Intro To SMS PDF
Intro To SMS PDF
An integrated circuit commonly referred to as an IC or a semiconductor chip, is a complex device
that consists of miniaturized electronic components and their interconnections. The production of IC's is
accomplished in a four‐stage process that begins with raw plate of silicon or, less commonly, gallium
arsenide and is called raw wafers. Wafers are grouped in lots, which travel together in a standard
container and are destined for conversion to the same final product. The maximum lot size, usually
between 20 and 100 wafers, differs from one production facility to another and may even differ from
one product to another within the same facility. The most commonly seen lot size is a set of 25 wafers.
The first stage of IC production is called wafer processing or wafer fabrication. It is conducted in a
so‐called clean room, where special means are employed to maintain low density of airborne particles.
The term wafer fab is commonly used to mean a clean room in which wafer fabrication is conducted.
Here the intricate miniature circuits for a number of identical chips are created on each wafer. The
individual chips‐to‐be are referred to as dice. The circuitry is created by a lengthy and complex process,
and the number of dice per wafer may vary from just a few to many hundreds.
This number depends on different factors including the diameter of the raw wafers, device
geometries and quality yield. Companies constantly make wafer size transitions and shrink device
geometries to improve their productivity. In high volume manufacturing the current industry standard
for wafer size is 300mm or 12in wafers. The latest achievement in device geometries is reported as
45nm, which indicated the width of an individual transistor on the device. Companies however,
constantly improve wafer size and device geometry because of overall cost benefits resulting from the
larger number of dice per wafer, thereby using the same number of process steps to produce more dice.
Wafer fabrication requires a long sequence of processing steps and involves many separate pieces of
equipment, through which lots of wafers are routed in the traditional job shop fashion.
In the second stage of IC production, commonly referred to as wafer probe the individual dice on a
wafer are tested for functionality by delicate electrical probes. Dice that fail to meet specifications are
marked with an ink dot, the wafers are scored and broken into separate individual dice and the
defective dice are discarded. In the third stage of production, called assembly, electrical leads are
connected to the individual dice, which are then encapsulated in plastic or ceramic shells called
packages. In the fourth stage of production, packaged chips are subjected to a final functional test and
burn‐in.
Perhaps the greatest single determinant of economic success for an IC manufacturer is the total
process yield, that is, the fraction of individual dice that survives all stages of production and testing to
emerge as salable packaged chips. Total yield may be 80% or higher for relatively simple circuits
produced with mature technologies, but figures below 10% are not uncommon for large, highly
integrated products in the early stages of production.
The wafer fabrication stage dominates the economics of IC production, and it is here that
semiconductor manufacturers concentrate their research and development efforts. Wafer fabrication
requires an enormous investment in plant and equipment. Because capital costs are high and variable
processing costs are relatively low, high utilization of wafer fabrication equipment is a generally
accepted goal in the semiconductor industry. Most competitive wafer fabs are operated on a two shift
(day and night) basis for 7 days per week, but the amount of time spent to actually process wafers is
limited by several factors, such as preventive maintenance, setup, absence of qualified operators, end of
shift effects, and frequent episodes of unscheduled downtime.
Some of this unscheduled downtime is due to the literal failure of equipment, but "process tuning"
is often a more important downtime category. If a manufacturer reduces any of these sources of
equipment unavailability, or the time required for actually processing wafers on any given piece of
equipment, a higher service or throughput rate, and hence, a lower unit cost can be achieved, provided
that process yields are not adversely affected.
A piece of equipment is idle if it is available for processing but is starved for work. Equivalently, the
equipment is idle if it is neither processing wafers nor rendered unavailable for one of the reasons
named above. The idleness rate for a piece of equipment is defined as the overall fraction of working
hours that it spends in the idle condition. The conventional wisdom among semiconductor
manufacturers is that the idleness rate for critical fabrication equipment should be no larger than 10%.
Given this, it will come as no surprise that wafers spend most of their time waiting rather than being
processed.
To set terminology, we define the fab cycle time for a lot of wafers as the total number of working
hours that elapse between its entry into the clean room and its exit. This same quantity will occasionally
be called the manufacturing cycle time or manufacturing interval for wafer fabrication. The toolset cycle
time is defined as the total working hours that elapse between the entry of a lot into the queue of a
particular toolset and its exit after being processed by that toolset.
To better understand the magnitude of the queueing effects in IC manufacturing, consider a wafer
fab dedicated to production of a single, reasonably complicated product such as a laptop processor.
Production of these processors involves a total of perhaps 600 distinct fabrication steps or operations. If
we add the total times required to complete all these steps, assuming tools and people are always
available for production, the total time might come to about one week. However, realistically it would
typically take between 5 to 10 weeks to fabricate the same product. In the semiconductor industry it is
common to describe this state of affairs by saying that the actual‐to‐theoretical ratio is between 5 and
10, or that the manufacturing interval is 5‐ 10 times the theoretical.
This is widely recognized as a major problem for semiconductor manufacturers. In the case of
customized products, the nature of the problem is obvious, since the order lead‐time imposed on
customers must be at least as large as the total manufacturing interval. On the other hand, standardized
products can be made to stock, but here again, long manufacturing intervals cause trouble because
production must be based on forecasts of market demand many months in the future, and major
demand shifts are commonplace.
Moreover, product life cycles are short in the semiconductor industry, so the risk of obsolescence
for finished goods inventory is always present. Finally, there is an established negative correlation
between manufacturing interval and yield in wafer fabrication, which provides another strong
motivation for reduction of throughput times.
Thus, it is essential that the designer of a wafer fabrication line has a means to predict key
performance measures, including average cycle time, given only processing system characteristics that
are known or can reasonably be estimated before the system goes into operation.
There exist two classes of wafer fabrication facilities in the industry. Research and Development
(R&D) fabs, are dedicated to development of new products and processes and are usually a smaller
version of a High Volume Manufacturing (HVM) fab. HVM fabs are designed for production of salable
chips but the equipment, operating procedures and process flows are essentially the same as an
equivalent R&D fab but in a larger scale.
In this research, we mainly focus on HVM factories since the queueing systems formed in those
fabs are far more complicated than in an R&D fab due to increased number of tools within each toolset
as well as higher volume of lots flowing through the fab. Effective queuing models for HVM fabs can be
easily modified for an R&D fab too. In the following sections we will explain in more details the process
of wafer fabrication in an HVM fab along with a brief introduction to major fabrication equipment,
processing steps and flow of the lots throughout the fab.
Wafer Processing
As stated earlier, wafer fabrication is done in a clean room, which is typically divided into U‐shaped
bays. A bay generally contains a major piece of equipment or a whole toolset on which one or multiple
operations are performed, plus ancillary equipment or facilities involved in closely related operations.
Operators process lots on the inside of the U, and most equipment is positioned so that maintenance
can be performed on the outside of the U. This area is called chase, which is usually outside of the clean
room. Companies have recently come up with newer layout plans such as ballroom setting in which
there are no bay/chase divisions and all equipment are put next to each other in a very large ballroom
shaped space.
Each lot entering the clean room has an associated process flow, often called a recipe, which
consists of precisely specified operations executed in a prescribed sequence on designated toolsets. If all
goes well, this exact sequence of operations is performed, but sometimes inspections reveal that an
operation was not executed to specification, in which case, some or all of the wafers in the lot are either
scrapped or reworked.
Integrated circuit fabrication involves the creation of multiple layers on a silicon wafer hence using
the terminology wafer for the silicon plates. The operations involved in the creation of each successive
layer are essentially the same, so lots can, and typically do, return repeatedly to the same toolsets for
completion of the subsequent layers.
Equipment Categories and Major Process Steps
As noted before, each piece of equipment can perform one or multiple operations on wafers. On
the other hand a fab can possess multiple pieces of the same equipment, commonly referred to as tools.
The terminology, toolset is referred to a fleet of similar tools that are capable of performing the same
operations. These toolsets usually belong to one of the five major categories of operations involved in
creating the successive layers on silicon wafers. Figure 1 shows a simplified graph of these steps and
changes that they make on the wafer. For more detailed description of each operation please see [2]
and [3].
1. Deposition: A thin layer of material is deposited on the surface of the wafer to form a layer of the
integrated circuit. The wafers must be cleaned within a specified time before the operation is
performed, to avoid particle contamination. Different deposition technologies include oxidation,
chemical vapor deposition, spin on glass and physical vapor deposition (sputtering).
2. Lithography: Also called photolithography, masking or patterning. The wafer is coated with a light‐
sensitive material called photoresist, which is then exposed to ultraviolet light through a mask (reticle)
that contains a pattern reflecting the intended geometry of the circuit. The exposure step, generally
referred to as photo‐exposure, is the most complex and delicate operation in wafer fabrication, and it
involves the most expensive pieces of equipment in the fabs. The importance comes from the fact that
the whole geometry of device is being mapped in this step. Even a minor distortion or misalignment
with previous layers can lead to malfunction of the transistors and thus result in a defective device.
Many factors play role in the lithography step, including quality of the lens that focuses the ultraviolet
light on the mask and positioning and quality of masks. At this stage of wafer fabrication manufacturers
encounter the most common and important type of rework. If inspection shows that the pattern
exposed in the photo‐resist does not meet the specifications (because of mask misalignment, for
example), then the entire photo‐resist layer must be stripped off and the wafer prepared for a repetition
of lithography operation. If the pattern is acceptable, then exposed photoresist is removed by a
developer, some minor additional operations are performed and the wafer goes forward with a
protective covering for hardened, unexposed photoresist covering its surface selectively in a pattern
dictated by the mask.
3. Etching: Circuits are defined by etching away the portion of the deposited layer that is not
protected by photoresist. There are two etching technologies used in the fabs; wet etching and plasma
etching, also known as dry etching. The latter (more precise) technology is more heavily used, but wet
etching is more complicated and needs more attention because of the possible inaccuracies that can
happen.
4. Ion Implantation: In order to change the electrical properties of the surface not protected by
photoresist, the wafer is exposed to accelerated ions which are implanted to a predetermined depth
and concentration. Boron, phosphorus, and arsenic ions are most frequently implanted because even a
small number will dramatically alter the electrical properties of the underlying silicon.
5. Photoresist Strip: Finally, the pattern of photoresist that remains on the wafer after lithography is
removed (stripped) using a process similar to etching. Equipment which carries out this operation is
commonly referred to as planar tools.
In addition to these five types of operations, many cleaning, measurement and inspection
operations are performed throughout the fab. The cleaning operations prevent contamination of the
wafer, and the inspection and measurement operations are performed to identify defective wafers,
which are then scrapped or reworked. Different kinds of integrated circuits require different processing
steps during fabrication, and thus, have different process flows through the fab. Production facilities (as
opposed to R&D labs) usually make more than one type of product. The processing technology changes
quite frequently as well and therefore there is substantial diversity in product routing when the
operations of any given fab are examined over a period of several years.
As stated earlier, an individual lot of wafers may visit one or more pieces of equipment repeatedly
during the course of its fabrication. This phenomenon is referred to as re‐entrant processes. In
particular, it is usual to use the same photo‐expose machine (stepper or aligner) in the creation of each
layer. Ion implanters and inspection stations may also be visited repeatedly. Deposition equipment and
plasma etchers are often dedicated to a single operation in a single process; it is therefore common for
them to be visited only once during the entire fabrication sequence.
Flow Control in HVM Fabs
Most of HVM fabs are designed as classic job shop operations, in which lots with diverse characters
are routed through a collection of general purpose work centers for execution of prescribed operations.
The performance of a job shop is affected by the flow control mechanisms employed, including the input
control, routing and sequencing rules built into the shop's operating system. Some of these routing and
sequencing rules include lot to lens dedication for lithography operations and ion‐implant run rules that
are described here briefly.
Lot to lens dedication states that if lithography process is done on a particular lot by a certain tool
that lot has to go through the exact same tool for its subsequent lithography operations on higher
layers. The reason is that lenses which are used in lithography to focus ultraviolet light on the mask and
subsequently onto the wafer through the patterns on the mask cannot be built quite identically in
microscopic scales. The difference, although minute, can cause misalignment between successive layers.
Therefore, when processing the critical layers (not all, but those layers which are critical to quality of the
device) on a wafer, each lot has to go through the exact same piece of equipment that has laid the first
layer onto the wafers in that lot.
Ion‐implant run rules call for certain sequencing of lots that go through an implanter machine. This
sequence depends on the material that is being implanted onto the wafer. For example, there is a limit
on the number of lots implanted by fluoride in a sequence or cascade. After the limit is reached, the
recipe should be changed on the tool and lots with a different implant ion should be started on that
machine.
Another factor that causes the flow of lots through the fab to be stochastic is the processing of
monitor lots or engineering lots. Due to high quality standards in most chip manufacturer companies,
after any maintenance of the tools or any modification in recipe a number of monitor lots or engineering
lots are run through the tool and possibly the tools that conduct subsequent operation. These lots are
not considered as production lots and will be scrapped eventually. Their role is to make sure that the
tools are qualified to manufacture within the required specifications after maintenance or recipe
change.
Also for operations that require high accuracy, such as lithography, after a number of lots have
been processed, one look‐ahead (or send‐ahead) lot is processed and tested before the rest of the lots
in the queue are processed on that tool. This action is taken to make sure that the tool is still
maintaining its qualifications to process the lots.
Hot lots belong to another category of productions lots. These are lots with higher priority than
any other lot waiting in the queue. Their priority may be caused by many different reasons, including
split lots that are going to be merged with the mother lot in a few steps from the existing operation. This
type of lots is especially common in R&D fabs where many experiments are performed on the process.
Processing of lots in a queue at a certain toolset is not always done on a First‐In‐First‐Out (FIFO)
basis. However, within a queue of lots from the same product type and same operation FIFO is
conducted as the general policy.
1‐ J. G. Shanthikumar, S. Ding, M. Zhang, "Queueing Theory for Semiconductor Manufacturing
Systems: A Survey and Open Problems", IEEE Trans. Automat Sci and Eng., to appear.
2‐ H. Chen, J. M. Harrison, A. Mandelbaum, A. V. Ackere, and L. M. Wein, "Empirical evaluation of a
queueing network model for semiconductor wafer fabrication," Oper. Res., vol. 36, no. 2, pp.
202‐215, 1988.
3‐ http://www.eas.asu.edu/~aar/research/intel/papers/fabspec.html