.
`(cid:14)
`Journal of Operations Management 18 1999 41–59
`
`www.elsevier.comrlocaterdsw
`
`The economics of yield-driven processes
`Roger E. Bohn a,1, Christian Terwiesch b,)
`a Uni˝ersity of California, San Diego, CA, USA
`b The Wharton School, Uni˝ersity of Pennsyl˝ania, Steinberg Hall–Dietrich Hall, Philadelphia, PA 19104-6371, USA
`
`Received 5 February 1998; accepted 6 April 1999
`
`Abstract
`
`The economic performance of many modern production processes is substantially influenced by process yields. Their first
`effect is on product cost — in some cases, low-yields can cause costs to double or worse. Yet measuring only costs can
`substantially underestimate the importance of yield improvement. We show that yields are especially important in periods of
`constrained capacity, such as new product ramp-up. Our analysis is illustrated with numerical examples taken from hard disk
`drive manufacturing. A three percentage point increase in yields can be worth about 6% of gross revenue and 17% of
`contribution. In fact, an eight percentage point improvement in process yields can outweigh a US$20rh increase in direct
`labor wages. Therefore, yields, in addition to or instead of labor costs, should be a focus of attention when making decisions
`such as new factory siting and type of automation. The paper also provides rules for when to rework, and shows that cost
`minimization logic can again give wrong answers. q 1999 Elsevier Science B.V. All rights reserved.
`
`Keywords: Production yields; Cost of quality; Product cost; Rework; Ramp-up; Location decisions; International operations
`
`1. Introduction
`
`Many modern production processes and services are driven by process yields. Not every unit of material that
`starts into the production process makes it to the end as a sellable, high quality product. Some ‘‘fall-out’’ along
`the way due to problems of various kinds. Often, some of the fall-out can be reworked, but always a fraction of
`it must be scrapped. This means that materials and effort go into making something that ultimately cannot be not
`sold.
`The effect of yield losses on the economics of the product, factory, and business can be dramatic. The
`comprehensive Berkeley project on semiconductor manufacturing has documented many examples of integrated
`(cid:14)
`.
`circuit factories with yields below 50% for years Leachman, 1996 . The impact of this is, crudely, that costs per
`good unit are multiplied by two compared with what they would be at 100% yield. The impact on profit is much
`greater.
`
`)
`Corresponding author. E-mail: terwiesch@wharton.upenn.edu
`1 E-mail: rbohn@ucsd.edu.
`
`0272-6963r99r$ - see front matter q 1999 Elsevier Science B.V. All rights reserved.
`(cid:14)
`.
`PII: S 0 2 7 2 - 6 9 6 3 9 9 0 0 0 1 4 - 5
`
`Applied Materials, Inc. Ex. 1010
`Applied v. Ocean, IPR Patent No. 6,836,691
`Page 1 of 19
`
`

`

`42
`
`R.E. Bohn, C. Terwieschr Journal of Operations Management 18 1999 41–59
`)
`(
`
`The main purpose of this paper is to analyze the economics of yield-driven production processes. Despite the
`widespread and important role of yields,
`their impact on economic performance is treated casually in
`management accounting systems, and has received little attention by operations management researchers. The
`result, we observe, is that some decisions are driven by analysis and intuition developed from inadequate
`models.
`A secondary purpose of this paper is to compare the importance of yields with that of labor costs.
`Specifically, we show that under common conditions in ‘‘high-tech’’ industries, the impact of direct labor wage
`rates can be overshadowed by the effect of yields. Even eliminating direct labor entirely can have less effect on
`profit than modest changes in yield levels. Thus, yields matter when asking questions such as ‘‘Where to site
`the next factory?’’ and ‘‘Should we automate a process?’’
`.
`(cid:14)
`Our analysis is illustrated with examples from a high-tech industry, hard disk drives HDDs . Disk drive
`.
`(cid:14)
`production starts with the fabrication of key components heads, media disks, and semiconductors . All of these
`fabrication processes are strongly yield-driven, i.e., much less than 100% of what goes ‘‘in’’ to the process
`comes ‘‘out’’ as good components. The components are then assembled in multi-step, labor- and testing-inten-
`sive processes. These assembly steps are also yield-driven. The industry is sensitive to yield issues, as illustrated
`by the following quotation. Nonetheless, it has not had good tools for quantifying their effects.
`
`It is how you can improve your yield that will get your productivity up. We are not in a business where you
`have a 99% yield. In many cases, there are initial yields on high-end products that are in the 50% range. So a
`(cid:14)
`5% or 10% improvement in these yields is significant Richard Downing, a senior VP of manufacturing at
`.
`Seagate, quoted in Electronic Business Asia, Feb. 1997, p. 35 .
`
`Section 2 of this paper reviews the existing literature on yield-driven processes. Section 3 analyzes yields in
`multi-stage production process. Section 4 motivates our analysis by describing the yield-driven nature of the
`disk drive industry, and the yield-related decisions its managers must make. Section 5 examines the economics
`of rework and scrap in detail for a simple process. It concentrates on variable cost and output as the main effects
`of yield. Section 6 gives our conclusions and points at needed future research.
`
`2. Prior research on yields
`
`The subject of process yields has received considerable attention in various disciplines. We can group this
`research into four streams. First, engineering reports describe yield problems in specific industrial processes and
`provide technical solutions. Second, operations management and operations research models support production
`management of yield-driven processes. Typical concerns are inventories, inspection plans, order releases,
`scheduling and sequencing, and other issues related to production planning. Third, there is an organizational
`learning literature on how to improve yields and reduce ‘‘waste’’. Much of it is empirical- or case-based.
`Fourth, quality management research outlines a number of principles to reduce the ‘‘cost of quality’’. Yield
`losses correspond to internal quality problems, i.e., problems caught before goods leave the factory.
`There are a number of engineering articles and technical reports describing methods of dealing with
`yield-driven production processes, especially in the semiconductor industry. For example, IEEE Transactions on
`Semiconductor Manufacturing has several articles per issue related to yields or ‘‘defects’’. The emphasis is on
`methods, concepts, and tools that will improve yields by detecting diagnosing and solving specific problems.
`(cid:14)
`.
`(cid:14)
`Examples include methods of defect classification Breaux and Kolar, 1996 , yield-loss modeling Stamenkovic
`.
`(cid:14)
`.
`et al., 1996 , in-line product inspection Wang et al., 1996 , statistical software to analyze process control data
`(cid:14)
`.
`(cid:14)
`.
`Burggraaf, 1996 , and expert systems to provide estimates on quality of certain batches Khera et al., 1994 .
`This literature is vital to continued technological progress in these industries. As new products and processes
`push the state of the art, yields fall, and new cycles of yield improvement are needed.
`
`Applied Materials, Inc. Ex. 1010
`Applied v. Ocean, IPR Patent No. 6,836,691
`Page 2 of 19
`
`

`

`R.E. Bohn, C. Terwieschr Journal of Operations Management 18 1999 41–59
`)
`(
`
`43
`
`The random nature of yield-driven processes and the resulting challenges for managing production have
`attracted a number of operations management researchers. Most of this literature takes the production
`technology, and thus,
`the yield problems, as given and provides models supporting standard production
`(cid:14)
`.
`decisions such as how to manage work-in-process and congestion e.g., Chen et al., 1988 , inspection plans and
`(cid:14)
`.
`(cid:14)
`.
`quality improvement e.g., Barad and Bennett, 1996 , scheduling and sequencing e.g., Ou and Wein, 1995 , and
`(cid:14)
`.
`other issues related to production planning e.g., Denardo and Tang, 1997 .
`A smaller group within the operations management literature argue that the overall yields of a production
`process can be improved by effective management of the process. Proposed methods for yield improvement
`(cid:14)
`.
`include inspection policies for quick feedback on the quality of the process e.g., Tang, 1991 , keeping the work
`(cid:14)
`.
`(cid:14)
`in progress level low e.g., Wein, 1992 , and effectively combining items from different batches e.g., Seshadri
`.
`and Shanthikumar, 1997 . In contrast to the engineering literature, these papers focus on improving various
`performance measures, including yields, without really changing the underlying production technology. This
`makes them more general across processes, but limits their potency.
`The third stream of yield research is at the intersection between production management and organizational
`research, especially organizational learning, and has contributed some in-depth empirical studies on yield
`(cid:14)
`.
`improvement. Mukherjee et al. 1995 categorized various quality projects undertaken at a major manufacturer
`(cid:14)
`´
`of wire cord depending on the type of learning approach taken in the projects. A follow-up study Lapre et al.,
`.
`(cid:14)
`.
`(cid:14)
`.
`1996 links these quality projects to waste reduction yield improvement . Bohn 1995a; b looks at factors
`(cid:14)
`.
`which influence the speed of yield improvement in semiconductor manufacturing. Kantor and Zangwill 1991
`give a theoretical model of waste reduction learning. Like the engineering literature, the organizational literature
`has little to say on the economic value of yield improvement. For the most part, yield improvement is implicitly
`treated as a way to reduce costs without looking at other effects.
`.
`(cid:14)
`Finally, under the ‘‘cost of quality’’ paradigm as outlined in Juran and Gryna, 1993 , yield losses are viewed
`as part of internal failure costs and thus, as one of the main drivers of the costs of quality. Juran and Gryna
`emphasize the need to assign economic values to these quality costs, to make them easier to understand for top
`management decision-makers. The cost of quality approach is valuable in its recognition of hidden effects from
`quality problems, and its emphasis on quantifying them. For example, this approach would show that when
`first-pass yields get high enough, in-process inspection can be eliminated, which has various desirable effects.
`However, one of the main benefits of yield improvement is ignored, namely the improvement in effective
`capacity and output.
`In the quality literature, yield loss is the extreme form of a defect — the product is unsalable. Therefore,
`much of the quality improvement literature is applicable to yield improvement. Probably most important are the
`tools and concepts of statistical process control to yield monitoring and improvement. Again, this is most active
`for semiconductors; see the surveyrtutorial by Spanos 1992 . Typical issues include how to detect a defective
`(cid:14)
`.
`machine quickly, what inspection policies to set, and how to modify SPC tools such as control charts to cope
`with the huge amount of data produced by automated semiconductor manufacturing lines.
`Although the literature reviewed above has substantially improved our understanding of yield issues in
`production processes, none of it has provided the basic economic analysis of how yields matter. We attempt to
`extend the literature in three directions:
`.
`(cid:14)
`fl we assign concrete economic values to yield issues Juran and Gryna, 1993
`fl we do not take yields as given, rather, we concentrate on the value of improvement
`fl we look beyond the cost impacts of yield improvement.
`This article can be viewed as an effort to evaluate the value of yield improvement.
`
`3. Multi-stage yield-driven production processes
`
`In this section, we discuss production processes consisting of a sequence of sub-processes, of which at least
`one has yield below 100%. Although defects can occur anywhere, they are detected mainly at test points. An
`
`Applied Materials, Inc. Ex. 1010
`Applied v. Ocean, IPR Patent No. 6,836,691
`Page 3 of 19
`
`

`

`44
`
`R.E. Bohn, C. Terwieschr Journal of Operations Management 18 1999 41–59
`)
`(
`
`important question in designing processes with yield losses is the positioning of tests or inspections. Tests are
`costly, and can sometimes reduce yields themselves. There are various formulations of where to put them.
`Common rules are to position them before expensive or irreversible operations, at the end of modules in
`(cid:14)
`.
`modular subassembly, after low-yield operations to avoid adding more value to bad units , or immediately after
`(cid:14)
`.
`to provide fast feedback for learning .
`operations targeted for process improvement
`At each test point, items are classified into ‘‘good items’’ and various categories of ‘‘defective items’’.
`Whereas good items can continue processing at the next operation, defective items are removed from the line.
`They can then be either reworked or scrapped.
`
`3.1. Yields and rework
`
`Rework means that some operations prior to the defect detection point must be redone, or defects must be
`otherwise repaired. Thus, rework changes the capacity utilization profile of the process. In analyzing the
`(cid:14)
`.
`influence of yields
`and rework on process capacity, we need to distinguish between bottleneck and
`non-bottleneck machines. If rework involves only non-bottleneck machines with a large amount of idle time, it
`has a negligible effect on the overall process capacity.
`(cid:14)
`In many cases, however, rework is severe enough to make a machine a bottleneck or, even worse, rework
`.
`needs to be carried out on the bottleneck machine . As the capacity of the bottleneck equals the capacity of the
`overall process, all capacity invested in rework at the bottleneck is lost from the perspective of the overall
`process.
`A second complication related to rework, which affects bottleneck and non-bottleneck machines, is related to
`the amount of variability in the process. A yield of 90% means not that every 10th item is bad, but that there is a
`10% chance that a given item is bad. Thus, yield losses increase variability, which is the enemy of capacity. The
`best stochastic case is that yields are Bernoulli, i.e., that the process has no memory. Suppose that bad items at
`an operation are immediately reworked by repeating the operation. Even if the actual processing time of the
`operation is itself deterministic, the yield losses force items into multiple passes, and thus make the effective
`(cid:14)
`. (cid:14)
`.
`processing time for a good item a random variable. Hopp and Spearman 1996 Section 12.3 show for this
`(cid:14)
`.
`case that the variability measured by the squared coefficient of variation in the effective processing time
`increases linearly with 1y y .
`(cid:14)
`.
`Capacity losses due to variability can be partially compensated by allowing WIP after each operation with
`yields below 100%. The larger these buffers, the more the capacity-reducing impact of variability can be
`reduced. However, additional WIP increases costs, lead times, and throughput times; it also can hurt problem
`detection and solution, thereby reducing yields.
`
`3.2. Yields and scrap
`
`Scrap occurs when bad items are discarded. Final output is correspondingly reduced. Rework is generally
`preferable, but sometimes, it is technically infeasible or uneconomic. An economic comparison of scrap and
`rework is given in Section 5.
`Strictly speaking, scrap is a special form of rework, where the rework loop includes all operations between
`the defect generating machine and the beginning of the process. The impact of scrap losses on system capacity
`are even stronger than in the rework case, since additional capacity must be added at all stations upstream of
`yield test points, with the most capacity needed at the start of the process. It does not matter where the defective
`unit is actually created, only where it is detected. In order to get 100 good parts at the end of the process, more
`than 100ry must be started at the beginning, where y is the cumulative yield all the way through the process.
`Further, the stochastic variation in load is felt at all stages downstream of the yield loss, not just at the stages
`involved in the rework loop.
`This points to the importance of capacity planning in yield-driven processes. If yields and resulting rework
`requirements are known at the time a line is laid out and remain roughly constant, then capacity planning and
`
`Applied Materials, Inc. Ex. 1010
`Applied v. Ocean, IPR Patent No. 6,836,691
`Page 4 of 19
`
`

`

`R.E. Bohn, C. Terwieschr Journal of Operations Management 18 1999 41–59
`)
`(
`
`45
`
`Table 1
`Summary of yield effects on cost
`Rework is done
`
`Scrap
`
`All material up to failed test is lost
`Incremental material to replace bad components
`Material-related costs
`All labor up to failed test is lost
`Rework labor
`Labor-related costs
`Capacity-related costs More capacity needed in the rework loops of process More capacity needed at all stages upstream of failed tests
`Variability-related costs WIP cost to buffer variability
`WIP still needed but less effective; more capacity needed
`to counteract
`Extra large lots needed in make-to-order systems
`Line never perfectly balanced; more capacity needed
`to counteract
`
`Lead time variability in make to order systems
`
`line balancing is done by increasing the capacity at each station enough to handle its anticipated yield-caused
`extra load. With scrap, it takes the form of increasing the capacity at all upstream stations enough that they can
`keep up with demand at the end of the process. Usually, however, yields are neither known accurately in
`(cid:14)
`.
`advance nor are they constant over time. Instead, the aggregate yield shows both a positive trend learning and
`a week-by-week variation which cannot be buffered out economically, even by finished goods inventory.
`Therefore, once a process starts up, the actual capacity at each stage usually will be ‘‘sub-optimal’’ by static
`criteria.
`A related complication arises in make-to-order situations with scrap. To respond to a customer order of N
`units, we must start Nry at the beginning to compensate for the expected yield losses. This approach would
`work fine, if yields were deterministic. However, since they are not, the production scheduler has to trade off
`the costs of making too much against the cost of making too little. Mathematically this is a newsboy-type
`(cid:14)
`.
`problem Table 1 .
`
`3.3. Cost and ˝alue at different stages of the process
`
`In addition to its effect on capacity, yields determine the value that a good unit of WIP has at various stages
`in the process. This information is, for example, important in deciding where to concentrate process improve-
`ment efforts. A two-point yield improvement has different value at different places in the process.
`The value of a good unit of WIP also help to decide whether it is more economical to scrap a defective item
`or to rework it. For example, suppose that after a test a defective item can be reworked for a labor cost of
`US$10, with a 90% chance of success and a 10% chance that the item must be scrapped. Is it better to pay for
`rework, or to scrap the item? Clearly if x is the value of a good item at that point, the decision rule is to rework
`if 10 -0.9 x. However, determining x is not trivial.
`At the beginning of the process, the value of a good item equals the cost of raw materials. At the end of the
`process, the value is given by the marginal revenue from a good item that can be sold. The value of a good item
`increases as it moves through the process, even if no additional material is being added. Let y be the yield at
`n
`the nth stage. If there are no binding capacity constraints, the value leaving stage n is approximately 1ry times
`the sum of the value entering stage n and the variable costs at stage n.
`This gives two different ways to calculate value: cost-based working forward, and price-based working
`backwards. The two will be equivalent if there is no binding capacity constraint, and differ if there is one. The
`(cid:14) .
`discontinuity in value comes at the bottleneck operation s . After the bottleneck, value is based on selling price;
`before the bottleneck, it is based on cost. An analogous effect will be formally discussed in Section 5. It can
`(cid:14)
`have surprising consequences when cheap raw materials are transformed into expensive products
`e.g.,
`.
`semiconductors .
`
`n
`
`Applied Materials, Inc. Ex. 1010
`Applied v. Ocean, IPR Patent No. 6,836,691
`Page 5 of 19
`
`

`

`46
`
`R.E. Bohn, C. Terwieschr Journal of Operations Management 18 1999 41–59
`)
`(
`
`4. Yield-related problems in disk drive production
`
`In this section, we discuss the managerial importance of production yields based on a particular industry,
`HDDs. We describe the production of HDDs as well as the managerial questions related to yields. The answers
`to these questions will be provided by the economic analysis in Section 5.
`
`4.1. Product and process technology
`
`A HDD is a magnetic data storage device that reads, writes and stores digital data. The main components are
`(cid:14)
`.
`(cid:14)
`.
`the head disk assembly HDA and the printed circuit board assembly PCBA . The HDA includes the head,
`(cid:14)
`.
`(cid:14)
`.
`media disks , head positioning mechanism actuator and the spindle motor. The HDA is sealed in an enclosure
`that shields the HDD from dust and other particles, keeping a contamination-free environment over the life of
`the product. The PCBA includes custom-integrated circuits, an interface connector to the main computer and a
`power connector.
`The manufacturing of HDDs is a complex process. The sub-micron flying heights of the head over the media
`make the HDD vulnerable to contamination by small particles, requiring a clean room environment for many
`steps in production. A second challenge in the assembly of a HDD is given by the high degree of
`(cid:14)
`.
`miniaturization of the components especially the head and the extremely small tolerances in putting the parts
`together. Third, magnetic tolerances are very tight. These challenges make the production of HDDs a
`yield-driven process.
`Assembly of HDDs starts with the assembly of the actuator mechanism, head sub-assembly, disks, and
`spindle motor in a housing to form the HDA. Although this process can be partially automated, it typically is a
`largely manual operation. After the HDA is assembled, an operation known as servo writing is putting a basic
`logical format on the disks. This is followed by several optical and functional tests, which are typically highly
`automated. Finally, the PCBA is added to the HDA and the completed unit is formatted and tested prior to
`packaging and shipment.
`Table 2 includes information about typical component prices and other production data, including yields. As
`we will discuss below, it is typically beneficial to conduct rework on HDDs. The information for the rework
`process is also given by Table 2.
`
`4.2. Competiti˝e pressures on HDD production
`
`The HDD industry is a typical ‘‘high technology’’ industry, meaning that to survive, companies must be on
`the cutting edge, with rapid product innovations. Most product generations last less than 1 year. Furthermore,
`because competitive pressure forces products to be brought to market before they or their manufacturing
`
`Table 2
`Typical HDD data
`
`Material cost
`Direct labor
`Yield rate
`Testing time
`Set of heads
`Selling price
`Demand
`Wage rate
`
`Initial production
`135 US$rdrive
`(cid:14)
`.
`0.9 hrdrive
`(cid:14)
`.
`(cid:14)
`.
`60 %
`1 hrdrive
`.
`(cid:14)
`1 unitrdrive
`.
`(cid:14)
`
`Rework
`27 US$rdrive
`.
`(cid:14)
`1.35 hrdrive
`(cid:14)
`.
`(cid:14)
`.
`70 %
`2 hrdrive
`.
`(cid:14)
`0.25 unitrdrive
`.
`(cid:14)
`
`300 US$rdrive
`.
`(cid:14)
`150,000 drivesrmonth
`.
`(cid:14)
`US$6rh
`
`Applied Materials, Inc. Ex. 1010
`Applied v. Ocean, IPR Patent No. 6,836,691
`Page 6 of 19
`
`

`

`R.E. Bohn, C. Terwieschr Journal of Operations Management 18 1999 41–59
`)
`(
`
`47
`
`processes are fully understood, production techniques are at low stages of knowledge. A low stage production
`(cid:14)
`.
`process is one that is not well-understood and may behave unpredictably Bohn, 1994 .
`This situation usually has two key consequences. First, production yields are well below 100%. Because the
`production process is poorly understood, inevitably much of what is made does not work properly. Over time, as
`more is learned and process problems are identified and solved, yields increase, but they never reach 100% and
`often never get close to it.
`The second consequence of being on the cutting edge is that the product is in short supply. Initial production
`volumes are usually low, because of a variety of problems at the manufacturer or its key suppliers. If the
`product is successful, demand exceeds supply. Over a period of months, the manufacturing plant strives to
`increase output through a process known as ‘‘ramp-up,’’ the gradual acceleration of manufacturing output from
`zero to full capacity. Although other forces also come into play, again the key driving force behind ramp-up is
`usually learning of various kinds. Machine downtime decreases as causes are identified and fixed. Bottlenecks
`are detected and circumvented. More workers are trained for the labor-intensive production steps.
`Notice that low-yields exacerbate the problem of short supply. After all, the other production problems are
`dealt with and units are produced, not all of them work properly. Thus, one way in which output increases is by
`increasing yields.
`
`4.3. Managerial questions
`
`The central research question of this article is ‘‘What is the economic ˝alue of an x% yield impro˝ement?’’
`The following three managerial decisions are driven by this economic value.
`First, the economic value of yield improvement is a crucial input in making process impro˝ement decisions.
`Most process improvement decisions in the HDD industry are geared towards yield increases. Several consulting
`(cid:14)
`.
`companies including a company called Yields-Up promise to improve production yields. Similarly, purchasing
`decisions of new equipment or formation of Kaizen teams can lead to higher yields. Whereas the economic cost
`of such projects can be computed easily, understanding their economic pay-back requires an answer to our
`value-of-yield-improvement question.
`Second, when companies make decisions about new plants and processes, they often have to choose among a
`range of geographic locations, technologies, and workforces. The disk drive industry is characterized by a
`strong separation into two geographic clusters: most product development is done in the US, whereas 65% of
`the assembly is done in Southeast Asia. Further, there is a trend towards moving some manufacturing to
`(cid:14)
`countries with even cheaper labor, such as the Philippines and mainland China for a detailed description of the
`.
`global patterns of this industry, see Gourevitch et al., 1997 . In many cases, moving into a new country has the
`potential to affect yields, particularly during ramp-up of advanced products. Workers and engineers in the new
`factory will not be as fast at debugging problems, or as able to communicate with developers for joint problem
`solving. In addition, infrastructure differences among countries may affect ramp-up and yields.
`To what extent is there actually a trade-off between wage rates and yields in HDDs? Evidence on this is
`sketchy and anecdotal, in part, because of the general confidentiality of yield information, and in part, because it
`is a lot easier to measure wage effects of a workforce than to measure yield effects. One disk media company,
`HMT, says publicly that it manufactures in California because it is easier to ramp-up new products to high-yield
`quickly there. However, many of HMT’s competitors are building their capacity additions near their customers’
`assembly plants in SE Asia. In HDD-assembly, there is general agreement that Singapore today has assembly
`capability and yield as good or better than anywhere in the world. One executive who played a key role in the
`early SE Asian migration of the industry stated that in 1988, yields in Thailand could not compete with those in
`(cid:14)
`Singapore, due to superior worker ability to ‘‘tweak’’ production processes Interview with S.G. Tien, cited in
`.
`Doner, 1998 . Therefore, the effects of yields need to be evaluated at the same time as other consequences of
`factory location, and are likely to have a similar magnitude of impact on the bottom line.
`
`Applied Materials, Inc. Ex. 1010
`Applied v. Ocean, IPR Patent No. 6,836,691
`Page 7 of 19
`
`

`

`48
`
`R.E. Bohn, C. Terwieschr Journal of Operations Management 18 1999 41–59
`)
`(
`
`Third, automation generally impro˝es yields, especially as components get smaller and smaller. At the same
`time, automation reduces the need for labor. Again, an informed choice concerning an automation equipment
`purchase decision requires a detailed understanding of the value of yield improvement.
`(cid:14)
`.
`The last two questions
`location and automation broaden the scope of our earlier discussion to include
`different wage rates. Thus, when comparing different locations or technologies, we not only need to consider the
`economic effect of yields, but also look at the effect of changes in wage rates. In order to support managerial
`decisions, our analysis therefore must be extended to: ‘‘What effect do wages ha˝e on the contribution? What is
`the effect of yields? Which of the two effects dominates under what conditions?’’
`A final question we aim to answer in our analysis is related to the above discussion of the value of a good
`unit. As yield losses in HDD processes occur at various stages in the process, including upfront operations like
`component fabrication, as well as back-end operations such as final assembly, the value of a good unit changes
`drastically throughout the value chain. This leads to the question ‘‘When is it beneficial to rework, and when to
`scrap, a defecti˝e item?’’
`
`4.4. Current practice
`
`In the past, HDD producers answered these questions using standard cost accounting techniques. However,
`accounting systems are quite poor at dealing with yield issues, both prospectively and retroactively. Scrap costs
`are often treated as a separate cost pool, which is not carefully allocated back to individual points in the process.
`Even more basic, accounting systems only look at
`the cost-based numbers, not
`the price-based values.
`Sensitivity analysis on the effects of alternative production methods with different yields is very difficult with
`conventional cost accounting. Because of these problems with accounting numbers, experienced managers in
`yield-driven industries often rely on intuition for relevant decisions, while inexperienced managers make
`mistakes. Even the decision on what to rework and what to scrap, seemingly a technical decision, turns out to be
`an economic choice, and one not captured in a cost-based accounting system.
`(cid:14)
`.
`More recently, high-tech companies are using cost-of-ownership COO analysis to address the above
`questions. Consider, for example, choosing between an automated and non-automated machine. As discussed
`above, the automated machine is likely to have higher production yield and lower labor cost, but will also
`require a larger upfront investment. To support the purchasing decision, the company performs a so-called COO
`(cid:14)
`.
`analysis of the two machines typically implemented in form of a spreadsheet . Yields and capacity utilization
`are important inputs for such COO models. The COO analysis computes the production cost of a good HDD
`from the automated machine and compares it with the cost from a non-automated process. The total economic
`value of the yield and wage differences is then computed using an estimated total volume of drives produced
`(cid:14)
`over the life of the equipment. If this lifecycle cost of owning the non-automated machine with lower yields
`.
`and higher labor cost exceeds the difference in purchasing cost, the automated machine is acquired.
`COO models are better than a pure accounting approach, but are still inadequate, as we will now show.
`
`5. Economic analysis
`
`To address the questions raised above, we have, based on research at several companies in the information
`storage industry, developed a simple mathematical model. The model is targeted towards a managerial audience
`and is based on a strong simplification of the complex HDD production process. It allows us to demonstrate
`how the current managerial practice of analyzing yield-driven processes dramatically underestimates the valu