9]. Generally the TDI
image sensor is composed of a 2-dimensional pixel array. Assum-
ing there are M pixels in one column and M indicates the number
of TDI stages. The signals of these M pixels in one column are
responded by the same object, and then the signals are accumu-
lated by some particular architecture, which means that the signal
level increases by a factor of M. Ideally, assuming the exposure
conditions of these pixels in one column are identical, the total
noise
r
total
accumulated by M pixels in one column is:
r
total
¼
X
M
i¼1
r
2
i
!
1=2
ð1Þ
where
r
i
is the noise of each pixel in one column, and both
r
total
and
r
i
are root mean square (RMS) values. Usually,
r
i
is Gaussian noise,
such as thermal noise and shot noise. It means that
r
i
is a constant
in noise histogram. Then the signal level increases by a factor of M
while the total noise level increases by a factor of M
0.5
. Conse-
quently, the signal to noise ratio (SNR) increases.
In TDI CIS, each pixel has its own source follower, which means
that the signals are processed by different source followers before
accumulation. And because of the wide distribution of RTS noise in
noise histogram,
r
i
of different pixels are not equal if RTS noise is
the dominant noise under low light conditions. This may lead to a
large deviation of SNR between the numerical results and testing
results, so a model is essential for the prediction of the RTS noise
characteristics and the effects on imaging in TDI CIS.
In this paper, characteristics of RTS noise in TDI CMOS image
sensor are studied. The first step is extracting the mathematical
feature of RTS noise parameters in time domain from traditional
four-transistor CIS, which is shown in Section 2.1. Then a large
sample is simulated to derive statistical regularity in Section 2.2.
In Section 3, results are discussed in detail.
2. Model of RTS noise in TDI CMOS image sensors
2.1. Mathematical description of RTS noise in time domain
It is widely accepted that RTS noise is caused by traps located at
the Si/SiO
2
interface. Assuming the number of traps follows the
Poisson distribution in geometrically identical devices, which is:
0026-2714/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.microrel.2012.10.003
⇑
Corresponding author. Tel./fax: +86 022 27890832.
E-mail address: xujiangtao@tju.edu.cn (J. Xu).
Microelectronics Reliability 53 (2013) 400–404
Contents lists available at SciVerse ScienceDirect
Microelectronics Reliability
journal homepage: www.elsevier.com/locate/microrel