MICRO: Yield Analysis

Determining cost of yield to monitor fab manufacturing processes

Daniel N. Maynard, David S. Kerr, and Cynthia Whiteside, IBM Microelectronics

Every IC manufacturer carefully manages fixed and variable costs to achieve profitability. For high-volume, single-product manufacturers, managing the costs of facilities, equipment, labor, and materials is usually well defined. However, allocating a factory's capacity across several products quickly complicates the picture, forcing the engineering staff to prioritize issues across product subsets. Cost of yield (COY) is a measure of the additional dollars debited or credited to production expenses when aggregate productivity falls below or exceeds the business plan. Based on the costs and volumes of each product, COY helps fab managers to understand the economic impact of yield variation from the manufacturing plan.

At IBM Microelectronics Division (Essex Junction, VT), COY analysis has been used since 1995 as an important business management tool. The management staff interprets COY data to validate resource allocations and to ensure that appropriate measures are adopted to correct the production of products with the greatest COY variances. COY can be used to analyze a range of fab processes and therefore reaches beyond the realm of accounting. A manufacturer develops cost targets during planning cycles by projecting productivity and yield. Measuring COY validates and audits the planning process, and large COY variances compel managers to question how fab profits are being generated. Moreover, based on COY analysis, manufacturers can make informed decisions about when to scrap misprocessed wafers.

This article discusses the mechanics of COY calculation and illustrates COY concepts with analysis examples. It suggests how COY may be integrated into fab management activities. Gaining an accurate understanding of how specific decisions affect the dollar value of yields forces the design community to account for yield-limiting decisions and provides a correct measure of design-for-manufacturing initiatives. Finally, senior business decision makers can assemble enterprise-level COY perspectives to guide future investments and activities.

Cost of Yield and Standard Costing

Most semiconductor manufacturers produce multiple products and support multiple technology nodes within the same factory. Using design metrics such as chip size and critical area, productivity expectations can be tuned to product and process characteristics. Based on that information, cost estimates for a specific product drive the return on investment calculations used to establish a business plan.

During the manufacturing phase of a project, the evaluation of actual performance (costs) must be reviewed frequently to ensure that profit margins are maintained. While all profitable fabs manage their costs successfully, many are unaware of the exact financial impact of yield on each lot when losses occur. Even high-volume facilities that produce a single commodity are faced with the inherent variability of complex manufacturing processes.

While not often observed in industry, the ideal yield management credo insists that all factory personnel understand the monetary value of one point of yield improvement (or loss). In fact, this principle is extendable to assessing the financial loss (or gain) of yield variation for every lot across discrete segments of the process. Yield management concepts, combined with management buy-in, induce manufacturing teams to minimize the effects of COY variations.

Cost, or managerial, accounting focuses on providing information to help managers achieve organizational goals. Once objectives (plans) have been established, internal accounting systems are designed to help evaluate the effectiveness and efficiency with which these objectives have been met. The primary goals of cost accounting are to value inventory and the cost of goods sold, ensure that proper controls and audits are in place, and provide accurate and timely financial analyses and reports.

To accomplish these goals, a company must choose an accounting system that is best suited to its needs. At the IBM Microelectronics facility in Essex Junction, VT, a standard cost system was chosen to value inventory and track financials each month. In a standard cost system, a budget is established to determine the unit cost of each product. The unit cost becomes the standard cost of a product for a given time period.

When production begins, actual costs vary from the standard costs, indicating whether production costs are over or under budget. Total performance variance is determined by comparing actual costs with the inventory's standard cost. If the value of the inventory is greater than what was spent, the fab has performed better than planned, but if it is less, the fab has performed worse. A variety of factors contribute to total performance variance, such as actual versus plan production volumes, actual versus plan cycle times, variable spending impacts, actual versus plan purchases, and vendor prices for materials and services. Probably the most important factor contributing to total performance variance, however, is COY variance, which can be directly associated with each product.

COY variance indicates whether the number of good chips per wafer is higher or lower than planned. If a fab can exceed its planned yields, it will be able to start fewer wafers to meet demand, contributing to reduced costs. In an industry in which 70 to 80% of costs are fixed, yields are one area in which fab managers can influence costs directly.

Mathematically, COY is defined as

where C_u is the standard unit cost, V is the volume of goods produced, Y_p is the planned (target) yield, and Y_a is the actual yield.

At IBM, COY is measured throughout the manufacturing line (during wafer processing, wafer finishing, wafer final test, and module burn-in and test) and at numerous sectors within each manufacturing area. Within the fab's cost system, macrosectors, or groups of actual operational sectors, have been established. For example, although wafer processing may encompass 100 to 150 actual manufacturing operations, the fab has four macrosectors encompassing wafer processing (two sectors covering the front end of the line and two covering the back end of the line). Macrosector organizations, which group related process areas, enable managers to determine costs accurately, since scrap can be tracked when it occurs instead of at the end of a process.

The COY formula is very simplistic in its application, since it compares planned yields with actual yields and only considers actual volumes. However, although planned volumes are not compared with actual volumes, the impact of COY is captured in other variances to some degree. For example, if actual yields are below plan, the site, in order to make up the difference, may start more wafers, which causes volume variances. The additional cost incurred to complete additional wafers is captured in the spending variance. Therefore, the true impact of yields is captured in several different variances. In a standard cost system, however, the COY formula is an excellent metric for determining the value of lost inventory and quantifying yield issues in the manufacturing line.

Scrap Decisions

In addition to its managerial accounting role, COY can be used by manufacturing engineers and technicians to make the correct economic decision about when to scrap work-in-progress (WIP). The semiconductor manufacturing industry is often plagued by WIP that does not meet a fab's baseline yield capability. Once the engineering staff has recognized an ailing process sector and identified the afflicted WIP, it must decide whether to scrap the WIP or continue processing. To accomplish that task, the limited yield resulting from the substandard WIP must be estimated, since it will not be known until functional testing at the end of the line. In the most extreme cases, an experienced fab manager might decide to scrap the WIP, but in many situations, a proper decision can be arrived at by examining COY.

Table I presents a hypothetical example featuring three unit-cost cases. In the first case, most value is added early in the process; in the second, value is distributed evenly across the process; and in the third, most value is added at the end of the process. For each sector, the target yield of 90%, loss of 50%, and resulting actual yield of 45% are the same. To understand the financial impact of scrap, COY is computed for each sector individually, assuming that loss is calculated per sector. In the last three columns, the additional cost incurred by carrying the loss over to the next sector is presented. Positive numbers indicate that the WIP should have been scrapped in the previous sector, while negative numbers indicate that a savings was incurred by deferring the scrap decision from the previous sector. In this example, case 3 losses should always be taken in the sector they occur. However, scrapping hardware for losses that occurred in sectors 2 and 3 would not be the best decision for the first two cases.

Sector	Unit Costs ($)			Volume Out	COY ($)			Additional COY Incurred ($)
Sector	Case 1	Case 2	Case 3	Volume Out	Case 1	Case 2	Case 3	Case 1	Case 2	Case 3
1	50.00	25.00	10.00	45	5000	2500	1000	—	—	—
2	75.00	50.00	25.00	40.5	5400	3600	1800	400	1100	800
3	90.00	75.00	50.00	32.805	4068	3390	2260	(1332)	(210)	460
4	100.00	100.00	100.00	23.914845	2434	2434	2434	(1634)	(956)	174

Table I: The effects of unit costs on COY. All three cases have the same final unit cost in sector 4, but the accumulated value is different. Consequently, a scrap decision may be deferred based on COY.

All of the other parameters in the COY formula influence unit cost and the decision to scrap or keep product. Table II illustrates three target-yield cases by sector with the same overall yield (productivity). As in Table I, the effect of a constant loss of 50% in each sector is examined. Case 1 shows that losses in sector 2 should be deferred until sector 4, while case 3 losses should always be taken in the sector where they occur.

Sector	Unit Costs ($)	Plan Yields (%)			COY ($)			Additional COY Incurred ($)
Sector	Unit Costs ($)	Case 1	Case 2	Case 3	Case 1	Case 2	Case 3	Case 1	Case 2	Case 3
1	25.00	70.00	90.00	99.00	2500	2500	2500	—	—	—
2	50.00	95.00	90.00	99.00	3500	4500	4950	1000	2000	2450
3	75.00	99.00	90.00	95.00	3491	5468	7277	(9)	967	2327
4	100.00	99.00	89.41	70.00	3065	5314	9034	(427)	(153)	1757
1–4	—	65.18	65.18	65.18	—	—	—	—	—	—

Table II: The effects of plan yields on COY. While all three cases have the same overall yield, they show different sector targets, indicating that the impact on COY of production in each sector will guide the scrap decision.

Combining the variabilities of more than one parameter from the COY formula quickly creates scenarios that do not necessarily have answers obvious to even the most seasoned fab manager, and additional subdivisions of the process to create more sectors adds further complexity. For these reasons, an automated technique for assessing wafer value and performing what-if analyses is required to empower engineering and manufacturing personnel to make scrap decisions. At IBM, a Web-based tool was developed to tag WIP with an expected limited yield, whereby the resulting COY options are computed and scrap options are identified. Depending on characterization and management direction, the decision to scrap becomes an informed, conscious, and financially viable choice.

Fab Metrics

COY is a high-level metric that should be reviewed weekly by a fab's senior management team. By monitoring overall COY, senior management can quickly recognize yields that vary from the plan. If a manufacturing area such as wafer processing, finishing, final test, or module burn-in and test is recognized as producing substantially higher yields than planned, measures can be identified to restore the yields to their proper level. Exceeding the plan in a manufacturing area or in the production of a specific product creates excess inventory and can result in unrealized additional capacity. If an area is recognized as producing substantially lower yields than planned, the planning process can be revisited and adjustments made.

In each manufacturing area, actual COY is broken down into components. For example, the fab tracks COY for wafer processing and wafer final test, which provides the overall cost to produce a chip as measured against a planned standard cost. Each product has its own standard cost based on the semiconductor process used to manufacture it and the planned final test yield expected. The COY formula computes a cost value for each wafer in the manufacturing line. That value changes as the wafer progresses from wafer processing to final test. The more a wafer is processed, the greater is the cost value assigned to it. The highest processing cost a wafer can have is at the end of the processing cycle or at wafer final test. After the wafer has been tested, the percentage of good chips is compared to the plan. If the number of good chips on a wafer is higher than the planned or expected yield, the wafer has a positive cost value.

Figure 1: Example of a COY chart delineating the COY values for a range of products. The chart breaks down IC manufacturing into its primary categories of wafer processing and wafer final test.

A typical measurement chart used to track COY is shown in Figure 1. Products have been sorted according to whether wafer-processing or wafer-test COY did better or worse than planned. The products on top had a negative cost applied to them, indicating that they returned value to the plan. Products on the bottom, which did not meet planned wafer-processing or wafer-final test yields, had a positive cost applied to them, indicating that they cost more than planned. Products with the highest positive COY values would be targeted for improvement by management.

In the case of the products with the poorest COY values (e.g., product U), management would question whether or not an adequate supply of the product is being produced, since actual yield diverges greatly from the planned yield. Customer commitments are based on planned yields before wafer production begins. When actual yields are off by large amounts, customer orders can be jeopardized. Actions such as running more wafers to compensate for low yields can be taken to avoid serviceability problems. In contrast, products that consistently outperform the plan (the products with the highest negative values in Figure 1) would be scrutinized as well. Perhaps they have benefited from process changes and yield improvements that could be applied to other products.

At IBM, the wafer characterization team uses COY as a metric to determine the health of individual products, common processes, and technology nodes. When the COY indicates that products or groups of products are not performing according to plan, actions are undertaken to drive the COY to zero. Resource allocation plans can then be made across products, technologies, or sectors, and task-force teams can be established. Based on COY values, products can be traced back to their business units, allowing unit managers to scrutinize their objectives, or can be used by characterization engineers participating in scrap decisions to settle serviceability or other customer-sensitive issues. In some cases, WIP may be purposefully continued, despite a poor COY scrap evaluation, if the product is known to have a high profit margin. Nevertheless, in most cases, COY is the primary metric guiding when to scrap WIP in the process line.

Since IBM insists that its macrosectors liquidate WIP in less than a month under normal circumstances, business reporting takes place monthly. At the end of each month, the financial COY analysis is reconciled with the manufacturing production data to ensure accuracy. The manufacturing teams analyze COY on a daily basis to run the business, and monthly reconciliation helps site managers to understand the financial impact of line yields. From a cost-accounting perspective, that procedure facilitates a better understanding of the manufacturing issues at a product level. Thus, the relationship between manufacturing and finance is a symbiotic one, welded together by common business objectives.

Conclusion

Once COY becomes a pervasive metric in the manufacturing area, pressure builds on planners to ensure achievable targets. At the same time, competition and ambitious customer expectations prevent the business from drifting, cornering management into a delicate balancing act. If the planning process is not accurate, an oversupply (inventory) or undersupply (customer serviceability) condition will result. Clearly, both situations are not planned and not desirable. Therefore, the science of yield prediction is extremely important to negating COY variances. That fact should be intuitively obvious to the skilled semiconductor manufacturer. Indeed, substantial research has been conducted on yield modeling and management.

COY's significance goes beyond day-to-day fab monitoring. It enables process managers to financially justify tool upgrades or process changes and enables them to perform process characterizations so that they can make financial decisions about which tools to use to process specific products. It also helps to valuate manufacturing-driven design changes and trade off marginal resource costs. In fact, COY is a useful thermometer of design for manufacturability initiatives that, when combined with the revenue picture, can provide fab personnel with a high level of motivation to meet production goals.

Acknowledgments

This article is based on a poster paper that was presented at the Advanced Semiconductor Manufacturing Conference, held March 31–April 1, 2003, in Munich. The authors wish to thank Dave Anger and Dan Dillner for their early contributions and ideas on the subject of cost of yield.

Daniel N. Maynard is a manufacturing engineer at IBM Microelectronics Division in Essex Junction, VT. A member of the staff since 1989, he has had responsibilities in functional characterization, reticle design engineering, and, for the last nine years, yield modeling, physical design characterization, and design for manufacturability. Maynard is a certified project management professional. He holds several patents and is a member of IEEE and the Project Management Institute. He received a BSEE in 1987 and an MSEE in 1989 from the University of Vermont in Burlington. (Maynard can be reached at 802/288-2042 or danielm@us.ibm.com.)

David S. Kerr works in cost accounting at IBM Microelectronics Division. He began in the company's finance department in 1990, working in the area of fixed assets. He received a BA in mathematics in 1986 from the State University of New York (SUNY) in Potsdam, a BA in accounting in 1988 from SUNY in Plattsburgh, and an MBA in 1990 from Northeastern University in Boston. (Kerr can be reached at 802/769-4666 or dskerr@us.ibm.com.)

Cynthia Whiteside is a manager of a yield-modeling and functional characterization team at IBM Microelectronics Division. In 1988 she began working at IBM's semiconductor manufacturing facility in East Fishkill, NY, where she was active in process manufacturing engineering and yield enhancement positions. She started her career at National Semiconductor in Santa Clara, CA, as a manufacturing engineer. She received BS and MS degrees in chemistry in 1985 from Rensselaer Polytechnic Institute in Troy, NY. (Whiteside can be reached at 802/769-8307 or whitesic @us.ibm.com.)