FabTime Cycle Time Management
Newsletter Abstracts - Volume 3 (10
Issues)
Our main article this month is about a
new performance metric we are proposing.
After discussing what attributes we believe
should be found in metrics for daily
production meetings, we propose Quality
Moves. Quality Moves measure, on a shift
basis, the best performance that can be
achieved given the fab’s WIP profile
and resource availability.
In this
month’s subscriber discussion forum
we have many responses to last
month’s main article about the impact
of staffing (particularly operator delays)
on cycle time. Most of the respondents
agreed that operator delays do have an
impact on fab cycle times, at least some of
the time. We also have new topics raised by
subscribers related to performing tool
qualification on the bottleneck and
estimating the impact of hand-carry lots on
other lots.
(
Return to newsletter subscription
page to subscribe and receive the current issue free each month)
We are interested in estimating the
impact of staffing on cycle time. In this
article, rather than tackle this issue in
detail, we focused on one particular aspect
- forced idle time on tools due to operator
delays. To look at this visually, we built
a very simple simulation model to study the
issue. We found that even in models with
only 3 tools, and light operator loading
(50% busy), operator delays may increase
cycle time significantly.
This month we
also have subscriber discussion on capacity
planning using simulation, as well as using
fab-level metrics for understanding
variability. We also present the results
from last month’s survey question
about the number of certifications per
operator that people have in their
factories.
(
Return to newsletter subscription
page to subscribe and receive the current issue free each month)
When a batch tool (e.g. diffusion
furnace) is available and there are one or
more lots ready to be processed, the
operator must decide whether to start the
batch immediately, or wait for more lots.
When a full batch of some recipe is
available, the decision to start that batch
is fairly easy. However, when less than a
full batch of lots is available, the
decision becomes more complex. On average,
it is usually better for cycle time to
start the batch immediately than to wait to
form a full batch. However, despite this
general rule, there are sometimes specific
cases where it makes more sense to wait for
the next lot before starting the batch. In
this article, we propose a simple rule for
deciding when to wait for the next lot, and
when to just start the batch.
In this
month’s subscriber discussion forum
we have continuing discussion on recipe
management, batch size decision rules, and
operator cross-training.
(
Return to newsletter subscription
page to subscribe and receive the current issue free each month)
In this month’s main article, we
have chosen to briefly review the topics
described in the FabTime newsletter issues
to date (both the main articles and the
subscriber discussion topics). The primary
reason for this is that we have many new
subscribers, who may not be aware of the
topics already covered. Even for long-time
subscribers, job descriptions and market
conditions change regularly. A topic that
wasn’t of interest to you when it
first came out may be more relevant now.
In
this month’s Recommendations and
Resources section, we review the many
resources available on FabTime’s
website (papers, tutorials, book reviews,
software demos, etc).
(
Return to newsletter subscription
page to subscribe and receive the current issue free each month)
In this month’s main article, we
propose three distinct cycle time
management styles, and describe how each
can be used to improve cycle time. We have
named these three styles: The Traffic Cop;
The Shepherd; and The Relay Coach. These
are management styles we have observed in
real fabs, although the names and
descriptions are our own. Each style is
suited to a particular cycle time focus.
Traffic Cops control starts and WIP flow
for production lots. Shepherds prevent
engineering lots from disappearing onto
shelves and hiding in corners. Relay
Coaches ensure that critical hot lots are
handed smoothly from one operation to the
next. Graphical examples, using charts from
FabTime’s software, can be found on
our website, at www.FabTime.com/ctmstyles.shtml.
Discussion topics in this issue include: responses on wafer starts methodologies, treating scrap in product costing, and ramp planning; a reference to a conference presentation about operator modeling; a question about how much is too much in reference to operator cross-training; a question about how people handle recipe management; and a request for benchmarks for gallium arsenside fab cycle times.
(
Return to newsletter subscription
page to subscribe and receive the current issue free each month)
Our main article this month is about
quantifying the bottom-line benefits of
cycle time improvement. We discussed one
particular benefit in a previous newsletter
issue. In this new article, we provide a
more comprehensive framework for linking
cycle time management to financial returns.
An Excel spreadsheet tool for what-if
analysis is provided on FabTime’s
website (here). There’s
both money to be saved and additional
revenue to be earned through cycle time
improvement. Under the assumptions in our
default example, the total annual benefit
of cycle time improvement could be more
than half a million dollars.
Discussion topics in this issue include: a request for information on wafer start methodologies; a request for research on staffing models; a request for literature on ramp models; a question about how companies treat cost of scrap; and a question about calculating mean time between assists.
(
Return to newsletter subscription
page to subscribe and receive the current issue free each month)
This month’s main article, Cycle
Time and the Core Conflict, is a guest
article, written by Dan Siems, of Philips.
Dan was recently appointed World Wide Wafer
Fab Cycle Time Manager for Philips
Semiconductors. This article represents
Dan’s thoughts on a core conflict
that often exists in managing wafer fabs -
trying to get lots out quickly, but having
to frequently stop the lots for quality
checks. Dan proposes the elements that he
believes must exist to weaken this
conflict, and maintain good cycle times
over the long term.
This month in the
subscriber discussion forum we have several
responses to last month’s main topic
of equipment dedication. Other topics
discussed in this issue include lot size
change, foundry performance data, and the
interaction of AMHS control and
dispatching.
(
Return to newsletter subscription
page to subscribe and receive the current issue free each month)
We talked back in Issue 1.8 about the
fact that single path tools tend to drive
up cycle times. The question is, how much
does tool dedication inflate cycle times?
The are sometimes important reasons to have
dedicated tools. What’s needed is a
way to explore trade-offs. In this article,
we present an approximation for queue time
as a function of number of machines in a
tool group. This approximation clearly
shows that queue time decreases as the
number of tools in the group increases (for
the same total traffic intensity of the
tool group).
Discussion topics in this issue include: a question about segregating downtime and idle time into "good" and "bad" for PEE calculations; a request for opinions on how to model single wafer lots; a question about the details of generating characteristic curves; a request for foundry performance data benchmarks; and several detailed responses to the Volume 3, Number 2 hot lot article.
(
Return to newsletter subscription
page to subscribe and receive the current issue free each month)
The article is drawn from a presentation
that Frank Chance made at Arizona State in
January. We present a formula for
estimating the average cycle time of lots
through a tool that processes lots with
different priorities (regular lots and hot
lots). We provide a numerical example that
shows how the cycle time of the regular
lots increases as the percentage of hot
lots is increased, and discuss implications
for managing hot lots in a wafer fab. An
example can be seen here.
Discussion topics in this issue include: a response to the question about performance measures regarding human resource to activity relationships; a request for cycle time reduction case studies; and an observation on production equipment efficiency (PEE) as a measure of tool variability.
(
Return to newsletter subscription
page to subscribe and receive the current issue free each month)
In a wafer fab, cycle time tends to
increase with increasing equipment loading
(with some exceptions for batch tools). In
large part to combat high cycle times, fabs
typically plan for some amount of idle time
on most tool groups. OEE, in its
traditional definition, is contradictory to
such planned idle time, since all standby
time (including planned idle time) drives
down OEE values. This puts fab personnel in
a tight spot when they are pushed to
simultaneously increase OEE values and
decrease cycle times. Production Equipment
Efficiency (PEE) is a related metric that
calculates equipment productivity only
during the time that product is available
at the tool. Improving PEE, therefore, is
not in conflict with reducing cycle times.
PEE only penalizes tools for standby time
during which lots are waiting (e.g. time
when WIP is present, but there is no
operator to load the tool). For
bottlenecks, there will likely be very
little time during which no WIP is waiting.
Therefore, for bottlenecks, PEE and OEE
will yield similar values. For
non-constraint tools, however, PEE values
will usually be higher than OEE values. The
important thing is that increasing PEE
values will not conflict with reducing
cycle times. For fabs trying to improve or
maintain cycle times, using PEE instead of
OEE may be more effective, at least for
non-constraint tools.
Discussion topics in this issue include: a request for information on measuring shift performance; a question about performance measures regarding human resource to activity relationships; and a question about model accuracy relative to actual performance.
(
Return to newsletter subscription
page to subscribe and receive the current issue free each month)
|
