Introduction
Rapid growth in several key technological areas such as communications,
transportation, computers, software and consumer electronics has been possible
due to tremendous advances in the integrated circuit manufacturing
technology. In the last 25 years, the semiconductor industry has grown from the
Intel 4004 chip containing 2300 transistors to the 9.3 million
transistors of Digital 21164 closely following Moore’s
Law. This has been possible due to the equally rapid advancements in
manufacturing process and IC design technology. Increasing demands for more
powerful and smaller computing machines have fueled the need to
manufacture such complex ICs. It is predicted that this need of the electronic industry
for even more complex ICs is going to grow at an even faster rate. By the
end of this century a state-of-the-art microprocessor may be expected to
contain upwards of about 50 million transistors.
From a cost
point of view, such a growth was unhindered in the past mainly because
manufacturers were able to maintain sufficient volume of production
to ensure low cost per fabricated unit. Increasing demand for
more IC products have, however, attracted more manufacturers making very
similar products resulting in a highly competitive market. At
the same time, the continuous drive towards smaller feature size on an
IC and larger die size itself has caused an increase in the cost of manufacturing.
Tough competition and increasing cost have thus made semiconductor
manufacturing a risky venture.
To be more
specific, the cost of a new VLSI fabrication line producing several different
products using several hundred processing steps, is now
estimated to be close to a billion dollars. Both the cost and
complexity of manufacturing have been observed in the past to increase
exponentially and there has been no indication that this trend is going to
slow down. To maintain a competitive edge, ICs must be
precisely manufactured within tight tolerances. Thus, in order to keep the cost
of manufacturing down one must ensure that no errors are made during
any of the stages of producing ICs from design to packaging.
Manufacturing ICs without any errors is a complex task. Whenever a new process
or product is introduced, the manufacturing yield or the
fraction of correctly manufactured ICs is usually low. One
has to ensure not only that the processes and products are designed to be high
yielding but also that errors in manufacturing are eliminated as
quickly as possible through continuous and timely improvements.
This correction process is known as yield learning.
Rapid yield
learning is key to manufacturing success. High yield not only translates to
lower cost per unit but also means that a larger number of ICs
can be delivered in time to maintain a competitive edge.
Therefore, one must be able to ramp up yield quickly using available resources
efficiently. The rate of yield learning is a function of a number
of inter-dependent attributes of IC design, fabrication process and the
failure analysis facility. These attributes in turn depend on a large number of
possible choices related to design style, products, equipment, technology, etc.
Therefore, the technology to debug the manufacturing line needs to
be more complex and advanced than the product technology
to be effective and, hence, very costly. Optimum exploration of cost-revenue
trade-offs requires adequate simulation and experimental models
to be developed taking into consideration the yield
learning process. Examples of rapid yield learning, which are dealt with in
this report are SSA (Spatial Signature Analysis) and the use of information
theory to improve knowledge extraction rate.