Full text: From pixels to sequences

ized imaging 
on an image 
the pixels is 
red in a CCD 
. Single-chip 
ctionality are 
convolutions 
ght fields for 
sign and the 
on the factory 
rarge-coupled 
published [1]. 
0) pixels was 
emiconductor 
e capacity. In 
yr technology. 
rithms on the 
nsor is just a 
mn, realization 
ng is not only 
ng functions. 
carried out by 
bination with 
rge-area CCD 
ne vision. To 
ns of modern 
"ith more and 
er limit to the 
srformance of 
amental noise 
otography are 
r applications 
media. Image 
called “seeing 
are offered in 
nitry [7]. It is 
. As shown in 
id 72% in the 
2-24 1995 
Today's most widely used semiconductor technology, CMOS [7], can be used without modifications for the manufacture of 
photodiode array image sensors [8], it can easily be adapted to the needs of CCD image sensor technology. As a consequence, 
image sensing can profit directly from the impressive 
advances made for the fabrication of CMOS circuitry. The 
most obvious measure of this development is the minimum 
feature size available in a certain technology, also called the 
design rules. This is illustrated in Fig. 2, showing the m 
evolution of the integration design rules over the past 25 
years. In the early 70's, typical design rules of 6-8 um were 
employed, while todays most advanced memory (DRAM) 
fabrication technology, employed for storage cells foreseen 
for 1 Gbit DRAMS, uses only 0.2 um design rules [9]. At the 
same time, the diameter of the silicon wafers used for 
fabrication has increased, making 20 cm (8 inch) the 
standard for modern silicon foundries. The technology 
employed for image sensors is less advanced, making use 
a 8 
à 
Quantum effidency [9g 
8 
  
today of 15 cm (6 inch) diameter silicon wafers, and using m 
larger design rules. Especially in large image sensors, for 10 
which 1X projection alignment lithographic equipment is 
usually used instead of step-and-repeat lithography, design 0 
rules of 1.2-2 pm are typical. Only the small video image 2 Sm m m a ae ue nm 
sensors, which can be fabricated with the same equipment Wavelength [nm] 
as used for RAMs, are already produced with sub-jm design 
rules, see for example [10]. 
These two simultaneous developments, on one hand the 
increase of the wafer diameter and, on the other hand, the 
reduction of the design rules, demand much more expensive 
fabrication equipment. As it turns out, advances in 
equipment manufacture just about compensate the increasing capabilities of this more expensive equipment: The cost per area of 
processed silicon has remained approximately constant over the years, on the order of $50 per cm”. 
Three major factor are responsible for the dramatic cost 
reduction of integrated electronics, experienced by the 
customer: Firstly, the shrinking design rules have made 
it possible to introduce much higher functionality with a 
larger number of devices and circuits on the same chip 
area. Secondly, the reliability of the fabrication 
processes has improved, leading to higher yields in the 
manufactured devices. Thirdly, modern circuit designs 
are redundant, i.e. more circuit elements (for example 
memory cells) are fabricated on a chip than necessary, 
so that the function of defective cells - inoperable due to 
fabrication imperfections - can be taken over by other, 
spare circuits. Image sensing can profit from all these 
developments, except from the last one: Image data that 
cannot be acquired at the position of a defective pixel is, 
in principle, irrevocably lost. This can be partly 
compensated by the interpolation of the missing data 
from its immediate surroundings. Nevertheless, such 
pictures will never be perfect, and, as a consequence, 
the customer would like to obtain image sensors with as 
few defective pixels as possible. 
Fig. 1 : Quantum efficiency (spectral response) of a backside 
illuminated frame-transfer CCD, thinned to 10 um. 
Reproduced from RCA Engineer, Vol. 29,(6), p.6, 
Nov./Dec. 1984. 
Minimum feature size [microns] 
  
A 
1970 1975 1980 1985 1990 1995 2000 
Year 
Fig. 2 : Evolution of the minimum feature size (design rules) in 
semiconductor technology over the past 25 years. 
3. MORE AND SMALLER PIXELS FOREVER ? LOWER LIMIT GIVEN BY OPTICS ! 
For the reasons stated above, the geometry of the pixels has been reduced continuously in the past 25 years, as shown in Fig. 3. 
This illustration of the design rules' evolution is based on the publication of papers, describing working image sensors with a 
record minimum of pixel size. Of course, it always takes some time to bring these R&D results to production, so that commercially 
available devices usually employ less advanced technology. Nevertheless, the consequences of this development can be seen quite 
clearly by the price drop experienced with standard (CCIR) video cameras. The Japanese manufacturer Sony, for example, offered 
a video camera with a 2/3" (8.8hx6.6v mm?) CCD 10 years ago at the price of around $2000. Today, a complete video camera 
module, offering the same resolution, but using a 1/3" CCD (4.9hx3.7v mm?) is offered at a price of around $250 [11]. Can one 
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop "From Pixels to Sequences”, Zurich, March 22-24 1995 
 
	        
Waiting...

Note to user

Dear user,

In response to current developments in the web technology used by the Goobi viewer, the software no longer supports your browser.

Please use one of the following browsers to display this page correctly.

Thank you.