Optimizing Fab Design and Deployment

Dario Benavides, Cornerstone Research Co., Menlo Park, Calif. James Duley, Hewlett-Packard, Palo Alto, Calif. Blake Johnson, Stanford University, Stanford, Calif. Michael O'Halloran, Industrial Design Corp., Portland, Ore. -- Semiconductor International, 11/1/2000

Fab investment decisions have grown increasingly complex in recent years because of ongoing increases in the scale, complexity, and cost of fabs, the uncertainty surrounding the costs, benefits and risks of 300 mm technologies, and the increasingly international nature of the facility location decision of many semiconductor manufacturing firms.

This article summarizes the analysis of the cost structure, optimal deployment policies for, value, and financial risks of a range of 200 mm and 300 mm fab design and deployment alternatives that was conducted to provide insight into the semiconductor capacity investment problem. The perspective taken is that of a manufacturer that seeks to maximize the expected net present value of its future profits by deploying capacity to meet uncertain future growth in demand at minimum cost.

To determine the optimal fab investment policy in an environment of uncertain demand, the manufacturer must balance the risk of costly periods of low capacity utilization with the risk of costly periods of capacity shortages. This evaluation is made difficult by the highly uncertain nature of the demand for semiconductors, where standard deviations of 20%-40% per year for individual firms are common. When combined with the 18-24 month construction lead time of a fab, this uncertainty requires firms to commit to fab deployment plans based on demand forecasts that could prove to be 30% above or below actual demand by the time the fab comes on-line.

1. This depicts the challenge of planning for capacity. Depending on demand, a fab could be over- or under-utilized by the time it is in operation.

Figure 1 illustrates the capacity planning problem given construction lead time and uncertain demand. It shows two scenarios: One scenario leads to an under-utilized fab and the other to costly capacity shortages.

In addition to the effect of lead time, the trade-off between the risk of having either too much or too little capacity to meet uncertain future demand is complicated by the economies of scale associated with large fabs. Large fabs are generally substantially more efficient than small fabs both because a range of fixed costs exists in fab construction and operation, and because tools can frequently be better utilized in a larger fab, where the capacity of each type of tool can be more effectively balanced (the "tool granularity" problem in reference 1). To add a large fab, however, a firm must either endure substantial capacity shortages before the fab comes on-line, substantial excess capacity immediately following the deployment of the fab, or some combination of the two.

To overcome the costs of bringing a large amount of capacity on-line at a single point in time, in recent years many firms have explored methods of deploying the capacity of a large fab in a "sequential" manner. This is most commonly done by constructing a large building "shell," which is initially only partially populated with tools, but which is laid out in a way that allows its capacity to be expanded incrementally over time by adding tools. Deploying a large fab this way allows the cost of tools, which comprise the majority of the capital expense of a fab, to be delayed until specific tools are needed.

Analysis conducted

The objective of the analysis was to quantify the impact of fab size, sequential deployment, and 200 mm vs. 300 mm technology on the expected value and financial risk a fab provides to its owner. Six different combinations of fabs, each of which together provide either 40K of wafer starts per month (wspm) of 0.18 µm 200 mm CMOS capacity or an equivalent amount of 300 mm production capacity, were evaluated.

2. Construction costs (including facility and tool costs) for a 200 mm fab (top) and a 300 mm fab (bottom) are shown as a function of capacity. 300 mm has a significant construction cost advantage over 200 mm.

The 200 mm capacity expansion alternatives considered were four 10K wspm fabs, two 20K wspm fabs, or one 40K wspm fab. The equivalent 300 mm capacity expansion alternatives considered were four 4.17K wspm fabs, two 8.33K wspm fabs, or one 16.7K wspm fab. Each 200 mm facility was assumed to be able to begin production with as little as 5K wspm of tools, after which its production could be ramped to its maximum capacity through the addition of tools in response to actually realized demand growth over time. Similarly, each 300 mm facility was assumed to be expandable incrementally from a minimum initial tool set capable of supporting 2.5K wspm of production to its maximum capacity. The lead time of the initial construction of all facilities was assumed to be 18 months, and the lead time of tools to be nine months. Once deployed, the useful life of a facility was assumed to be seven years.

The analysis was conducted from the perspective of a manufacturer that must deploy additional capacity to meet uncertain growth in demand over time. The expected rate of growth of the firm's demand was assumed to be 17% per year, and the standard deviation of its rate of growth to be 20% per year. Both of these parameters were selected to be consistent with SIA data for the growth of the semiconductor industry as a whole.² Finally, the firm was assumed to begin with both initial capacity and initial demand of 30K wspm of 200 mm equivalent capacity.

Fab cost structure assumptions

The cost estimates for each facility considered were prepared by Industrial Design Corp. based on SEMATECH's 0.18 µm copper low-k CMOS process specification. The facility and tool cost for each alternative, including the automated material handling and metrology equipment, are shown in Figure 2 as a function of the amount of capacity installed. Under the base case assumption that the cost of 300 mm tools is on average 1.3 times the cost of comparable 200 mm tools, the figures show that 300 mm fabs have a significant construction cost advantage.

3. Fixed recurring costs for a 200 mm fab (top) and a 300 mm fab (bottom) are shown as a function of capacity. They are about the same.

The fixed recurring costs of each alternative as a function of the amount of capacity installed are shown in Figure 3. This cost component is very similar across the 200 mm and 300 mm alternatives.

The average per-wafer variable cost of production over the lifetime of the capacity installed is assumed to be $220 per 200 mm wafer (of which $125 is the raw wafer), and $807 per 300 mm wafer (of which $600 is the raw wafer). On a per-die basis, 300 mm fabs are thus at a variable cost disadvantage unless and until the cost of 300 mm raw wafers decreases.

The cost of capacity shortages was represented in the analysis by a per-wafer cost that may either be interpreted as the cost of outsourcing the production of a wafer or the cost of the lost sales that result from the wafer shortage. In the base case analysis this cost was assumed to be $2000 per 200 mm wafer and $4800 per 300 mm wafer.

Optimal deployment

The first step in the evaluation of each capacity expansion alternative was to determine the optimal demand level at which to deploy each increment of capacity that is to be deployed under the alternative. The optimal deployment policy was determined both for initiation of the construction of each fab to be deployed under the alternative, and for the deployment of the tools within each fab. An important component of the latter analysis was the determination of whether tools should optimally be deployed sequentially and, if so, how.

4. The optimal demand level at which to deploy the facility is when the maximum of the curve is achieved.

To illustrate the methodology, which was based on "real options" techniques, note that, to find the optimal demand level at which to initiate the deployment of a given increment of capacity, the manufacturer must balance the risk of costly periods of under-utilization with the risk of costly periods of capacity shortages. To illustrate this trade-off, Figure 4 shows the net present value (NPV) of a 20K wspm 200 mm facility as a function of the demand level at which construction of the facility is initiated under the assumption that the facility is deployed all at once. In the figure, the manufacturer is assumed to have 10K wspm of other 200 mm capacity already in place.

There is an optimal demand level at which to deploy the facility, with this level being the demand level at which the maximum of the curve in the chart is achieved. Note that this demand level is not the demand level at which the expected NPV of the facility first becomes positive, which is about 14,000 wspm. The penalty for deploying the facility too early is substantially larger than the penalty for deploying it too late, a much more conservative deployment policy (at demand of about 24,500 wspm) is optimal. This is consistent with previous work, which quantified the impact of low capacity utilization on wafer cost.¹ It can be shown³ that there exists a unique demand level at which it is optimal to deploy each increment of capacity specified by the six different capacity expansion alternatives considered.

5. Optimal deployment policies for 200 mm (top) and 300 mm (bottom) alternatives.

The optimal deployment policies for each of the six alternatives are shown in Figure 5. The images reflect the assumption (noted above) that the firm has 30K wspm of other 200 mm capacity already in place. The line for each capacity expansion alternative in the figure shows the relationship between the firm's demand and the optimal amount of capacity for the firm to deploy the first time it reaches each given demand level. For example, the top panel of Figure 5 shows that when demand first reaches 40K wspm the firm should initiate deployment of about 10K wspm of capacity. It is important to note that, because of the lead time of capacity deployment (nine months for tools, 18 months for new fabs), the firm's demand when the deployed capacity comes on-line, although uncertain, will on average be significantly larger than the capacity deployment "trigger levels" shown in the figure.

Figure 5 shows that it is optimal to initially deploy each 200 mm fab with about 10K wspm of capacity. For the 20K and 40K wspm fabs, tools should subsequently be added incrementally in response to realized demand growth over time. This implies that about 10K wspm of tools is the "minimum efficient scale" of 200 mm fabs under the base case assumptions, and that sequential deployment of larger 200 mm fabs through the addition of tools following this initial deployment of 10K wspm of capacity is optimal. Similarly, Figure 5 (bottom) shows that under the base case assumptions the sequential deployment of 300 mm fabs is optimal following the initial deployment of about 4K wspm of capacity.

Economic benefit

Figure 6 shows the financial value of adding 40K of 200 mm equivalent capacity under each of the six capacity expansion alternatives considered when the optimal deployment policies identified above are followed.

Because future demand is uncertain, the value of each alternative is uncertain. The curves in the figures reflect this uncertainty by showing the probability distribution of the value of each alternative. For comparably sized fabs (measured by die output), the 300 mm alternatives are substantially more valuable than the 200 mm alternatives, with this difference ranging from 28% to 44%. Consistent with reference 4 and reference 5, for both the 200 mm and 300 mm technologies, larger facilities are significantly more cost-effective than small facilities.

Also note that, although the variances of the values of the larger facilities are larger, the risk of loss they generate is comparable because of their substantially larger expected values.

Sensitivity analysis

Sensitivity analysis was conducted to determine how the results of the base case analysis reported above vary with changes in the principal parameters of the model. The Table shows both the value of the parameters in the base case and how the parameters were varied in the sensitivity analysis.

Table. Parameter Values for Sensitivity Analysis
Parameter	Min	Mean	Max
Cost of 200 mm shortage	$1650	$2000	$2800
Cost of 300 mm shortage	$3960	$4800	$6720
Cost of capital	10%	15%	20%
200 mm firm size (in wspm)	10K	30K	100K
300 mm firm size (in wspm)	4.17K	12.5K	41.7K
Expected growth rate	7%	17%	27%
Std. Deviation of growth rate	10%	20%	30%
Lead time of new fabs	12 months	18 months	24 months
Lead time of tools	6 months	9 months	12 months
300/200 mm tool cost ratio	1.1	1.3	1.5
300 mm raw wafer cost	$350	$600	$850

The results of the sensitivity analysis are summarized in the "tornado" diagrams of Figure 7, which show the sensitivity of the value of the best 200 mm and 300 mm alternatives, respectively, to each of the principal parameters of the model. In no scenario is the best 200 mm alternative more valuable than the best 300 mm alternative. The dramatic impact of the cost of capacity shortages and the cost of capital illustrate the central role that capacity utilization plays in fab economics, and the extremely capital-intensive nature of semiconductor manufacturing, respectively.

Similarly, the impact of firm size and growth rate on the results reflects the importance of scale in semiconductor manufacturing. Finally, the results show that the conservative nature of the optimal fab deployment policies limits the impact that demand volatility and capacity lead time have on fab value. For the 300 mm alternatives, uncertainty about tool cost and raw wafer cost are also of limited significance.

Conclusions

The results obtained suggest that 300 mm facilities are significantly more cost-effective than 200 mm facilities, and also that for either wafer size large facilities are significantly more cost-effective than smaller facilities. The difference in value between the 200 mm and 300 mm facilities is smaller than analysis of their capital cost alone would suggest. This is due to the limited nature of the operating cost advantage of 300 mm facilities and, under the assumption of a $600 raw wafer cost, their variable cost disadvantage. Since the actual cost structure of 300 mm facilities has yet to be established in practice, these results must of course be interpreted with caution, and qualified by their assumptions. In addition, the results suggest that the sequential deployment of large fabs in response to actually realized demand growth over time following the initial deployment of a "minimum efficient scale" of tool capacity is optimal.

Finally, the results of the sensitivity analysis highlight the sensitivity of the results obtained to the cost of capacity shortages, and to the manufacturer's cost of capital. The sensitivity to the cost of capacity shortages implies that manufacturers that produce semiconductors with high gross margins should deploy capacity significantly more aggressively than those that produce semiconductors with low gross margins. The sensitivity to the cost of capital implies that manufacturers with low costs of capital should deploy capacity significantly more aggressively than those with high costs of capital. By doing so, manufacturers with low costs of capital can place themselves at a competitive advantage based on capacity availability in addition to cost. .

Dario Benavides is an associate at Cornerstone Research Co., where he specializes in cases involving derivatives, complex securities and valuation. During the last two years of his Ph.D., he worked at HP Labs as a research engineer, developing a compound options-based method for determining optimal sequential capacity expansion policies for a semiconductor firm.

James Duley has been employed at IBM, Allen Bradley and Danish Technical University, Denmark. He is currently employed at Hewlett-Packard, where he has spent most of his career in the area of design automation of computers and ICs. For 10 years, he was the director of the design technology lab at HP.

Blake Johnson is an assistant professor at Stanford University, where his research focuses on the application of financial tools and concepts to the development of better methods for valuing, financing and managing investments in physical assets. In recent years, his work has focused on investments in capacity and intellectual property in the semiconductor industry.

Michael O'Halloran is director of technology for Industrial Design Corp. (IDC), where he is responsible for analyzing technology migration within the industry and anticipating future design requirements for manufacturing facilities.

---

REFERENCES

1. J. Duley, V. Varma and S. Wood. "Sense and Sensibility: The Scaleable Minifab," Proceedings of the 1997 IEEE International Symposium on Semiconductor Manufacturing, October 1997, p. A45-48.
   
2. Semiconductor Industry Association, "Year-End Blue Book Summaries."
   
3. D. Benavides and B. Johnson. "Sequential Capacity Expansion with Increasing Returns to Scale and Path-Dependencies: A Compound Options Approach," Proceedings of the Second Annual Conference on Real Options, Chicago, June 1998.
   
4. D. Benavides, J. Duley and B. Johnson. "As Good as it Gets: Optimal Fab Design and Deployment," IEEE Transactions on Semiconductor Manufacturing, Vol. 12., No. 3, August 1999, p. 281-287.
   
5. J. Duley, B. Johnson and M. O'Halloran. "Financial Comparison of 200 mm vs. 300 mm IC Manufacturing Facilities," Proceedings of the 1999 IEEE International Symposium on Seminductor Manufacturing, October 1999, Santa Clara, Calif.
   

Talkback

We would love your feedback!

Post a comment

» VIEW ALL TALKBACK THREADS

Related Content

• Topics
• Author

Related Content

 

By This Author

There are no other articles written by this author.

SPONSORED LINKS