Вы находитесь на странице: 1из 4

Impacts of Small Lot Manufacturing on AMHS FD-O-045

Hiroshi Kondo
Asyst Technologies Japan, Inc.
Ise-city, Mie prif., Japan,
HKondo@asyst.com

section, what causes decline of tool utilization is not long


Abstract – Demands to reduce semiconductor fab lot cycle delivery time itself but Carrier Exchange Time (CET).
time are increasing more and more. Small lot manufacturing
(SLM) is proposed as a promising way to reduce cycle times
considerably. However there are detractors against realizing
30

frequency of carrier transport request


effective SLM. Some of them are related to Automated
Manufacturing Handling Systen (AMHS). This paper 25
describes these detractors and discuss potential solutions for 20
them. We conclude that AMHS solutions for SLM should

(relative)
stress more transport capacity and shorter Carrier Exchange 15

Time (CET) rather than shorter delivery time. We also 10


conclude that mathematical models are more intuitive and
useful than simulation models for such studies. 5

0
0 5 10 15 20 25
INTRODUCTION lot size (# of wafers/lot)

Demands to reduce semiconductor fab lot cycle time are


increasing more and more because lifecycles of products that
use semiconductors become shorter and shorter. Small lot Graph 1. Frequency of carrier transport requests v.s. lot size.
manufacturing (SLM) is proposed as a promising way to
reduce fab lot cycle times considerably [2]. However three
detractors, 1) First Wafer Delay (FWD), 2) transport capacity
limitation and 3) decline of tool utilization due to carrier
exchange time (CET), make it difficult to install smaller lot
manufacturing than it is done in current fabs. Among them, 2)
and 3) are issues related to AMHS. So, we discuss them in
this paper. The principle for the study we adopt is to use
mathematical models rather than simulation models. The
reason is that if AMHS layout candidates are narrowed down,
simulation studies can perform effective works but in order to
solve more general questions such as what kind of
configuration AMHS should have in SLM environment,
Graph 2. Sample of components of lot cycle time in a fab.
mathematical models are more intuitive than simulation
models because we cannot narrow down AMHS layout
candidates. CARRIER EXCHANGE TIME

Next, we will discuss about decline of tool utilization due to


CET. A mechanism of decline of tool utilization due to CET
TRANSPORT CAPACITY LIMITATION
was investigated in [1]. According to it, the causes of decline
First, we discuss item 2), issue of transport capacity. If we of tool utilization are long CET and current carrier operation
perform smaller lot manufacturing while keeping fab scenario. Therefore the solutions are shorter CET and a new
throughput constant, volume of carrier transport requests will carrier operation scenario. Some ways to reduce CET are
increase inversely proportional to the lot size (Graph 1). So, described in [1]. New two carrier operation scenarios are
SLM requires AMHS to have larger transport capacity. On the proposed in [2].
other hand, in current typical fabs, delivery times account for
only small percentage of total lot cycle times (Graph 2). So,
Current carrier operation scenario is as follows; 1) a carrier
reduction of delivery time itself does not contribute to
is set down on the load port. 2) It waits for wafers to be
reduction of lot cycle time so much. However, Enhancement
brought into the tool inside. 3) Wafers are brought into the
of transport capacity enable smaller lot size and therefore it
tool/ Wafers are processed/ Wafers are brought back into the
may lead to shorter lot cycle time, even if it may cause longer
carrier. 4) The carrier is picked up from the load port.
delivery time. Therefore, AMHS solutions for SLM will stress
transport capacity rather than delivery time. One may think
that longer delivery time yields decline of tool utilization, and
therefore decline of fab throughput. But, as described next
New carrier operation scenario 1 is as follows; 1) a carrier where, C is the tool capacity loss rate and p (t ) is the
is set down on the load port. 2) It waits for wafers to be probability density function of CET. τ is defined by
brought into the tool inside. 3) Wafers are brought into the
following equation for each carrier operation scenario.
tool. 4) The empty carrier is picked up so that the load port
becomes available for another carrier with wafers. 5) Wafers
are processed. 6) The empty carrier is set down on a load port
τ = tint [S (LP − 1) − N ] …..(2) for current carrier operation
and waits for processed wafers to be brought back into it. 7)
Wafers are brought bacj into the carrier. 8) The carrier is scenario.
picked up from the load port.

LP
New carrier operation scenario 2 is as follows; 1) a carrier
τ = tint × S × …..(3) for scenario 1.
2
is set down on the load port. 2) It waits for wafers to be
brought into the tool inside. 3) Wafers are brought into the
tool. 4) The empty carrier waits for processed wafers that τ = tint [S (LP − 2) − M ] …..(4) for scenario 2.
comes from ANOTHER carrier to be brought into it. 5)
Wafers are brought into the carrier. 6) The carrier is picked up
from the load port. This scenario no longer keeps “Carrier
Integrity”. Where, tint is process interval of a wafer and next wafer, S is
lot size, LP is number of load ports of the tool, N is number
of wafers that can be inside the tool and M is residue modulo
Scenario 1 and 2 show that shorter CET is necessary to realize
them. So, shorter CET is a major enabler for SLM. What
S of N .
extent AMHS should shorten CET? But since delivery time
vary statistically, actual CET has its distribution like as shown
Graph 3. So. it is hard to answer the above question. DERIVATION OF THE APPROXIMATION FORMULA

This section shows outline of derivation of approximation


0.35 formula (1) only for current carrier operation scenario.
0.30
Derivations for the other scenarios are similar as this. First we
consider the case that CET can has only two values; one
0.25
(denoted by “CET1”) is less than τ and the other (denoted by
Probability density

0.20 “CET2”) is greater than τ . Then, to simplify the explanation,


0.15 we consider the case of LP =2 and N =0 only. Let
0.10 te represent time to process one lot. Since lot size is S and
Tail
0.05
now N =0, te = tint S . Meanwhile since LP =2 and N =0,
τ = tint [S (LP − 1) − N ] = tint S = te .
0.00
CET

Graph 3. Example of distribution of CET.


LP1 te CET1 te
LP2 CET1 te
Considering this fact, we need to change the question like this.
“What distribution of CET is required in order to ensure that
LP1 te CET2 te
tool capacity loss rate is less than x%?” The term, “ tool LP2 CET1 te ⊿
capacity loss rate” means ‘one minus upper limit of tool
utilization’. It might be common to answer this question to LP1 te CET1 te
utilize simulation studies. But for the reson mentioned before, LP2 CET2 te
we think mathematical models are more intuitive and useful
than simulation models. So, in this paper, we present LP1 te CET2 te
approximation formula (1) to calculate tool capacity loss rate, LP2 CET2 te

S × t int …(1)
C = 1− Fig. 1. Gantt charts in the case that CET has possible two values and
∞ 1
S × t int + ∫ tp (t + τ )dt ⋅ ∞ the tool has two load ports.
1 + (LP − 1)∫ p(t + τ )dt
0
0
Fig. 1 shows parts of Gantt charts that represent the behaviors
of load ports for this case. By looking at them, we can easily
find that sometimes process finish time of a lot is the same as te
process start time of next lot and sometimes not. In case that C =1− ….(6).
p
process finish time of a lot is not the same as process start te + Δ
time of next lot, we can find that the difference of the two 1+ p
time points, Δ is CET2 − te by looking at the chart. However
by considering cases that LP > 2 , we find that it is better to Then, we extend (6) to the cases for any LP and N . Here,
write Δ = CET2 − τ . We express the case that a process we omit the derivation process, but by considering it in similar
finish time is the same as next process start time in “0” and way, finally we have
the case that it is not the same in “ Δ ”. As founded in Fig. 2,
if process fnish time point with “0” is followed by “CET1”,
then next process finish time point is also in case of “0”. S × tint
C = 1− ….(7),
p
S × tint + Δ
1 + (LP − 1) p
LP1 te CET1 te
LP2 CET1 te
where Δ = CET 2 − τ and τ is defined by (2).
0 0

These above descriptions are considerations for the case that


Fig. 2. “0” time point and CET1 yield another “0” time point. CET has only two values. From here we try to construct
approximation formula for tool capacity loss rate for the cases
We can express this relationship by “0 and CET1 Æ 0”. By
considering in similar way, we can get following results. “0 that CET has a distribution of its probability density p (t ) .
and CET2 Æ Δ ”, “ Δ and CET1 Æ 0” and “ Δ and CET2 Æ We will do this by rewrite p and pΔ in equation (7) by
0”. Let p represent the probability that CET has value CET2 using p (t ) . First, because p in (7) can be considered as the
(> τ ). Then we can consider above behaviors of Gantt chart probability that CET is greater than τ , then we can replace
as a Markov chain and express this Markov chain with Fig. 3. p with

∞ ∞

∫τ p(t )dt =∫ p(t + τ )dt ….(8)
0
1 p

Then, because pΔ in (7) can be considered as the average of


0 min( CET − τ ,0) , then we can replace p with

∞ ∞
Fig. 3. Graph of Markov chain for the case shown in Fig. 1.
∫τ (t − τ ) p(t )dt =∫0 tp(t + τ )dt ….(9)
We can have the probability that Δ occurs, P (Δ ) , by
calculating steady state probability of this Markov chain. The ∞
calculation yields By replacing p in (7) with ∫ p(t + τ )dt
0
and pΔ in (7)

p
with ∫ tp(t + τ )dt , we can have approximation formula (1).
0
P(Δ ) = ….(5).
1+ p
ACCURACY OF THE APPROXIMATION FORMULA
From (5), tool capacity loss rate C is given by
In order to verify the accuracy of approximation formula (1),
we compared the results of the formula with that of simulation
studies in some cases. In comparison, we set LP = 2 and
N = 0 . We used exponential and uniform distributions as
distribution of CET in case that CET ≥ τ , i. e. p (t ) , t ≥ τ .
Note that p (t ) , t < τ does not impact tool capacity loss. So, CONCLUSION
we need not consider shape of CET distribution in t < τ . The We have been studying impacts of SLM on AMHS. We found
results are shown in table 1. Left 2 columns of the table show and discussed two detractors against realizing effective SLM
the conditions and rest of columns show the results. Column that are related to AMHS; transport capacity limitation and
“distribution” indicates shape of p (t ) , t ≥ τ . load port neck. We conclude that AMHS solutions for SLM
t should stress more transport capacity and shorter CET rather
1 −m
“exponentiaonl” means p (t ) = e , t ≥ τ , where m is than shorter delivery time. We also conclude that
m mathematical models are more intuitive and useful than
its average. “uniform 1” means p (t ) = 1 , 1 ≤ t ≤ 1 + a and
simulation models for studying suitable AMHS for SLM
because we cannot narrow down AMHS layout candidates. In
p (t ) = 0 , 1 + a < t . “uniform 2” means p (t ) = a , order to estimate effect of tail of CET distributions on tool
1 ≤ t ≤ 2 and p (t ) = 0 , 2 < t . Column “probability of utilization, we propose an approximation formula, which is an
example of a mathematical model. These considerations will
CET>tau” indicates values of expression (8). be guidelines for designing AMHS that contribute to effective
SLM.
tool capacity loss rate
distribution
probability of simulation
calculation difference REFERENCES
CET>tau result
exponential 5% 0.8% 1.6% -0.8%
exponential 10% 2.2% 3.8% -1.6% [1] H. Kondo, "Requirements for AMHS in 450 mm era."
exponential 15% 4.0% 6.4% -2.4% Proceedings in International Symposium on
exponential 20% 6.0% 9.4% -3.4% Semiconductor Manufacturing – ISSM, Sept. 2006,
uniform 1 5% 0.1% 0.1% 0.0%
uniform 1 10% 0.5% 0.5% 0.0% pp.189-192.
uniform 1 15% 1.0% 1.0% 0.0%
uniform 1 20% 1.8% 1.6% 0.2% [2] O. Zimmerhackl, J. Rothe, K. Schmidt, L. Marshall and A.
uniform 2 5% 1.4% 2.3% -0.9% Honold, "The Effecrs of Small Lot Manufacturing on
uniform 2 10% 3.0% 4.3% -1.3%
uniform 2 15% 5.0% 6.1% -1.1% AMHS Operation and Equipment Front-End Design"
uniform 2 20% 7.0% 7.7% -0.7% Proceedings in International Symposium on
Semiconductor Manufacturing – ISSM, Oct. 2007, pp.
Table 1. Comparison between results of calculation and simulation 185-188
for tool capacity loss rate.
[3] J. Pettinato and D. Pillai, “Technology Decisions to
From the results we conclude that approximation formula (1) Minimize 450mm Wafer Size Transition Risk”
is accurate enough to use for AMHS specification design. Proceedings in International Symposium on
This approximation formula can reply answers quickly for Semiconductor Manufacturing – ISSM, Sept. 2004.
various combinations of values of condition parameters and
[4] ISMI Factory Architecture Focus Team "Unified
reveal impact of change in condition parameters on tool
450mm/300mmPrime Factory Guidelines Workshop –
capacity loss more clearly than simulation.
ISMI Next Generation Factory Vision." – July 2007

AUTHOR BIOGRAPHY
NEAR TOOL BUFFERING
Hiroshi Kondo joined NEC Corporation in 1982 and had
SLM requires shorter CET. One of means to realize it is to been engaged in development of Factory Automation System
utilize near tool buffering (For example, see [3] and [4]). In (especially real-time Dispatching system) for semiconductor
order to determine optimal capacity of near tool buffers, we Fabs. In 1998 he joined Selete (Semiconductor Leading Edge
can utilize similar consideration as what described in section Technologies, Inc.) and developed AMHS simulation model
“CARRIER EXCHANGE TIME” and “DERIVATION OF and common communication specification between AMHS
THE APPROXIMATION FORMULA”. The basic idea of and MES. In 2002 he joined Asyst Shinko, Inc. (now, Asyst
calculation of optimal buffer capacity is to consider buffer Technologies Japan, Inc.) and now is in charge of AMHS
capacity instead of number of load port LP in similar simulation researches.
approximation formula as (1).

Another interesting issue related to near tool buffering is to


develop the method to calculate direct delivery rate, at least
approximately. Direct delivery rate means the rate AMHS
delivers lots directly from previous processed tools to tools for
current process steps without passing through buffers and
stockers. We don’t find the solution for it yet. This will be our
future work.

Вам также может понравиться