Table of Contents
Not with a bang, but with a whimper:
Understanding delays in semiconductor supply chain dynamics
Paper for
International System Dynamics C onference 2003
May 2003
Jan Jaap Bezemer Henk Akkermans
Eindhoven University of Technology Eindhoven University of Technology
Technology Management Technology Management
PO Box 513, 5600 MB Eindhoven PO Box 513, 5600 MB Eindhoven
Netherlands Netherlands
E-mail: j,j.bezemer@ student.tue.nl (also Minase B.V., Netherlands)
Tel: +31-13-5443468
E-mail: h.aakkermans@tm.tue.nl
ABSTRACT
The semiconductor industry is characterized by high volatility: rapid increases in market demand
are followed by sharp downturns. Therefore, one would expect its supply chains to be very fast in
adjusting to changes in demand. However, empirical data from one leading semiconductor firm
suggest that delays in adjusting to the latest downtum of the market in 2001 have been
considerable. For instance, inventory levels have taken two years to come back in line. Generally,
these delays and the dynamics that are causing them are not well understood within the industry.
This paper presents research that explains these delays by means of a system dynamics
simulation model that captures the overall supply chain structure, the generic decision-making
processes and the associated supply chain dynamics typical for this industry. The model is based
upon pre-existing and well-tested generic supply chain models from the literature. It has been
tailored and validated with representatives from a major European IC manufacturer. Its dynamic
performance has been calibrated using four years of data on key performance aspects such as
inventory levels, cycle times, demand flexibility and delivery quality.
With this model, several SCM policies are explored that are effective in improving both sales
and supply chain performance, such as more aggressive capacity build-up, lower capacity
utilization targets and higher end product buffer stocks.
WORK IN PROGRESS VERSION
Not with a bang, but with a whimper:
Understanding delays in semiconductor supply chain dynamics
Jan Jaap Bezemer and Henk Akkermans
Eindhoven University of Technology,
Technology Management
Eindhoven, The Netherlands
This is the way the world ends
This is the way the world ends
This is the way the world ends
Not with a bang but with a whimper.
[From T.S. Elliot, The Hollow Men]
Introduction
The semiconductor industry is characterised by high volatility: long-term market growth rates
have been impressive, but demand can go up and down substantially from one year to another.
For instance, from 1998 to 2000, semiconductor industry sales rose more than 35%, only to go
down almost just as much from 2000 to 2002 (WSTS 2003). This is a “high-clockspeed” industry
(Fine 1998), where rates of innovation are high, product life cycles are short and new products are
pushed out into the market as fast as possible.
In such an industry, one would expect the supply chains that are responsible for the production
and distribution of IC’s to be very fast in adjusting to variations in demand. One would expect
major shocks to the supply chain to be absorbed “with a bang, not with a whimper’. Also, one
would expect delays in responding to changes to be under close scrutiny of company management
and, overall, fairly well understood. Interestingly, our dealings with managers in a variety of
firms in this industry (Akkermans et al. 1999, Akkermans 2001, Akkermans and van der Horst
2002, Akkermans et al. 2003) suggest that neither expectation is correct. Delays in responding to
demand variability can be very substantial, in some cases several years. Here, the response to a
shock to the supply chain indeed ends “not with a bang, but with a whimper”. Also, the extent of
these delays is usually not well recognised. Moreover, what is driving them is generally not well
understood.
This then becomes the goal of this paper: to shed more light on delays in adjusting supply
chain performance to variations in market demand, on root causes for these delays and on policies
for improving supply chain performance in view of this. For this purpose, we present a generic
system dynamics simulation model that captures the overall supply chain structure, generic
decision-making processes and associated supply chain dynamics typical for the semiconductor
industry. We show how we have validated and calibrated this model with representatives and
quantitative data from one leading European IC manufacturer.
We take advantage of the unprecedented IC market demand peak of 2001 and the subsequent
period of demand stability and use it as a unique real world “test signal” in four years of time
series data on key performance indicators. On the basis of our analysis, we suggest that there are
three different types of delays in response of supply chain performance, which we have labelled
as operational, tactical and strategic. We identify policies to dramatically improve supply chain
performance at each of this levels, based on policy experiments with our generic simulation
model.
SEMICONDUCTOR SUPPLY CHAIN DELAYS - JJ BEZEMER & HA AKKERMANS
2. Literature review
Delays in reacting to variations in market demand for different aspects of supply chain
performance have long been studied. The seminal works here are those of Forrester (1958, 1961),
who investigated the effect of changes in market demand on a four-level supply chain. Forrester
incorporated delays in both the physical structure of his supply chain as in the managers’
decisions and policies goveming inventory adjustment. Forrester found that these delays were
responsible for much of the oscillatory behaviour of many supply chains. He found that the nature
of the managerial policies concerned could modify these delays and the resulting instabilities of
supply chains considerable. His model was later on converted into the well-known “Beer Game”
(Sterman 1989), in which these policies and delays generate well-known supply chain dynamics
such as the Bull-whip effect (Lee et al. 1997a,1997b, Fransoo and Wouters 2000).
In the field of system dynamics, this work has been followed up by several other authors over
the years, such as Mass (1975) who investigated the interactions between inventory-production
policies and Morecroft (1983) who looked at the impact of implementing Material Requirements
Planning (MRP) systems on supply chain performance.
In the world of supply chain management, these supply chain dynamics insights were taken up
in Europe from the late nineteen seventies onwards (van Aken 1978, Hoekstra and Romme 1991,
Evans et al. 1993). In the last decade however, in US mainstream OR research interest in the
impact of delays on supply chain performance has risen considerably. There is Lee et al.’s
(1997a, 1997b) publication on the bull-whip effect of upstream demand amplification in the
supply chain. Baganha and Cohen (1998) look at the stabilising effect of inventory on supply
chain performance. Chen (1999) considers at the impact of information delays on decentralised
supply chains, Cachon (1999) investigated the relation between demand variability and ordering
policies, Chen et al. (2000) explore the impact of forecasting, lead times and information sharing
on the bullwhip effect in a simple supply chain.
Recently, there is a renewed stream of publications from the field of system dynamics proper
on the nature of delays in supply chains. Anderson et al. (2000) look at upstream supply chain
volatility in the machine tool industry. Anderson and Morrice (2000) describe a teaching game in
a service supply chain. Berends and Romme (2001) look at cyclicality in the paper industry and
Akkermans and Vos (2003) at demand amplification in a service supply chain. Goncalvez (2002)
has recently completed ongoing research on the semiconductor supply chain at Intel, to which the
current paper is closely linked.
3. Research method
Selection of research design
The research reported here combines a simulation modelling research design with case study
research. As the topic of this paper is the nature of delays in responding to demand changes in
semiconductor supply chains, and as this topic appears to be not well understood, exploratory,
theory-building research was required. In general, research of this type has been found to be
underrepresented in POM research (Flynn et al. 1990, Meredith 1993, Neely 1993). Case studies
are often employed for exploratory, theory-building research (Yin 1989, Eisenhardt 1989,
Meredith 1993).
But, given the specific nature of the issue at hand, simulation modelling seems equally
relevant as a research approach. A central premise of this paper is that the main reason why
delays in system supply chain response are not well understood is the dynamic nature of the
phenomenon (c.f. Repenning 2002). It is generally accepted in the literature that the ways in
which supply chains deal with variations in market demand is an inherently dynamic process, in
WORK IN PROGRESS VERSION
which numerous variables interconnected in multiple feedback processes play a part.
Unfortunately, human ability to reliably infer the behaviour of even low-order dynamic systems is
exceedingly limited (c.f. Sterman 1989). So, we present in this paper a simulation model whose
variables and linkages in themselves are mostly well-documented in the literature (mainly
Forrester 1968, Sterman 2000, Hopp and Spearman 2000) but whose interactions in this specific
context are not well understood.
Selection of case setting
Our selection criteria for the company to be studied were driven by our research questions (Yin
1989, Eisenhardt 1989, Meredith 1993). This implies that we needed a real world semiconductor
supply chain where:
a) aclear pattem of demand variability had been observed;
b) supply chain performance was significantly affected by this pattern in different respects;
c) sufficient time series and other quantitative data were available to trace this behaviour
over time and calibrate the model to this particular real world setting;
d) knowledgeable company representatives were willing and able to provide input to the
modelling process and validate the overall structure and dynamics of the model.
The dramatic boom-and-bust cycle of 1999-2001, with 2001 being the “worst year ever” in the
semiconductor industry provided an ideal opportunity for the first requirement. As Figure 1
shows, here demand rose spectacularly in 1999, only to drop equally dramatically from late 2000
onwards. Since this drop, the market has been relatively stable. This comes as close to giving a
major “test signal input” to a real world system as one is ever likely to get. The Semiconductor
Company whose empirical data we present here in our case study is where the first author
conducted his Master thesis research (Bezemer 2003). In this company, knowledgeable SCM
professionals and managers were willing to share their insights with him in a series of interviews
and group modek building workshops (Akkermans 1995, Vennix 1996, Akkermans and V ennix
1997). Also, four years of time series data on key aspects of supply chain performance could be
distilled from company records.
250
BS
200
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002
Figure 1: The semiconductor boom-and-bust cycle of 1998-2001, from WSTS (2003)
SEMICONDUCTOR SUPPLY CHAIN DELAYS - JJ BEZEMER & HA AKKERMANS
Research model and research questions
Our conceptual research model is visualised in Figure 2. First of all, this figure illustrates that we
look at different aspects of supply chain performance: quality, cost, time and flexibility. These
four generic categories are normally used to assess supply chain performance in particular and
business process performance in general (c.f. Neely et al. 1995, Neely 1998).
Supply Chain
Performance:
Corrpeny's response
Figure 2: Conceptual research model of drivers of supply chain performance
A central finding from the literature on supply chain management is that the match of demand
versus supply is key in determining supply chain performance. The more demand nears maximal
capacity, the more supply chain performance suffers: lead times become longer, costs in terms of
inventories increase, quality drops and further flexibility is reduced. As demand drops, over time
performance will recover. This leads to our first research question:
Q1: If demand is quickly raised to exceed capacity, and then brought down again,
what are the delays in responding to this variability in demand in terms for supply
chain performance in terms of cost, quality, time and flexibility?
As we will see, different supply chain performance indicators have very different adjustment
delays for their performance. Some take several years to come back in line, others retum quickly
back to normal. Why is that? This leads to our second research question in this paper:
Q2: If there are differences between the adjustment delays for different aspects of
supply chain performance, how can these be explained?
Finally, were are interested in more than just understanding why supply chain performance does
not recover quickly, we want to improve supply chain performance. Hence, our third research
question becomes:
Q3: What managerial policies can reduce the degree in which supply chain
performance suffers as a result of major variations in market demand, and hence
reduce the adjustment delays for each of the key SC performance aspects?
WORK IN PROGRESS VERSION
4. Supply chain model
The supply chain model that was developed for this research was triggered by group model
building workshops with SCM experts from the semiconductor company studied. Nevertheless,
its formal structure is strongly based upon the existing literature. As shown in Figure 3, this
model consists of the following interacting sectors:
1. Goods flow: production and shipment of products
2. Order flow: order acceptance and order fulfilment
3. Customer demand: generation of forecasts and actual order rate
4. Capacity management adjustment of available capacity
The overall structure of interactions between goods flow and order flow is based upon Chapter 18
of Sterman (2000). The throughput and cycle -time calculation formulae come from Hopp and
Spearman’s Factory Physics (2000). The interactions with customer demand and changes in
capacity are drawn from Forrester’ s (1968) market growth model.
industry tend
(3) Customer
Demand A aN
customer orde rate customer satisfaction - Pawar
Hi jlivery delay
\. (2) Order
accepted order rate Fiow
+ oc Se
forecast horizon Zoe Back >
ofr ofr
‘ intake fulfillment.
(4) Capacity rate rate
Management
+ tage!
change in save iat ae delivery
forecast change in delay
capacity va
worden,
Peel shipment rate
Demand
Forecast rin. order
Capacity processing time
Capacity rf
utilization RUP
Nee
shipment rate
+
¢
> WP pe Senice Sioa
production rate shipment rate
awe)
production start rate
(1) Goods
Flow
avarage
cycle-time ‘svarage order
processing time
raw cycle-time
inventory
stacked lead-time
Figure 3: Simplified stocks-and-flow diagram of supply chain model structure
SEMICONDUCTOR SUPPLY CHAIN DELAYS - JJ BEZEMER & HA AKKERMANS
Goods flow
The goods flow is split up into two stages. These stages are separated by the die-bank, which is
the Customer Order Decoupling Point (CODP) (Hoekstra and Romme 1991) most commonly
used in the semiconductor industry. Production control and inventory management decisions
made by the company are endogenous. With respect to this, the production system is organized as
an hybrid push-pull production system (Hodgson and Wang 1991a, 1991b; Spearman and Zazanis
1992). It combines a push system at the upstream stage (WIP) and a pull system at the
downstream stage (Service Stock).
There is only limited capacity available for production. On one hand, this sets the theoretical
maximum production rate. As with a simple production line, it is not possible to make more
products than the bottleneck allows for. On the other hand, cycle-times are dependent on the
realized capacity utilization. It is found that higher levels of capacity utilization result into longer
queue-times (Hopp and Spearman 2000).
A distinction is made between raw cycle-time and queue-time. Raw cycle-time is mainly
influenced by aspects like process-technology and product-design and is therefore considered to
be exogenous. However, queue-time is endogenous. For calculating queue-time, the production
system is modelled according to the “practical worst case”, in which maximum randomness is
assumed (Hopp and Spearman 2000). Infinite WIP is necessary to fully use the bottleneck
capacity. As a result, queue-times explode for high levels of capacity utilization (Figure 4).
Details about the model structure are given in Appendix 1.
Raw Cycle Time (T,): average time it takes a single job to traverse an empty line
Bottleneck Rate (1): rate of the process center having the least long-term capacity
Critical WIP (Wo) is
W, =hy,
Practical Worst Case cycle-time for a given WIP-level, w, is
w-1
CT =Ty —
‘b
Practical Worst Case throughput for a given WIP-level, w is
w
———r
me W, +w-1?
Capacity Utilization (%) Cycle-Time (weeks)
100%
80% f-
60% zai | i
40% ye ye
20% [ wz
0%
———_* 0% 20% 40% 60% 80% 100%
wip Utilization
Available capacity and raw cycle-time are constant
Figure 4: Key "Factory Physics” for cycle-time and WIP levels (Hopp and Spearman 2000)
WORK IN PROGRESS VERSION
Order flow
In contrast to the upstream stage, the downstream stage, i.e. assembly and distribution, operates as
a pull system, with shipments based on current demand signals. In order to prevent the system
from overflowing, order acceptance is limited to the capacity available for production. When
accepted, endogenous order fulfilment determines the adequacy of the inventory available at the
service stock.
In order to account for the processes carried out downstream of the die-bank, an exogenous
minimum order processing time is defined. As orders cannot be shipped instantly, customers
allow for a delivery delay. All unfilled orders remain in a backlog. It is assumed that, once orders
are in the order backlog they cannot be changed or cancelled. Details about the model structure
are given in Appendix 2.
Customer demand.
As show in Figure 3, customer demand is endogenous. It is modelled as a function of both an
industry trend and customer satisfaction. The industry trend is exogenous and determines the base
volume and some extemal variability. However, customer demand is endogenous, since it is
determined by the delivery performance perceived by the customer.
Delivery performance is basically determined by product availability. Any incoming orders
need to be accepted first. A ccepted orders state the desired shipments to be made in terms of time
and volume. In the simulation model it is assumed that orders can only be shipped once they are
complete. Through taking volume for granted, shipment gaps are only related to on-time delivery.
Consequently, delivery performance can be modelled as a function of order acceptance and on-
time delivery. Details about the model structure are given in Appendix 3.
Customer Order Rate = Industry Trend * Customer Satisfaction
Customer Satisfaction = f (Order acceptance, On-time delivery)
Capacity management
In Figure 3 it is pointed out that pressure to change capacity arises as a result of the perceived
supply chain performance. In the model, a distinction is made between available capacity and
desired capacity. The desired capacity is influenced by delivery performance (order acceptance
and on-time delivery) and capacity utilization. After a delay, actual capacity will be aligned with
the desired capacity. Details about the model structure are given in Appendix 4.
Capacity = f (Desired Capacity, Capacity acquisition delay)
Desired Capacity = f (Order acceptance, On-time delivery, Capacity utilization)
SEMICONDUCTOR SUPPLY CHAIN DELAYS - JJ BEZEMER & HA AKKERMANS
Key performance indicators
For tracking the system’s performance, six key performance indicators are identified in Figure 3
(shown in red italics). These are in line with the generic categories found for assessing supply
chain performance (c.f. Neely et al. 1995, Neely 1998).
The supply chain delivery quality is reflected in the CLIP, or “confirmed line item
performance”. This is an industry-specific term for measuring delivery quality, meaning the
percentage of order elements that was delivered as promised. The equivalent used in the
model, is directly related to the workload. It is defined as a non-linear function (f,) of capacity
and desired shipments.
TES ae =
The supply chain cost are reflected in the total on-hand inventory and capacity utilisation.
The total inventory is defined as the sum of WIP and service stock. The capacity utilization is
the ration of the actual production rate and the available capacity.
Inventory =WIP +Service Stock
Prodoution Rate
Fee
Capacity Utilisation. C
The supply chain delivery timeliness are reflected in the stacked lead-time and the delivery
delay. The former meaning the total time necessary for production and distribution. The latter
meaning the number of weeks between order acceptance and actual delivery at the customer.
Both measures are derived using Little’s Law:, which provide a fundamental relation between
WIP, cycle-time and throughput.
WP Service Stodk
Gey deril lead Tire & =
Production Rate Shiprent Rate
z Order Badd
Average Ddivery Dday =. ath
Order Fulfillmen t Rate
The supply chain delivery flexibility is reflected in the RLIP, or “requested line item
performance”. This an industry-specific term for the degree in which market demands could
be met, measured as the percentage of order elements requested by customers that are
confirmed by the company. The equivalent used in the model, is defined as a non-linear
function (f) of the maximum shipment rate and the desired shipment rate.
_ Maximum Shiprent Rate
FUP =6 9 Deseed Sia as a:
WORK IN PROGRESS VERSION
5. Case Setting
Case company profile
For production and distribution, a standard supply chain structure is used (Figure 5).
Technologically speaking, the most complex production process is diffusion (Diff). This is the
longest step in the total production process, and the most costly one. With a series of steps pieces
of silicon are transformed into a set of wafers containing numerous ICs. The wafers produced are
tested (WT) and put on stock (Die Bank). In response to customer demand, the ICs are cut from
the wafer, assembled (Assy) and tested once more (FT). They are moved to an industrial
warehouse (IWH) and from there either to a regional distribution centre (RDC) or directly to one
of the major customers within the context of a specific customer program.
Material Stock Metal Bank Die Stock Finished Product Stocks
> Diff V Diff >) WT >W Assy/FT > V-V —— 4
IWH RDC
Stock point V>
\ Process Guctomer
Proarams
Figure 5: Semiconductor manufacturing supply chain structure
Semiconductor industry trends
The market and supply chain setting this case company was in is typical for its industry. One list
of key trends in this major industry is the following:
¢ Rapidly changing technologies: In general, the semiconductor industry can be characterised
by a relatively high rate of change in products, processes, technologies, and organisational
structures (Fine, 1998). Great advances in IC-design and process technology have triggered
this rate of change over the last two decades. This industry has witnessed an exponential
growth in the number of transistors per IC. With a doubling period of approximately 18
months, this phenomenon, which is called “Moore’s Law” holds until today. As products
becoming increasingly complex, more products and services emerge. However, product
lifecycles decrease rapidly.
* Cyclic and volatile market behaviour: Semiconductor companies operate in markets where
demand is cyclic. The industry is faced with periods of rapid market growth followed by
periods of declining markets. Besides that, as relatively upstream companies in the supply
chain, high variability in demand volume needs to be taken into account. Forecasting is very
difficult and customers do not always commit to their demand. However, in case of a stock-
out there is a great probability that sales will be lost.
¢ Asset-intensive production Capital utilisation is an aspect of great importance in this industry,
due to the fact that huge capital investments are required to build new plants. For example, an
investment of about EUR 2 billion is required to build a new diffusion plant, or “wafer fab”
as they are also called. Therefore, investments in factories are typically related to long-term
planning; decisions are taken on a four-year basis.
¢ Long manufacturing lead-timres: In the semiconductor industry, the time necessary for
production and distribution of semiconductor products is normally between 10 and 15 weeks.
This time is much longer than the lead-time customers expect when putting in an order. This
forces the industry to strategically manage inventory at various stages in the pipeline.
10
SEMICONDUCTOR SUPPLY CHAIN DELAYS - JJ BEZEMER & HA AKKERMANS
¢ Differentiating customer demands. Customers have become more and more demanding, as a
consequence of increased business globalisation and competition. As products evolve faster
and non-traditional competitors enter the market, customer focus becomes a vital driver of
loyalty (Tracy & Wiersema 1997). As a result, a “one approach fits all” -policy cannot be used
anymore. Products become increasingly customised and a mix of operating models needs to
be adopted in order to support the customer's demands. In short, “differentiation” is
becoming an industry standard.
Fragmentation. Traditionally, business fulfilment in semiconductor companies is very
complex as a result of the large number of products, process steps, and routings through the
supply chain. However, complexity is multiplied by customer intimacy and the need for
differentiation, global dispersion of production facilities and customers, and outsourcing of
production processes. As a result, demanding customers are increasingly faced with
increasingly fragmented supply chains. This implies that companies in the semiconductor
industry face great challenges in managing their supply chains.
6. Empirical model refinement and partial model estimation
Strictly speaking, “all models are wrong, but some can be useful” (Sterman 2000). All models are
imperfect representations of reality, and one can only strive to make them as good an
approximation as one possibly can. In system dynamics modelling, this is usually done in a
strongly iterative process of, three separate but interlinked activities:
a) Identifying model structure through structured conversations with knowledgeable experts
working within the system being modelled, usually through structured workshops and group
mode building sessions (Akkermans 1995, Vennix 1996, Akkermans and Vennix 1997).
b) Collecting quantitative data on key parameter settings and time series of performance over
time for key model variables;
c) Developing and testing simulation models that represent the structure identified under a) and
generate simulated behaviour that can be compared with the historical performance as
collected through b).
All this leads to new rounds of a), b) and c), as discrepancies between simulation and historical
teality lead to additional data analyses and to additional sessions with real world system experts.
This iterative process is visualised in Figure 6.
PROGRAM-
MING
model adaptions,
values, functions,
scenarios
time series,
functions,
values
ANALYSIS interesting
findings
search
questions
Figure 6: The iterative process of model refinement and validation (from Akkermans 1995)
aa
WORK IN PROGRESS VERSION
This is precisely what has been done by the first author, who conducted his Master Thesis
research at the central SCM competence centre of this semiconductor company, from where he
had frequent contacts with the local business units serving particular market segments (Bezemer
2003).
Most of the key parameter settings for processes, policies and models of human behaviour could
be elicited from discussions with representatives from the company. A list of some of the key
settings in A ppendix 5 and 6. On the basis of these settings, the simulation model as described in
Section 4 has been simulated. To calibrate the model further, and also validate if its overall
pattems of behaviour fit sufficiently with company data on historical performance, a number of
partial model estimation tests have been conducted.
Sales and production output
The overall behaviour of the model is assessed, through visual comparison of the distributions of
teal world data with simulated date. The assumption made here is that of the simulation model as
a black box. Under this assumption, it should hold that when the model is run under the same
conditions (exogenous variables) as the real world system, the outputs (endogenous variables)
should be broadly similar. In our model, this generates a problem, as almost all variables are
endogenous. Similar to Oliva and Sterman (2001), we have chosen to treat customer demand
temporarily as exogenous.
Production rate (units/month) Sales (units/month)
2.0 250
15 CaN 200
‘ WS
= Ne q 100
0.5 50
0.0 0
1999 2000 2001 2002 1999 2000 2001 2002
Time Time
Actual
Simulated
Figure 7: Comparison of historical and simulated data on production and sales
At the case company, time series on supply chain performance from the period of 1999 up until
and including 2002 are available. This period reflects the highly cyclic character of the
semiconductor industry, since it includes a spectacular uptum and a downturn in the market of
over 30% in one year. In order to generate a customer demand pattern over this period, available
sales data were translated one month back in time, so as to represent incoming customer orders.
Although no hard data are available on this, we know from our interviews that the actually
received customer order rate for 1999 and the beginning of 2000 was considerably higher than
this translated sales data series suggests. This is because during this period, frequent shortages
occurred in the industry (c.f. WSTS 2003). Nevertheless, as can be seen from Figure 7, both
simulated sales and production output show a reasonable visual fit with our historical data. The
12
SEMICONDUCTOR SUPPLY CHAIN DELAYS - JJ BEZEMER & HA AKKERMANS
sales data first show a rapidly growing market, which peaks in October 2000. A fter that, sales
decline dramatically.
Supply chain cost: inventory and capacity utilisation
In the long run, the fixed cost of the advanced production facilities is the biggest cost driver in the
semiconductor business. As Figure 8 shows, both historical and simulated data on capacity
utilisation show that the factories were blowing at full capacity in 1999 and 2000, only to fall
dramatically in 2001 and to recover somewhat in 2002 again. As production rates were adjusted
downwards later than that the market tumed, inventory reached its peak in 2001. Also, as we have
seen in Figure 4, higher workloads during this period result in an exponential increase in work in
progress, hence, in inventory.
Inventory (units) Capacity utilization (%)
S25 =
——s | aie
40%
10
0.5 20%.
0.0 0%
1999 2000 2001 2002 1999 2000 2001 2002
Time Time
Actual
Simulated
Figure 8: Comparison of historical and simulated data on inventory and capacity utilisation
The fit between simulated and empirical data for capacity utilisation is fair, but the fit for
inventory is less clear. For this there may be a number of reasons, of which four key ones are:
Value versus units: The inventory data available is not in units but in value, calculated as
units times factory cost price. This cost price is not updated frequently, which creates
information distortion. More importantly, when expected sales are high, the fixed cost
allocated per product can be low. When expected sales drop, fixed cost allocations will drop.
Hence, the strange phenomenon occurs that, in the accounting books, when demand for
products goes down, their accounting value goes up.
“Factory physics”: A fundamental relation between WIP, available capacity, and queue-time
is used in the model (Hopp, Spearman, 2000). This relation assumes maximum randomness.
It seems plausible that, compared to the real world, the model assumes higher levels of WIP
(and therefore inventory) are necessary for realising a specific level of utilisation
Generic product In the simulation model, one generic product is assumed. This makes it
possible to “flush” the supply chain empty. In the real world however, a product mix prevents
from doing this. It is likely one has the wrong products on stock.
WIP versus final product: Finally, the historical data add up WIP and final stock into one
monetary number. As Figure 9 shows, final product stock rises to its peak later than WIP
does. As this final stock also has a higher intemal cost price, this effect may play out stronger
in value terms.
13
WORK IN PROGRESS VERSION
Most likely, all four of these factors contribute to the mismatch between simulation and reality,
but the relative strength at which they do is not well known. What is important to note here
though is that, in the real world, it has taken inventory levels at this company almost two year to
come back in line again with the sales rate. In our simulation model, this is at least a year, so still
quite long compared to a normal stacked lead-time of some 10-14 weeks.
Inventory (units)
a0 WIP
25 ge SN Service stock
: vk ii Customer demand
2.0 —
154
10
0.5 Wan
0.0
1999 2000 2001 2002
Time
Figure 9: Distribution of WIP and final (service) stock over time
Supply chain flexibility: RLIP
When customer demand is higher than the capacity of the production system can handle,
customer demand has to be refused. Obviously, when this happens it suggests inflexibility of the
supply chain. At the case company, this was measured as RLIP, or requested line item
performance (see Section 4). An RLIP of much more than 80% is always problematic, as this
assumes that the company would be able to deliver all products in its portfolio within the target
delivery delay of three weeks.
RLIP (%)
100%
80%
60% +
40%
20%
0%
1999 2000 2001 2002
Time
Actual
Simulated
Figure 10: Historical and simulated data on supply chain flexibility in meeting customer demand
14
SEMICONDUCTOR SUPPLY CHAIN DELAYS - JJ BEZEMER & HA AKKERMANS
As could be expected, during the peak sales period of 2000, RLIP could not be maintained at its
target level but dropped considerably. The simulation model replicates this neatly. Interestingly,
the historical RLIP also dropped once more early 2002, and again the simulated data follow this
trend.
Supply chain delivery quality: CLIP
In theory, CLIP, or confirmed line item performance, could be well at its target level of 95% all
the time, even during periods of peak demand. This is because all the company has to do is to
deliver the products it knows it has on stock or can assemble shortly to the customer on time.
However, as Figure 11 illustrates, this has not been the case. The explanation for this is well-
known from the literature on service quality (Oliva and Sterman 2001, Akkermans and Vos 2003)
and has to do with the workload that the employees are experiencing: they are well aware that
there is a pile of orders waiting to be filled and start making mistakes as a result of this. Figure 11
shows the CLIP as measured at the company versus the inverse of workload as it is calculated in
the simulation model. A gain, the fit is reasonable, with a slightly less dramatic drop in delivery
quality in the model than in reality in 2000, but a swift recovery as market demand reduced
sharply towards the beginning of 2001.
CLIP (%) & 1/workload
110% 25
100%
(TONG 1s
90% eX FissAl
80% V
70% :
1999 2000 2001 2002
Time
Actual
Simulated
Figure 11: Historical and simulated delivery quality (1/workload)
Supply chain timeliness: cycle times and stacked lead time
Finally, we look at supply chain timeliness. Here we unfortunately only have very limited data on
stacked lead-time performance. We know from anecdotal evidence that around the first half of
2001 the stacked lead-time for some key products was about double of what has been measured
since the beginning of 2002. This is shown in Figure 12.
15
WORK IN PROGRESS VERSION
Cycle-time (days/ML) SLT (weeks)
14 14
12 12
10 J 1 ao,
0.8 :
0.6 ee eS 0.6 ny
0.4 \ —_ 0.4 \ =
0.2 Nd 0.2
0.0 0
1999 2000» -2001.-—S 2002 1999 2000». 2001S 2002
time lime
Actual
Simulated
Figure 12: Simulated and historical behaviour of cycle-time and stacked lead-time
More historical data are available on cycle time, but here the issue is that most of what drives
cycle time, or “days per mask layer”, as the technical term is within the case company, is not
telated to material flows but to technical progress. Hence, most likely, the relative flat descent of
this performance indicator from late 1999 onwards. The sudden drop in the historical data line at
the end of 2000 suggests that some workload-related effect has been present, but only at extreme
levels, much as with CLIP.
8. Supply chain response delays at different levels
Different delays response levels for different supply chain performance indicators
The topic of this paper is delays in responding to market changes for different supply chain
indicators. Based on the behaviour over time charts of Section 7, what can we conclude from this.
First of all, in line with our first research question, the following:
Proposition 1: Regarding delays in responding to market changes for different supply chain
performance indicators, the following distinction can be made for different levels ofsupply chain
management:
Operational level: Order acceptance ratio and delivery quality have relatively short response
delays (10-20 weeks or 1-2 times the normal stacked lead time).
Tactical level: Inventory levels and stacked lead time have relatively longresponseddays(1-
2 years, or 4-8 times the normal stacked lead time)
Strategic level: Capacity utilisation rate has a very long response delays (2-3 years, or 8-12
times the normal stacked lead time).
So, we can group supply chain indicators in three levels: those that respond quickly to market
changes, those that respond slowly and those that respond very slowly indeed. For instance,
capacity utilisation at the case company is, well over two years after the market downtum, still
not at its target rate. This is illustrated in Figure 13. This shows how, during the market uptum of
1999-2000, a significant amount of capacity was lined up (simulated data only) to meet this
demand growth. As demand dropped in 2001, this capacity did not just disappear ovemight.
Indeed, only recently has the case company closed some of its most generic and outdated
16
SEMICONDUCTOR SUPPLY CHAIN DELAYS - JJ BEZEMER & HA AKKERMANS
production facilities. This type of delayed response really classifies as , “not with a bang, but with
a whimper”.
Volume (units/month)
Tree
ee
—— Customer Demand
0.5 ——-Production Rate
—Capacity
o 82 toa 156 208
Time (weeks)
Figure 13: Capacity development and production rates over time
Different drivers for supply chain performance indicators with different delay response levels
But, what is causing these major differences in lengths of delays? Here we have to retum to our
model structure, as outlined in Section 4. What we find there is that the drivers for these three
types of delays are very different, that these drivers themselves have very different response
delays and that this explains the differences in response delays for the various performance
indicators. Proposition 2 summarises our argument:
Proposition 2: The fundamental drivers for supply chain behaviour, including the magnitude of
the delay in responding to market changes, are, for the different levels of supply chain co-
ordination
Operational level (CLIP, RLIP, delivery delay): Order backlog level / Capacity;
Tactical level (Stacked lead-time, stock value): Inventory level / Capacity;
Strategic level (capacity utilisation rate): Market demand level/ Capacity.
CLIP and RLIP are both driven by some ratio of, on the one hand, the current backlog of orders,
and, on the other, the available capacity. As this accumulation of recent demand increases or
decreases, so do they follow, almost without delay. Available capacity is also the denominator at
the tactical level. Here though the key accumulation is inventory, which is the result of all the
production activities initiated in the past periods minus all the sales activity during that same
time. As this accumulation rises and falls, so does stacked lead-time. We have seen that the cycle
time in the wafer fabs is more or less constant, so the main variable in stacked lead-time is queue
time at various stock locations. Finally, at the strategic level there is the capacity utilisation rate.
As we have seen, there are delays of several years in bringing new capacity on line and also in
dismantling it. So, it is not surprising that the supply chain reacts very slowly indeed to drastic
changes in market demand, as it is the ratio of market demand and available capacity that
determines capacity utilisation.
Tradeoffs in policies for improving system performance at different SCM levels
So, what can be done to improve performance? Here, no general answers are sufficient since there
are fundamental trade-offs to be made. This line of thinking is summarised in Proposition 3:
17
WORK IN PROGRESS VERSION
Proposition 3: Shortening response delays for different supply chain performance indicators
always requires making tradeoffs between performance improvement in one area and
deterioration in another, usually at two different levels of supply chain management. The net
result on customer satisfaction and future sales, on cost levels and hence on bottomline
profitability is therefore unclear.
The fundamental drivers for supply chain behaviour, including the magnitude of the delay in
responding to market changes, are, for the different levels of supply chain co-ordination
- More (less) restrictive order acceptance leads to lower (higher) backlog and lower (higher)
WIP levels and hence to higher (lower) delivery quality and shorter (longer) lead times, but
also to lower (higher) demand flexibility.
More (less) restrictive WIP control leads to lower (higher) WIP levels and hence to higher
(lower) delivery quality and shorter (longer) lead times, but also to longer (shorter) delivery
delays.
More (fewer) buffer stocks lead to shorter (longer) delivery delays, but also to higher
(lowever) inventory costs and longer (shorter) stacked lead times.
More (less) aggressive capacity adjustments: lead to higher (lower) supply chain flexibility
(RLIP) and lower (higher) WIP levels during market upturns, which leads to higher (lower)
delivery quality and shorter (longer) lead times, but also to lower (higher) capacity
utilisation levels during market downturns.
Order acceptance. The low supply chain performance at this company, and within the industry as
a whole, during the 1999-2000 period is not just attributable to the unexpected increase in
demand. Companies could have said “No” to some of that demand. As they did, their workloads
would have remained acceptable, so delivery quality would have remained good. This would have
delighted customers and marketeers. Also, as workloads remained under control, WIP levels
would stay at modest levels and hence inventory would not have risen so dramatically, as well as
stacked lead times. This would have pleased the accounting department. But, there is a trade-off
to be made: when customers are told “No, you can’t have your product’, they are rightfully upset.
In our model this tradeoff is not symmetrical: customer dislike it more when they first get their
order accepted and later do not get it delivered on time than if their orders are refused straight
away.
WEP control. A similar trade-off applies to workload control, so, accepting the orders but not
taking more in production that the manufacturing system can handle. Many of the same effects as
described with order acceptance would apply: better CLIP, lower stocks and shorter lead-times.
But, also, this would lead to dissatisfied customers as their delivery delays would increase as the
order backlog, now not immediately channelled through into production, would rise higher and
higher. If those orders would still have materialised during the downtum of 2001, when it became
clear that much of the offical demand was strongly inflated due to shortage gaming (Lee et al.
1997a, 1997b) one cannot tell.
Buffer stocks In these days of apparent supply chain anorexia, it is good to mention the
blessings of well-placed buffer stocks to dampen demand volatility (c.f. Hoekstra and Romme
1999, Gongalvez 2002). Buffer stocks can secure short delivery delays even in times of rapid
market growth, which pleases customers. Unfortunately though, buffer stocks also cost money,
which means a trade-off. More importantly, buffering is only possible for products which are not
fully customised, and a significant portion of IC production is precisely that.
Capacity adjustments. With a focus on customer intimacy (Treacy and Wiersema 1997), a
“capacity cushion” should be created in order to accommodate surges in demand. The line of
reasoning here is not just to be ready for the next uptum, but also to win additional demand by
pleasing customers with very short lead-times, made possible by planned under-utilisation. On the
other hand, if the managerial focus is on cost control, capacity should always be trimmed to be in
18
SEMICONDUCTOR SUPPLY CHAIN DELAYS - JJ BEZEMER & HA AKKERMANS
line with demand. Through being conservative with regard to the demand forecast, high
utilisation and hence high return on investments can be guaranteed. However, lagging customer
demand can lead to erosion of market share. This strategic trade-off can be visualised as shown in
Figure 14.
Lead Demand Meet Demand Lag Demand
“a A\
4
7 / a
Time Time Time
Capacity
Capacity
Capacity
Figure 14: Different strategies for determining capacity adjustment delays
This Figure shows that companies that aggressively pursue increases in market share can follow a
lead strategy, in which they always have access capacity (the staircases in Figure 14). Or, the can
choose to lag market demand, and only have their capacity in line with the market during a
downtum.
9. Conclusion
The semiconductor industry is a highly dynamic one, where major surges and subsequent
downtums in market demand will remain a fact of life. Semiconductor supply chains must be
flexible in dealing with such market volatility. Delays in responding to market changes should be
as short as possible. But, such flexibility comes at a price. This paper restates what is commonly
known in supply chain management: there is always a trade-off. Disappointing some customers
to satisfy others better? Investing more in buffer stocks for greater responsiveness but at higher
costs? Leading or lagging in adjusting capacity to market demand? How to decide when faced
with such multiple, interrelated tradeoffs, is difficult enough for management of semiconductor
firms if all the relevant facts and interrelations are known.
This paper has argued that much of the relevant information is not apparent to executive
decision-makers in the semiconductor industries. As this is a fast-paced industry, it is often
difficult to take a few steps back to see what is actually happening over longer periods of time.
This paper has shown that the length of this period can vary considerably. It can be just one or
two quarters, in the case of supply chain performance criteria that affect customer satisfaction
directly, such as order acceptance or delivery quality. Or this period can be one to two years
before inventory levels and stacked lead times recover after a major boom-bust period in the
market. It can even be several years, in the case of adjustments of costly production capacity to
market fluctuations. Understanding the dynamics of these and others semiconductor supply chain
delays is essential if the right trade-offs are to be made. This paper is intended as a contribution to
such understanding.
19
WORK IN PROGRESS VERSION
References
Akkermans, H.A. (1995) Modelling with Managers. Participative Business Modelling for effective strategic
decision making. Ph.D. Thesis, TU Eindhoven.
Akkermans, H.A. & Vennix, J.A.M. (1997) “Clients’ Opinions on Group Model-Building: An Exploratory
Study”, System Dynamics Review 13(1): 3-31.
Akkermans, H.A., Bogerd, P., Vos, B. (1999) ‘’ Virtuous and vicious cycles on the road towards
international supply chain management’ Intemational Jounal of Operations and Production
Maragerent Vol. 19 Nr. 5/6, pp.565-581.
Akkermans, H.A. (2001) “Renga: A systems approach to facilitating inter-organizational network
development,” SystemDynamics Review 17, 179-193.
Akkermans, H.A. and van der Horst, H. (2002) Managing IT Infrastructure Standardisation in the
Networked Manufacturing Firm. Intemational Jounal of Production Eoononics 75(1) 213-228.
Akkermans, H.A. Bogerd, P., van Doremalen, P. (2003): Travail, Transparency and Trust: A case study of
computer-supported collaborative supply chain planning in high-tech electronics” Forthcoming in
European J ourmal of Operational Research.
Akkermans, H.A. & Vos G.CJ.M. (2003) “Amplification in Service Supply Chains. An exploratory case
study from the telecom industry.” Forthcoming in: Production and Operations Management.
Anderson, E.G. & Morrice, D.J. (2000) “A simulation game for teaching service-oriented supply chain
management: Does information sharing help managers with service capacity decisions?”, Production
and Operations Managenent, 9, 1, 40-55.
Anderson, E.G., Fine, C.H. & Parker, G.G. (2000) “Upstream volatility in the supply chain: The machine
tool industry as a case study”, Production and Operations Management, 9, 3, 239-261.
Baganha, M. and M. Cohen (1998) “The Stabilizing Effect of Inventory in Supply Chains.” Operations
Research 46: S72-S83.
Berends, P.A.J. & Romme, A.G.L. (2001) “Cyclicality of capital-intensive industries: A system dynamics
simulation study of the paper industry”, Omega, 29: 543-552.
Bezemer, J.J. (2003) Modelling the dynamics of business fulfillment: improved understanding of stacked.
lead tine Master Thesis, TU Eindhoven (work in progress)
Cachon, G. (1999) “ Managing Supply Chain Demand Variability with Scheduled Ordering Policies.”
Science 45(6): 843-856.
Chen, F. (1999) “ Decentralized Supply Chains Subject to Information Delays.” Management Science
45(8): 1076-1090.
Chen, F., Z.Drezner, J.Ryan, and D.Simchi-Levi (2000) “ Quantifying the Bullwhip Effect in a Simple
Supply Chain: The Impact of Forecasting, Lead Times, and Information.” Management Scene 46(3):
436-443.Eisenhardt, K.M. (1989) “Building Theories from Case Study Research”, Acadeny of
‘Review, 14, 4, 532-550.
Evans,G.N.,Naim,M.M.,Towill,D.R.(1993) Dynamic supply chain performance: sessing the impact of
information systems.Logistics Information Management 5 (4), 15-25.
Fine, C.H. (1998) Clockspeed; winning industry control in the age of temporary advantage, Perseus books,
Reading, MA (USA).
Flynn, B.B., Sakakibara, S., Schroeder, R.G., Bates, K.A. and Flynn, E.J. (1990) “Empirical research
methods in operations management”, Journal of Operations Managenent, 9, 2, 250-283.
Forrester, J.W (1958) “Industrial Dynamics. A major breakthrough for decision makers,” Harvard Business
Review July-August 1958.
Forrester, J.W. (1961) Industrial Dynamics. MIT Press, Cambridge.
Forrester, J.W. (1968) Market growth as influenced by capital investment. Industrial Management Review,
Vol. 9 Nr. 2, 83-105.
Fransoo, J.C. and Wouters, M.J.F. (2000) “Measuring the bullwhip effect in the supply chain”, Supply
Chain Managerent An intemational journal, 5,2, 78-89.
Goncalvez, P. (2002) The impact of Shortages on Push-Pull Production Systems . MIT oan School of
Maragerent Working Paper, Operations Management/System Dynamics Group.
Hodgson, T. J. and D. W. Wang. (1991a) "Optimal Hybrid Push-Pull Control Strategies for a Parallel
Multistage System .1." International J oumal of Production Research 29(6):1279-1287.
Hodgson, T. J. and D. W. Wang. (1991b) "Optimal Hybrid Push-Pull Control Strategies for a Parallel
Multistage System .2." Intemational Joumal of Production Research 29(7): 1453-1460.
20
SEMICONDUCTOR SUPPLY CHAIN DELAYS - JJ BEZEMER & HA AKKERMANS
Hoekstra, S. & Romme, J. (1992) Integral logistics structures: Developing customer-oriented goods flow,
London: McGraw-Hill.
Hopp, W J., Spearman, M.L. (2000) Factory Physics. 2"* Edition, McGraw Hill, Boston.
Lee, H.L., Padmanabhan, V., and Whang, S. (1997a) “Information distortion in a supply chain: The
bullwhip effect”, Management Science 43, 4, 546-558.
Lee, H.L., Padmanabhan, V., and Whang, S. (1997b) “The bullwhip effect in supply chains”, Soan
Review, 38, 3, 93-102.
Mass, N. (1975) Economic Cydes: An analysis of underlying causes. MIT Press, Cambridge MA.
Meredith, J. (1993) “Theory Building through Conceptual Methods”, Intemational Joumal of Operations
and Production Managerrent, 13, 5, 3-11.
Morecroft, J.D.W. (1983) "A systems perspective on material requirements planning", Decision Sciences,
14: 1-18.
Neely, A., Gregory, M. and Platts, K. (1995) Performance measurement system design. Intemational
Journal of Operations and Production Managenent 15: 80-116.
Neely, A. (1998) Measuring Business Performance. Profile Books Ltd: London.
Oliva, R., & StermanJ.D. (2001) “Cutting comers and working overtime: Quality erosion in the services
industry.” Management Science 47: 894-914.
Repenning, N.P. (2002) “A simulation-based approach to understanding the dynamics of innovation
implementation.” Organization Science 31: 109-127.
Spearman, M. L. and M. A. Zazanis (1992) “ Push and Pull Production Systems - Issues and Comparisons.
Operations Research 40(3): 521-532.
Sterman, J.D. (1989) “Modeling management behavior: Misperceptions of feedback in a dynamic decision
making experiment”, Management Scien 35: 321-339.
Sterman, J.D. (2000) Business Dynamics: Systems Thinking and Modding for a Complex World New
Y ork: McGraw-Hill.
Treacy, M en F. Wiersema (1997) The discipline of mariet leaders Perseus Books, Reading MA.
Van Aken, J.E. (1978) On the control of complex industrial organizations, Leiden: Martinus Nijhoff Social
Sciences Division.
Vennix, J.A.M. (1996) Group Modd-Building: Facilitating Team Leaming Using SystemDynanics.
Chichester: Wiley.
Yin, RK. (1989) Case Suudy Research: Design and Methods, London: Sage.
WSTS (2002) Word Samiconductor Trade Satistics. A ccessed at wsts.org, May 2003
21
WORK IN PROGRESS VERSION
Appendix 1, Structure for production system
Capacity
>| ow > Utilization
Production Production
Start Rate Rate eS
+
‘Capacity
‘i Average CT Raw CT Target Capacity
Utilization
WIP Control .
Deskedt Adjustment +
Production for WIP Upper
ad 7 ™\ _
Desired WIP
we
Adjustment + Adjustment for
Time :
ServiceStock
Desired an Forcasted
Production JT Ore Rate
Production rate
The formula for calculating the production rate is derived from Hopp & Spearman’s (2000)
Practical Worst Case. It is determined by the available WIP, the installed capacity and the raw
cycle time. The average cycle time and the capacity utilization follow logically the realized
production rate.
Production Rate = (WIP * Capacity) / ((RawCT * Capacity) + WIP)
AverageCT = WIP / Production Rate
Capacity Utilization = (Production Rate / Capacity) * 100%
Production starts
The production start rate is the desired production start rate, constrained to be nonnegative.
The desired production start rate is equal to the desired production rate adjusted by the adequacy
of the WIP inventory. The desired production rate is the forecasted order rate adjusted to bring the
service stock position in line with the target inventory level.
Production Start Rate = MAX (0, Desired Production Start Rate)
Desired Production Start Rate = (Desired Production + Adjustment for WIP)
Desired Production = MAX (0, Forecasted Order Rate + Adjustment for Service Stock)
Adjustment for WIP
The desired quantity of WIP is proportional to the average cycle time and the desired production
tate. However, in order to prevent the pipeline from overflowing, an upper: limit is set. This limit
is determined by the target utilization rate.
Desired WIP = MIN(AverageCT * Desired Production, UpperLimit)
UpperLimit = (Capacity * RawCT * Target UR) / (1 - Target UR)
22
SEMICONDUCTOR SUPPLY CHAIN DELAYS - JJ BEZEMER & HA AKKERMANS
Appendix 2, Structure for order fulfilment
Cinnnee
Order Rate
eee ine
Capacity ‘Order Rate aaciiog
we order order
Max. Workload Ratt Fulfillment
Rate
- GC:
Target : 8
Delivery Desired
+
Delay Sniener Order
Fulfilment
+
Service
5 Stock : ae)
Production Shipment
Rate Rate
‘G) *
Desired
Shipment
Stchest Order Rate>
a \ Fulfilment
Maximum Ratio ©
Shipment
Rate +
mati Table for
Order Order
Processing Fulfillment
Time
Accepted order rate
In order to prevent the system from overflowing, order acceptance is limited to the capacity
available for production.
Accepted Order Rate = MIN(Customer Order Rate, Capacity*Max Workload)
Shipment rate
The desired shipment rate is determined by the backlog and the target delivery delay. However,
endogenous order fulfilment determines the adequacy of the inventory available at the service
stock. In addition to that, in order to account for the processes carried out downstream of the die-
bank, the maximum rate of shipments the company can achieve is given by their current service
stock and the minimum order processing time.
Shipment Rate = Desired Shipment Rate * Order Fulfilment Ratio
Desired Shipment Rate = Backlog / Target Delivery Delay
Order Fulfilment Ratio = Table(Maximum Shipment Rate / Desired Shipment Rate)
Maximum Shipment Rate = Service Stock / Minimum Order Processing Time
23
WORK IN PROGRESS VERSION
Appendix 3, Structure for customer demand
Table for Effect of Satisfaction Table for Effect of
‘SeAccepetd on Customer ‘Adjustment Time Delivery Delay on
Market Norm
sbAcceptance Effect of %Accepted
‘on Customer
Satisfaction Customer Satisfaction
Satisfaction |
fo _‘Biectorp Delvery poly Market Norm
Change ia Delivery Delay
Satisfaction ‘Sutstacton.
TeRccepted
>| Perceived by
Time for Market
ae ( 4)
Delivery Delay
Perceived by bq Time for
Market Market to
= Perceive
} Delivery
sAcceptance
Delay
sbAccepted
Perceived by
‘| Company customer’ Delivery Delay
Time for Order Rate Perceived by
‘Company to XN. ee 7 ~ climate
Perceive ‘ompany
soheceptance voncceptes pereelve
livery
Delay
=;
‘Accepted \
oN
_-
me
a ‘onder
Workload Backlog
‘Order: i Rate ‘Order! A.
Customer order rate
Customer order rate is endogenous. In addition to an exogenous industry trend, which determines
the base volume and external variability, customers’ satisfaction (CS) with the perceived delivery
performance is taken into account. Delivery performance is determined by two feedback loops
reflecting product availability (order acceptance and on-time delivery).
Customer Order Rate = Industry Trend * CS
Change in CS = (CS* Effect of %Acceptance * Effect of Delivery Delay) - CS)
/ CS Adjustment Time
Effect of % accepted / delivery delay on customer satisfaction
The attractiveness of the firm's products depends on their availability. Long delivery delays (DD)
relative to the market norm erodes attractiveness. The delivery delay perceived by the market lags
behind the quotes given by the company, which in tum lag the true availability of the product.
First-order smoothing is assumed. The same account for the effect of order acceptance on
customer satisfaction.
Effect of DD = Table (DD Perceived by Market / Market Norm DD)
DD Perceived by Market =
SMOOTH(DD Perceived by Company, Time for Market to Perceive DD)
DD Perceived by Company = SMOOTH(DD, Time for Company to Perceive DD)
DD = Backlog / Shipment Rate
24
SEMICONDUCTOR SUPPLY CHAIN DELAYS - JJ BEZEMER & HA AKKERMANS
Appendix 4, Structure for capacity management
‘Order
ate)
‘Customer ‘Order Intake Backlog ‘Order
saan Tate rultoeet ate
seine Wee
‘wonond
™e/ Axcepted ————P Accepted
uct rn tr coy
ber Perceive Delivery
Time for Company to Belay:
Poceive ncceped 9 CEE
Company [Delivery Delay
a ‘Target %Accepted Company
™, Ae)
Ba Fressure.ct arget Deliver
4s) suaccaptad's Adhst Target Dalvery
‘Capacity
a Table forEtactot
Arena “Aces oressure othe
saya
Effect of %Accepted Copectty
on Desired Capacity
Tala foretectot
Desired Capacity
Effect of Delivery
Belay on Desired <7
= capacity no sels
cme esired age
Production: Capacity
res eee
apact WEIGHT Delivery
acquit Delay vs Capacity
Uilzaton
y vec cs
Capacity Utilization on Desired
Utilization Bs Capacity
Table for Effect of
Capacity Utilization on
Desired Capacity
Pressure of Capacity
capaciy Uainion
Ties cap pets Uiaon Ast
Perceive Capacity ——™] — company [le
‘Target Capacity
wot
Capacity
Pressure to adjust capacity arises when the perceived delivery performance (Yaccepted and on-
time delivery) or the perceived capacity utilisation are not in line with the targets set. After a
delay, actual capacity will be aligned with desired capacity. The weight determines the relative
importance of delivery delay versus capacity utilisation.
Capacity = SMOOTH3I(Desired Capacity, Capacity Acquisition Delay, Initial Capacity)
Desired Capacity = Capacity*Effect of %Accepted * (Weight * Effect of DD)
* ((1-Weight) * Effect of Capacity Utilization)
Effect of Y accepted / delivery delay / capacity utilization on desired capacity
For all three feedback-loops, the modelled structure is basically the same as in A ppendix 3. First a
perception delay is taken into account. A fter that, a table is used (non-linear function) for
calculating the effect on desired capacity.
Effect of DD on Desired Capacity = Table (Pressure of DD to Adjust Capacity)
Pressure of DD to Adjust Capacity = (DD Perceived by Company/Target DD)
25
WORK IN PROGRESS VERSION
Appendix 5, Parameters and values
Where possible, “company facts” and “laws” from the literature have been used to determine the
values of the model parameters. Typically, process times, company targets and industry standards
have been relatively easy to determine. In contrast, the exact details of those parameters reflecting
human behaviour have been more difficult to establish. In these cases, the necessary time-series
and parameters have been estimated with the help of knowledgeable company
representatives.
Parameter Value Source
Process times All these parameters are more or less “facts”,
- Raw cycle-time (before die-bank) 3 weeks which are available in the company. However,
- Minimum order processing time 1 week often the available level of detail is too high.
Information has been aggregated.
4 Internal targets - Process times are derived from internal
B|- Maximum order acceptance 95% measurements.
- Target capacity utilization 80% - Internal targets are estimated with the
- Target safety stock coverage 3 weeks help of the SCM competence centre and
- Target delivery delay 3 weeks business units.
- Industry standard is derived from
Industry standard available benchmark information.
- Average delivery delay 3 weeks
Adjustment delays Although these parameters are explicitly used
- Forecast 26 weeks | in the simulation model, they have the form of
- Capacity 26 weeks | policies in the real world. Therefore they are
- WIP 20 weeks | hard to establish.
- Service-stock 8 weeks - Adjustment delays are estimated with
the help of business lines and the SCM
Strategy competence centre.
- Strategy chosen (lead = 1, lag =0) 0.5 - Company strategy is derived from
relevant documents and presentations.
Company perception delays All these parameters refer to human
- Time to perceive customer demand 4 weeks behaviour. Since specific data is unknown, it
5] - Time to perceive %orders not accepted 4 weeks is most important to use appropriate
&|- Time to perceive delivery delay 4 weeks magnitude.
2 - Time to perceive capacity utilization 4 weeks - Company perception delays are
3 assumed to be dependent on monthly
Market perception delays internal reporting.
§|- Time to perceive “orders not accepted 13 weeks | - Market perception delays are assumed
4 - Time to perceive delivery delay 13 weeks to be dependent on quarterly reviews.
=
Adjustment delays
- Customer Satisfaction 4 weeks
26
SEMICONDUCTOR SUPPLY CHAIN DELAYS - JJ BEZEMER & HA AKKERMANS
Appendix 6, Time-series and non-linear functions
Order Fulfilment
- Table for Order Fulfilment
The non-linear function reflects the following:
* Given the large number of products and frequent changes in the product mix, the
company will never be able to have all the right products in its service stock. At best,
85% of the requests can be met instantly.
Order Fulfillment Ratio
0 1 2 3 4
Max. Shipm / Desired Shipm
Customer Demand
- Table for Effect of %Accepted on Customer Satisfaction
- Table for Effect of Delivery Delay on Customer Satisfaction
The non-linear functions reflect the following:
¢ Itis much earlier to lose a customer than to win one. In relation to this, it is modelled that
customer satisfaction decreases easier than it increases.
* On-time delivery has a stronger effect on customer satisfaction than order acceptance.
This is related the fact that only after order acceptance concrete promises are made.
Customers expect the company to fulfil these.
Effect on Satisfaction Effect on Satisfaction
12 12
1.0 1.0
0.8 08
0.6 0.6 : x
0.4 0.4 : x
SS
0.2 0.2
0.0 0.0
0% 5% 10% 15% 20% 0 1 2 3 4
%Not Accepted Delivery Delay / Market Target
Effect > 1: Customer Satisfaction increases, Effect <1: Customer Satisfaction decreases
27
Capacity Management
WORK IN PROGRESS VERSION
- Table for Effect of %Accepted on Desired Capacity
- Table for Effect of Delivery Delay on Desired Capacity
- Table for Effect of Capacity Utilization on Desires Capacity
The non-linear functions reflect the following:
* In response to low delivery performance, pressure arises to add capacity.
¢ In response to over utilization, pressure arises to add capacity.
+ In response to under utilization, pressure arises to remove capacity.
¢ Adding capacity is easier than removing capacity.
Effect on Desired Cap. Effect on Desired Cap. Effect on Desired Cap.
2.0 2.0 2.0
15 J 15 15
10 f 1.0 10
0.5 0.5 0.5
0 0.5 1 15 1.0 15 2.0 0% 5% 10% 15% 20%
Utilization / Target Delivery Delay / Target %Not Accepted
Effect > 1: Desired Capacity increases, Effect <1: Desired Capacity decreases
28
Back to the
