Hans Ehm, Thomas Ponsignon, Thomas KauImann Corporate Supply Chain InIineon Technologies AG Neubiberg, Germany hans.ehminIineon.com, thomas.ponsignoninIineon.com, thomas.kauImann1inIineon.com
Abstract-The globalization of the world economy as well as progresses in information technology made global supply chains a new paradigm for high-tech and semiconductor manufacturers like Infineon Technologies. Consequently supply chain operational excellence has become a key competitive advantage. Along with it comes the need for an agile, adaptable, and aligned global manufacturing network for mastering the volatile market demand - known as the triple-A challenge. A high degree of integration and automation is required across all stages of the value chain: equipment, factory, and supply chain levels. In this paper we describe several factors from shop-floor to corporate level addressing this challenge, and we outline a successful example of frontend-backend integration. !"#$%" '())"* +,%-./ 0%.(1%23(4-.5 6738#49/ :4-)"7;< +,%""7.57/ 7=$7>>7> ?%174 @7A7" B%""/ I. SUPPLY CHAIN OPERATIONAL EXCELLENCE: A KEY COMPETITIVE ADVANTAGE IN A GLOBALIZED WORLD In the last decades global supply chains have become a standard due to diminished trade barriers that made resource sharing attractive. Also tremendous progresses in inIormation technology now allow managing and controlling complex systems (Stadler and Kilger, 2007). As a result the scope oI operational excellence is no longer conIined to single Iabs, but it encompasses nowadays the concept oI manuIacturing network: the global supply chain is our new Iab (Chien, 2007; Chien et al., 2008). Achieving an eIIicient utilization oI the resources oI the global supply chain is a must Ior those surviving in the semiconductor business but it remains quite challenging due to several Iactors. First, it usually involves dozens oI in-house Irontend and backend Iabrication sites plus silicon Ioundries and subcontractors organized as a dynamic network that is spread all over the world. Secondly, it has to be dealt with long lead times while products tend towards shorter liIe cycles and steeper production ramp ups. In addition, semiconductors are certainly in one oI the most volatile markets with cyclic up- and downturn phases, which make the demand very diIIicult to predict. Finally, capacity expansions typically take long times and are expensive, which comes in contradiction with the need Ior Ilexibility (Uzsoy et al., 1992; Gupta et al., 2006). Consequently, supply chain operational excellence along the Plan, Source, Make, Deliver, and Return processes (Supply Chain Council, 2011) has become a key competitive advantage in the globalized world oI semiconductors. To overcome and also to take advantage oI the characteristics oI the semiconductor industry an Agile, Adaptable, and Aligned supply chain is required. It has been made Iamous as the Triple-A Challenge (Lee, 2004). Agility describes the ability to deal with demand and supply uncertainties, Adaptability depicts a dynamic supply chain that copes with shortening product and technology cycles, and Alignment balances the interests oI multiple players in the supply chain. Looking at the Ilows oI materials that occur between the workstations oI a waIer Iabrication site (Hopp and Spearman, 1997), similar complexity can be observed between the locations oI a semiconductor network (Hopp, 2008). In the Iollowing we show an example oI a successIul Agile, Adaptable, and Aligned global supply chain at InIineon Technologies oI a highly integrated chip Ior a platIorm chipset. The product is initially manuIactured not prototyped at the only technically qualiIied Iactories, i.e. waIer Iabrication in Germany, bumping in Taiwan, testing back in Germany, assembly in Korea, Iinal test back in Germany. Whilst the product matures and penetrates the market, it requires more capacity and new routing opportunities including production partners since the manuIacturing cost begins to dominate Iurther growth. Thus, aIter a year oI booming demand the chip has successively used more than IiIteen diIIerent supply chains (Fig. 1). Figure 1. Example oI a Triple-A supply chain, i.e. Agile, Adaptable and Aligned, oI a highly integrated chip Ior a platIorm chipset at InIineon Technologies: the chip went through more than IiIteen successive supply chains in a one year time period to be able to satisIy booming market demand. 978-1-61284-409-1/11/$26.00 2011 IEEE
Each new route allowed either increasing throughput or decreasing cost (Fig. 2). Obviously, the cost would have been even lower using one single low cost production site Ior the entire value chain, but it is technically hardly achievable within the short product liIe cycle, also the risk would be too high since the manuIacturing Ilexibility is lost, and the learning Irom best practices oI neighbor routes would be lost. Hence, by harvesting the opportunities oI a global network, cost reduction oI mid double digit percent ranges within one year can be reached. Figure 2. On the example oI the chip Irom Fig. 1, the Triple-A supply chain allowed a much steeper production ramp up than initally Iorecasted, i.e. 100 throughput aIter Iour quarters, while achieving drastic cost reductions. II. CHALLENGES FROM EQUIPMENT TO SUPPLY CHAIN: INTEGRATION AND AUTOMATION Mastering the triple-A challenge as described above requires a certain degree oI automation Ior dealing with Ilows oI goods and inIormation at each level oI the global manuIacturing network as well as an enhanced integration across all stages oI the value chain. In the Iollowing we outline several key Iactors Irom equipment, Iactory, and supply chain levels, which contribute to the operational excellence oI InIineon Technologies (Fig. 3). Figure 3. Integration and automation challenges need to be tackled across all levels, i.e. Irom equipment, to Iactory, to supply chain levels, in order to achieve operational excellence. The best example oI operational excellence in the recent past is probably the management oI the threeIold increase oI the Iactory loading in certain technology nodes in only 12 month time aIter the 'cold steel phase in 2008. The steep production ramp up was oI major importance Ior a timely increase oI the market share. The practical consequences imply Ior example integrating new equipment on the shop-Iloor in short delays, dealing with a tremendous amount oI data at each Iabrication site, and keeping track oI lots during intra- and inter-site movements. A. Equipment and Single Process Level To guarantee the quality oI the chips under the constraint oI dynamic, complex, and globally distributed means oI production, advanced inIormation Ilow strategies are required. It is achieved by implementing Statistical Process Control (SPC) as well as Advanced Process Control (APC) techniques directly at the equipment level (Montgomery, 2001; Moyne et al., 2001). A Iully automated data acquisition is a pre-requisite to be able to handle the important amount oI data that is needed. SPC and APC methods are part oI an approach that enables to speciIy and improve quality metrics oI general application (Ryan, 2000), to detect violations and process trends in time, to react quickly, and to prevent Iuture abnormal occurrences. Among others Western Electric rules allow reacting beIore an impending quality metric violation could stop the material Ilow (Western Electric Co., 1985). By means oI these techniques a consistent monitoring report oI the Overall Equipment EIIiciency (OEE) across the shop-Iloor is provided to planning organizations. Consequently, production recipes and test programs are adjusted accordingly. Furthermore, the mentioned well-known best practices are extendable to other levels oI the manuIacturing network Ior documenting the quality oI all processes, not only Ior the Iabrication but also supply chain planning processes. (Russland et al., 2011) shows an example on how to monitor demand Iorecasts with the help oI SPC methods. Finally, the reliable control oI processes on the equipment level and single process level with the help oI SPC and APC methods is the Ioundation Ior Ilexibility options at higher levels. B. Factorv Level Paperless manuIacturing is becoming standard in semiconductor manuIacturing. The main challenge not only consists in having it done at the start oI a manuIacturing line; more oIten the challenge comes when Ilexible processing is required as IT systems have to prove that they are able to organize routine processes as well as to handle agile production. Besides others, this is made possible at InIineon Technologies via an indoor GPS system Ior easy and reliable material identiIication and also with an access control system Ior operators, i.e. only those who are qualiIied to run a certain type oI equipment or process are allowed to operate it. The Iactory challenge in semiconductor manuIacturing has always been to achieve both high utilization rates and Iast processing times (Leachman et al., 2007). The operating curve management is the toolset that is used at InIineon Technologies to reduce the alpha value (Fig. 4). Indeed, alpha is the measurement oI the variability, which means, a low alpha value enables higher capacity utilization and higher throughput Ior a given X-Iactor value, i.e. the ratio between cycle time and raw process time. State-oI-the-art semiconductor manuIacturing Iabs typically run their production with an X- Iactor value between two and three (Fowler and Robinson, 1995; Robinson et al., 2003). Figure 4. For a given ratio between cycle time and raw process time, known as the X-Iactor, a higher capacity utilization U2~U1 is reached via lower variability 21. In order to decrease the overall alpha value oI the manuIacturing network, the variability within each Iab needs to be diminished. Variability may be due to equipment downtimes (scheduled or unscheduled), operator availability (physical presence, trainings.), process issues, cleanroom disturbances, raw material availability. All these disruption parameters need to be as low as possible, and those which are unavoidable should be synchronized. Thus, the synchronization can be done using the 4- (or X-) partner method (Fig. 5). It is shown that the variability oI the entire system goes down when synchronization is achieved, e.g. the break oI an operator occurs at the same time as a scheduled maintenance oI the equipment and the scheduling system takes into account that the work-in-progress in Iront oI the machine has to be processed beIore the maintenance starts. Figure 5. The 4-partner method is used to synchronize disruptions that occur during the processing; it allows decreasing the variability (i.e. alpha). In addition, Automated Material Handling Systems (AMHS) as well as automated cleanroom monitoring are Iurther enablers Ior stabilized processes in manuIacturing areas. C. Global Supplv Chain Level Having statistical control Ior equipment and process as one pillar, and operating curve management, X-Iactor and X- partner method as a second pillar on the Iactory level, the top area is now the supply chain level. Hence, the challenges Iaced in each Iactory are now transposed on a global scale. In other words, equipment on Iactory level becomes an entire Iabrication site among the whole production network, and a manuIacturing route linking equipment on shop-Iloor with respect to the Iabrication recipe becomes a global supply chain that describes the sequence oI Iabrication sites being used. The challenge on global supply chain level becomes integrating all elements oI a supply chain. The material Ilow, the inIormation Ilow, and the value Ilow need to be synchronized. For a semiconductor company like InIineon Technologies, where almost every produced chip is traveling once around the world Irom raw waIer until entering the production site oI the customer, the value Ilow managed in all aspects is one oI the key enablers in order to always IulIill custom and Iinancing requirements and Ior taking Iull beneIit oI the globalization. Global process change management (including qualiIications), transit time measurements Ior regular Ilows, as well as standardized emergency shipment procedures Ior latest demand changes are other aspects oI managing global supply chains. In addition, a master data management that is globally harmonized and interlinked (Fig. 6) is a mandatory enabler to support blended best-oI-breed IT Tools (Fig. 7). Figure 6. An eIIective master data management based on a consistent structure is necessary to support suply chain processes and IT tools. Enablers oI the enabler are the skills needed Ior the human interactions. Indeed one common 'language based on the Supply Chain Operations ReIerence (SCOR) model (Supply Chain Council, 2011) is required to allow having an aligned supply chain. Furthermore, a global online training system called the Supply Chain Academy is also very beneIicial to keep the same level oI knowledge across all worldwide sites.
Responsible Location PLP RGB BE PLP RGB CC PLP MAL DS PLP RGB FE (ECD) PLP GLP SIN PLP Final Product Production Start Location Line Planning Functions RGB PLP Interface Figure 7. Various planner communities are involved to perIorm the diIIerent planning processes as shown on this simpliIied representation oI InIineon`s planning landscape (SP: Sales Planner; MP: Marketing Planner; VP: Volume Planer; CP: Capacity Planner; GLP: Global Logistics Planner; PLP: Production Logistics Planner; CLM: Customer Logistics Manager; ALM: Allocation Manager). A common understanding and level oI knowledge is required company-wide to achieve an aligned supply chain. III. A SUCCESSFUL EXAMPLE OF INTEGRATION AND AUTOMATION: BACKEND MEETS FRONTEND One oI InIineon`s successIul examples oI integration and automation among all three levels is the introduction oI the embedded WaIer Level Ball (eWLB) technology. This waIer level packaging solution uses a combination oI traditional Irontend and backend manuIacturing techniques (Fig. 8), which has the potential to reduce the package size, to diminish manuIacturing costs, and to speed up production since the backend process is literally integrated into Irontend manuIacturing steps (STATSChipPAC, 2007). Figure 8. The introduction oI the embedded WaIer Level Ball (eWLB) technology was and is still challenging Irom equipment, to Iactory, to supply chain level due to its modiIied process Ilow. The implementation oI the new production process was quite challenging across all levels due to its modiIied process Ilow. In the Iollowing, we describe the diIIiculties encountered during the eWLB Supply Chain Task Force Iocusing mainly on chipsets Ior the wireless communication market segment. Note that InIineon Technologies sold its Wireless business division to Intel Corporation in the Iirst quarter oI 2011 (Intel and InIineon Technologies, 2010). A. Equipment Level The biggest challenge we Iaced at equipment level while introducing eWLB was to enable the traceability oI the chips up to the silicon waIers. Traditionally, the necessary data are lot and waIer identiIication numbers. However, during the eWLB reconstitution process each silicon waIer is cut into dices that are picked and placed onto the mold waIers. The Iinal test maps Ior silicon waIers can only be reconstructed Irom inIormation available Irom mold waIer test maps. ThereIore, a dedicated database has been created with data Irom die placement equipment i.e. identiIication numbers oI silicon and mold waIers and x,y-coordinates oI each die on these waIers and Irom test equipment i.e. identiIication numbers and x,y-coordinates Ior mold waIers. This single die tracking concept guarantees the chip traceability. B. Factorv Level The manuIacturing sequence oI eWLB gets beyond the traditional distinction between Irontend and backend since Irontend equipment is being used Ior the redistribution process. The resource sharing brought challenges on Iactory level. First, in terms oI production planning and scheduling a logistics interIace organization dedicated to eWLB production control has been introduced along with the creation oI a responsible location Production Logistics Planner role that coordinates the utilization oI Irontend and backend shared capacities (Fig. 9). Figure 9. The responsible location Production Logisitcs Planner (PLP) coordinates the utilization oI Irontend and backend shared resources Ior eWLB manuIacturing (GLP: Global Logistics Planner; RGB: Regensburg prod. site; SIN: Singapore prod. site; MAL: Malacca prod. site; CC: Chip Card product segment; DS: discrete product segment). Secondly, Ior lot tracking purpose we needed visibility oI the Irontend redistribution manuIacturing step in the backend tracking. This was only possible by re-programming the lot tracking tool, the yield tracking tool, and the traceability tool in order to work in a 'pool concept, which means, to give Before: Lot tracking not possible for redistribution process After: Lot tracking possible for redistribution process in frontend & backend visibility simultaneously to Irontend and backend tools (Fig 10). Thirdly, in terms oI ownership oI yield Iailures, a deIiciency happening during the redistribution process is due to a Irontend technology but is detected during the backend manuIacturing stage. Hence, Ior the very Iirst time the collaboration between both Irontend and backend quality department was needed to identiIy and assign Irontend Iailures to the backend production. This is the only way to guarantee the outmost quality oI the products. Figure 10. With the tools available the lot tracking during the redistribution process was not possible in backend. Thanks to the re-programmation oI the tools to a pool concept approach, visibility was enabled in both Irontend and backend (ECD: Lithography and Electrical & Chemical Deposition equipment; Workstream is the tool being used Ior WIP monitoring) C. Global Suplv Chain Level On supply chain level, the Iirst challenge encountered was the deIinition oI a new product structure and the corresponding data handing process when creating a new product. Traditional assembly speciIications (e.g. waIer data and technology, packing sequence, package type, ball apply material, back side protection material.) were extended to include both reconstitution (e.g. molding material) and redistribution process data (e.g. process line, pad metallization material.). Also, the release oI this new assembly speciIication implied the creation oI a new assembly data creation workIlow involving nine departments in three locations. Another challenge was to agree on the splitting and merging rules Ior the reconstitution process since it impacts the extent oI a quality issue, i.e. the number oI lots that need to be blocked or the number oI waIers that need to be scrapped. We also needed to agree on the Country OI Assembly (COA) since the assembly process steps (i.e. dicing oI the silicon waIer, reconstitution, rounding, redistribution, marking, ball apply, and dicing oI mold waIer) are physically done in multiple locations, but Ior legal purpose the main assembly location has to be deIined. DiIIerent approaches have been considered (e.g. looking at cycle time, cost.). The Iinal decision has been taken to align with other products and to choose the ball apply production site as the main assembly location. Another hurdle was to adapt InIineon`s key perIormance indicators (i.e. conIirmed line item perIormance, yield control, WIP and cycle time monitoring.) in order to include both reconstitution and redistribution process; e.g. changes to the yield reporting structure had an impact on all InIineon planning processes Irom the calculation oI waIer starts Ior a single process step up to the monthly volume rolling Iorecast process on corporate level. CONCLUSION In this paper we have shown that today`s progress in semiconductor manuIacturing goes beyond the strict control oI physical and chemical processes in equipment and single process steps controlled via Statistical Process Control. It even goes beyond the particle-Iree manuIacturing in Iabs that are continuously optimized via operating curve management. Semiconductor manuIacturing beneIits Irom the Ilexibility options oI global supply chains more than other industries since transportation costs are more negligible than Ior most oI other high-tech products. Agile and adaptable supply chains emerge when the complexity in material, inIormation, and value Ilows is understood, reduced to a minimum, and is perIectly managed. II those that work in semiconductor supply chains can be used in aligned end-to-end supply chains Irom suppliers` suppliers to customers` customers, then we are very close to the saying oI Dr. Hau Lee: 'instead oI company to company competition, we are now in an era oI supply chain to supply chain competition (Lee, 2010). ACKNOWLEDGMENTS The authors would like to thank Geraldine Yachi Ior her contribution to the third part oI this paper. REFERENCES |1| C.-F. Chien, 'Made in Taiwan: shiIting paradigms in high-tech industries, Industrial Engineer, vol. 39, no. 2, pp. 4749, February 2007. |2| C.-F. Chien, S. Dauzere-Peres, H. Ehm, J. W. Fowler, Z. Jiang, S. Krishnaswamy, L. Mnch, and R. Uzsoy, 'Modeling and analysis oI semiconductor manuIacturing in a shrinking world: challenges and successes, Proceedings oI the Winter Simulation ConIerence, pp. 2093 2099, December 2008. |3| J. W. Fowler and J. K. Robinson, 'Measurement and Improvement oI ManuIacturing Capacity (MIMAC) Designed experiment report, SEMATECH Technology TransIer #95062861A-TR, July 1995. |4| J. N. D. Gupta, R. Ruiz, J. W. Fowler, and S. J. Mason, 'Operational planning and control oI semiconductor waIer Iabrication, Production Planning and Control, vol. 17, no. 7, pp. 639647, October 2006. |5| W. J. Hopp, Supply Chain Science, New York: McGraw-Hill, 2008. |6| W. J. Hopp and M. L. Spearman, Factory Physics, Yearbook oI Science and Technology, New York: McGraw-Hill, 1997. |7| Intel and InIineon Technologies, 'Intel to acquire InIineon`s wireless solutions business, Joint news release, August 2010. |8| R. C. Leachman, S. Ding, and C.-F. Chien. 'Economic eIIiciency analysis oI waIer Iabrication, IEEE Transactions on Automation Science and Engineering, vol. 4, no. 4, pp. 501512, October 2007. |9| H. L. Lee, 'The triple-A supply chain, Harvard Business Review, vol. 82, no. 10, pp. 102112, October 2004. |10| H. L. Lee, 'Matching supply and demand with sensible sense` and responsive response`, SCM World Live: RaptureWorld webinar, February, 2010. |11| D. C. Montgomery, Introduction to Statistical Quality Control, 4th ed. New York: Wiley, 2001. |12| J. Moyne, E. Del Castillo, and A. M. Hurwitz, Run-To-Run Control in Semiconductor ManuIacturing, Boca Raton: CRC Press, 2001. |13| J. K. Robinson, J. W. Fowler, and E. Neacy, 'Capacity loss Iactors in semiconductor manuIacturing, working paper, March 2003. |14| T. P. Ryan, Statistical Methods Ior Quality Improvement, Wiley Series in Probability and Statistics, 2nd ed. New York: Wiley, 2000. |15| T. Russland, H. Ehm, and T. Ponsignon, 'A worIlow management system Ior cross-system processes on the example oI InIineon Technologies AG, unpublished, Feburary 2011. |16| H. Stadler and C. Kilger, Supply Chain Management and Advanced Planning: Concepts, Models, SoItware, and Case Studies, 4th ed. Berlin: Springer, 2007. |17| STATSChipPAC, 'Embedded waIer level ball grid array, http://www.statschippac.com/services/packagingservices/waIerlevelprod ucts/ewlb.aspx, February 2011. |18| Supply Chain Council, 'Supply Chain Operations ReIerence (SCOR) model, http://supply-chain.org/scor, February 2011. |19| R. Uzsoy, C.-Y. Lee, and L. A. Martin-Vega, 'A review oI production planning and scheduling models in the semiconductor industry part I: system characteristics, perIormance evaluation and production planning, IIE Transactions, vol. 24, no. 4, pp. 4760, September 1992. |20| Western Electric Co., Statistical Quality Control Handbook, 11th ed. DC: Delmar Printing Company, 1985.