Operational Modeling & Simulation in Semiconductor Manufacturing

We present a panel session on the role of simulation in improving semiconductor fab operations. The participants include three principal investigators (PIs) from the recently awarded three-year, $1.2 million contracts sponsored jointly by the National Science Foundation (NSF) and the Semiconductor Research Corporation (SRC) on Operational Methods in Semiconductor Manufacturing; the Factory Sciences Program Director from SRC; and industry representatives from the semiconductor manufacturers and from discrete-event simulation vendors. Included here in these proceedings are initial position statements from the various participants, which formed the basis for the panel discussions. For the industry participants, the statements may include, but were not limited to, specific important problems related to the role of simulation in operations in their respective companies, noting any significant technical, managerial, market, or other barriers. The position statements of the academic PIs describe the role that simulation is expected to play in their ongoing research in semiconductor manufacturing and/or their views on the key to successful application of simulation in the industry. 1 INDUSTRY POSITION STATEMENTS Disclaimers: The views expressed in these position statements are those of the individual panel participants, 1035 and do not necessarily represent the views of the employee’s company or other individuals employed by the company. Steven Brown, Siemens AG I believe Siemens Semiconductor Division is today just in the fledgling stages of using simulation to improve factory performance. The emphasis at Siemens is to integrate simulation activities into the decision-making process of the factory managers. For current factories the greatest potential lies in sensitivity analysis of operating policies, with a focus on meeting new production goals while avoiding equipment purchases. For Siemens, there is particular benefit to come from a better understanding of the impact of product mix changes; staffing levels (particularly in the back-end); and people utilization. For future factories simulation should be used effectively with specialized software to evaluate and analyze solutions for equipment layout, material flow, and automated material handling systems to minimize tools, space, costs, and cycle time. Simulation should also be used in conjunction with overall equipment effectiveness (OEE) efforts to evaluate the impact of changes in equipment parameters. The goal here is to produce a priority list of realistic equipment improvement programs for manufacturers of advanced-technology tools (i.e., 300mm). Fowler, Fu, Schruben, Brown, Chance, Cunningham, Hilton, Janakiram, Stafford and Hutchby For all factories, simulation can be linked directly with other common programs: • Scheduling systems can be implemented to improve on-time delivery and can be integrated with systemlevel logistics programs to improve capacity allocation/analysis for business plans. • Simulation can be used to set the targets/goals of factory-level OEE programs and cycle-time reduction programs. Frank Chance, C2MS Productivity Solutions “Credibility is not a gift – it has to be earned. It is built up one step at a time and supported by facts, and by consistency. Even more, credibility is never owned; it is rented, because it can be taken away at any time – Pedro Aspe, 1993." I think that simulation suffers from a credibility gap in semiconductor manufacturing. This is not to say that all simulation projects have been failures – there have been notable successes. But there have been many more cases where results did not meet expectations. As a vendor of simulation and factory analysis software, this is obviously an issue that concerns me. To address it, I believe we need to examine our expectations for simulation users. First, we expect users to be proficient with multiple pieces of computer software – certainly our own, and then probably Microsoft Word, Excel, and PowerPoint, if the user is to ever do in-depth analysis and presentation of simulation results. Second, we expect users to have sufficient project management skills to oversee an implementation of our software. Third, we expect users to be to politically adept, as they must often interact and negotiate with three or more functional business units for most implementations. And let’s not forget that we also expect users to understand simulation, statistics, and even a little probability! Is it any wonder that simulation users face an uphill battle to meet the expectations placed upon them, and in turn, that simulation projects often fail to meet expectations? We could place blame on the university for not providing graduates with this well-rounded skill set. We could place blame on the simulation vendors for products that require too broad an array of user skills. We could even place blame on the client for expecting too much from simulation, and not providing enough skilled users. But shouldn’t we in the simulation community view these expectations as a challenge to be met, rather than as a bar to be lowered? Given these constraints, I would say a good simulation software product is a necessary, but not sufficient, condition for success. My approach is to treat the first several projects with a new client as apprenticeships. An experienced analyst from our company 1036 serves as project manager, and the project team includes members from our company and from the client company. These initial projects concentrate on delivering measurable, concrete results – to build credibility – and on preparing the client to run his/her own projects. For I believe that the best way to train a simulation analyst is to have them work for and emulate a simulation analyst who is successful. It’s slower, surely, and takes more resources, but is there ever a shortcut to credibility? Sean Cunningham, Intel Corporation Fast, cheap, good: pick two. This is the dilemma of discrete-event simulation modeling as applied in operational settings. Fast, cheap models damage the credibility of our discipline. Implicit assumptions are not well understood. Verification and validation are left undone in the rush to results. Customers are disappointed when the results of the model do not match their operational realities. Fast, good models require inordinate computing, customer, and developer resources. Exquisite coordination of operations staff, software developers, and management is required. Several individuals must know their roles immediately and execute on them. Cheap, good models have timelines that exceed the time horizon of the problem itself. The typical factory trains one or two simulation developers in the software technology; their progress is gated by their need to train their peers in the assumptions, needs, and interpretations of their models. The single developer can quickly and easily become overwhelmed by the detail of the problem, and can become frustrated when progress is slow. How to achieve fast, cheap, and good? First, we must realize that not everyone will be a simulation modeling expert. No quantity of additional features added to standard software packages will save inexperienced modelers; experienced modelers will almost always choose to write their own features. It is the duty of management to recognize and reward those individuals who excel in modeling, and to weed out those who do not. Second, we must insist upon model, software code, and developer re-use. A general model that approximates several scenarios is better than several specific models. Object-oriented software tools are a step in the right direction. Intact development teams that persist through several modeling projects are becoming a necessity. Finally, we must pick the right problems. Simulation modelers can tend to live out Maslow’s comment that, to those skilled with a hammer, all problems are nails. Valuable simulation resources must be spent on projects for which simulation is the best feasible solution methodology. Where appropriate, queueing theory, linear programming, Operational Modeling & Simulation in Semiconductor Manufacturing and analytic optimization should be preferred to simulation modeling. In the rapid turn-around time environment typical of operational settings, modelers should be rewarded not only for the simulation projects they perform, but perhaps also for those they prevent. Courtland Hilton, Intel Corporation We simulate complex semiconductor factories simply because they are too large, too complex, and too costly to optimize and refine any easier way. Increased complexity in our products as well as in our manufacturing systems is the natural result of the market and business pressures of today coupled with the hard limits of physics. Our goal is to often have our simulation studies return 100x their investment (although I will do them for a 10x return). We need help to accomplish this. We need tools that allow complex customization but at the same time are robust, bug free, and are supported with detailed and complete documentation and training. We need languages and environments that allow us to increase code reuse by easily exchanging modules, rules, and functions between models. We need languages that abstract us above the code level and let us "speak" in terms of strategies and plans. We would benefit from well-developed industry specific frameworks so that third-party companies and equipment vendors could develop plug and play modules. This framework would also assist many of our simulation engineers who do not have the software engineering background necessary to develop coherent and global frameworks. Data quality and availability is a tremendous problem. Much of that we own internally. But we would benefit from better designed interfaces and database links to facilitate model data loading and complex data management. Validation is a key issue. Models must be provably correct if they are to be used with confidence in high cost decisions. The rub, of course, is how to validate the model and the simulation, especially when one may be modeling a factory of the future that does not yet exist. Such a validation is often a hybrid of comparisons to physics models, to portions of existing manufacturing processes, to specialized experiments, and to intuition. Tools and practices to facilitate these for industry specific models would be of value. Simulation is not always the right tool to use. Even when it is, it is not always clear what level of abstraction should be used within the model. Carefully developed guidelines for generic industry problems would be most useful in reducing the time to solution. We also need vendor support for this. For example, people are generally handled abominably in most packages which treat people as little more than a jig or fixture. We need to represent 103 people who think, who preempt work, who set dynamic goals and who behave, in general, like real people do. Simulation is a wonderful tool and besides, is a lot of fun. As we work to develop these further capabilities we will expand the user base, shorten the time between new question and new model, and increase the effectiveness of our operations. May that be our lot! Mani Janakiram, Motorola Many operational decisions are made in the industry purely based on prior knowledge, experience and intuition. Operational modeling and simulation is being used in all industries to perform factory analysis. The semiconductor industry, in particular, the wafer fab operations, pose several challenges to modeling and simulation, due to varied complexities which include reentrant flows, use of cluster tools in fab operations, etc. Given the high cost of building a fab, high equipment cost, dynamic market changes and technology innovations, it is imperative that Motorola should position itself to be a market leader in all their product portfolios. However, in order to understand the true stochastic implications of an operation, it is necessary to build a meaningful model and perform simulations to study the operation in question. Several models have been build and many simulation studies have been performed to improve fab operations and to achieve the goal of keeping the company profitable and providing world class customer service. At Motorola SPS, factory simulation is performed for capacity planning, scheduling, bottleneck identification, impact of new product/process flow, additional equipment justification, layout analysis, functional equipment modeling, cost modeling, yield modeling, lot size sensitivity analysis, operator modeling, factory ramp-up modeling, etc. The performance measures normally analyzed are: cycle time, throughput, WIP, equipment usage and cost. Like every other industry, Motorola uses simulation for making rational decisions and stands to gain from these virtual factory operations with the help of simulation. Several simulation packages are used at Motorola but the simulation packages by Tyecin Systems (now part of Manugistics) and AutoSimulations are extensively used in addition to Cost Resource Model (from SEMATECH) and others. It is critical that the simulation models provide meaningful data which depends primarily on understanding fab operations, input data accuracy, proper model building and output data validation. It is also essential that the model be kept up-to-date in order to reflect the current factory scenario. This can be accomplished by having a good, user friendly interface between the simulation package and the manufacturing execution system.