to empty the FD from electrons. The column-level amplifier
auto-zero (AZ) is performed in order to reset the feadback
capacitance, cancel the offset and reduce its low frequency
noise [9]. The floating diffusion voltage reset level is sampled
at the output of the column-level amplifier. The transfer gate is
then switched-on in order to transfer the accumulated photoelectrons
to the floating diffusion. The voltage level after
this operation is again sampled at the output of the columnlevel
amplifier and differentiated with the reset level voltage.
In some low noise CIS the SF transistor is replaced by a gain
stage like a common-source [1] or transimpedance amplifiers
(CTIAs). It has been shown in a previous work [10] that a readout
chain based on in-pixel gain does not necessarily introduce
a lower noise compared to a readout chain with an in-pixel
source follower scheme and column-level amplification.
The CIS readout chain noise mechanisms are depicted in the
timing diagram of Fig. 1. It shows two types of noise. The first
is the sampled noise constant in time during one readout of
the pixel and randomly changing at the next readout and from
pixel to pixel. It includes the noise sampled at the sense node
after its reset (mainly kTC noise) and the noise sampled at
the integration capacitor and transferred to the output after the
colum-level amplifier AZ [11]. This sampled noise originates
from the thermal and low frequency noise of the SF stage
and the column-level amplifier before opening the AZ switch.
The second type of noise is the random fluctuation of the
signal during the readout originating from the SF stage and
the column-level amplifier after opening the AZ switch. All
these noise sources are uncorrelated and add to each other at
the output of the readout chain.
The double sampling and differentiation (CDS or CMS)
performed at the output of the CIS readout chain has two
main advantages regarding its noise performance. The first one
consists in the cancellation of the sampled noise. The second
one consists in the reduction of the 1/ f noise of the SF stage
and column-level amplifier. After the double sampling, the
dominant noise source is the random fluctuation originating
from the thermal, 1/ f and random telegraph signal (RTS)
noise of the readout chain transistors. RTS noise is a severe
problem that especially appears in small sized transistors like
the in-pixel SF. But for current CIS processes, it only appears
in a minority of pixels.