Image Processing Reference
In-Depth Information
but time
T
when amount of integrated signal charge quantity reaches a predetermined
amount Δ
S
q
at pixel
r
k
, that is,
T
(Δ
S
q
,
r
k
).
Because processing circuits are necessary at each pixel in pixel-level ADC sensors and
therefore many transistors are required for each pixel (among other problems), they are
still under investigation and are only used for special purposes. But if it was possible to
use them widely, noise in the readout operation would not be a problem and they would
be very easy devices to deal with. This prediction is reminiscent of the way that CMOS
sensors have become accepted and mainstream as a result of progress such as transistor
shrinkage despite needing multiple transistors and FDAs in each pixel, while CCDs need
only one common FDA.
At the advanced phase of realization of pixel-level ADC sensors, no noise except optical shot
noise would not be allowed. It seems doubtful to think that a solution of the pixel-level ADC
sensor problem would be an extension of the shrunken version of the present ADC circuit.
In photoelectric conversion, one photon is absorbed and one signal charge is generated.
Since a photon is a quantized particle of light, the most precise measurement of light inten-
sity should be to count photons one by one, if possible. Although there are methods of count-
ing photons, they are available only in situations where photons come with enough temporal
intervals, that is, very-low-level illuminance in the present situation. These methods cannot
be applied to high-illumination situations, in which many photons arrive to a sensor part
concurrently or continuously. If it were possible to count each photon directly, this would be
an ideal solution. It is hoped that new concepts and technology will emerge.
5.3.3.3 Sensitivity Improvement Technology
Sensitivity in terms of SNR is the most important and eternal performance issue for image
sensors. While noise reduction approaches were discussed in Sections 5.3.3.1 and 5.3.3.2,
signal-increasing approaches try to decrease optical loss using techniques such as reflec-
tion, absorption, and cross talk through paths to the PD.
5.3.3.3.1 Lightguide (Lightpipe)
On-chip micro lenses (OCLs), discussed in Section 5.1.2, have been very successful as a
means to leading incident light to the sensing aperture, especially in CCDs. In the early
1980s, OCLs were proposed for situations where pixel pitches were around 10 μm with
an aperture of about 5 μm, about ten times wider than the present pitch and aperture.
Focusing at the aperture means sharpshooting it by incoming light to focus on it. But
the incident light angle includes not only the perpendicular component but also a lot of
the oblique component, which decreases the efficiency of the light focus. An inner lens was
developed to suppress degradation of light focus efficiency caused along with aperture
shrinkage and shortening of lens focusing length.
In CMOS sensors, light focus efficiency decreases more because the distance from the top
portion of the OCL to the PD is lengthened by applying plural metal layers for wiring as well
as CMOS logic LSIs. Although three-dimensional wiring by multiple metal layers enables
smaller chips for cost reduction in logic devices, it inhibits optical performance such as lower
sensitivity and higher cross talk. A lightpipe was developed to suppress this type of degra-
dation. Especially in fine-pitch pixels, as optical path widths are close to light wavelength, it
is not easy to arrive at a corresponding PD for incident light because of shielding, reflection,
and diffraction caused by metal wiring through the paths shown in Figure 5.66a.
Some incident light reaches the next PD and this is
cross talk
. In lightpipe sensors, the
optical paths between the color filter and photodiode are filled with high refractive index
