MES and CIM: At the Center of Productivity
Computer integrated manufacturing will play a vital role in the planned increase in productivity.
John Baliga, Associate Editor -- Semiconductor International, 7/1/1998
 |
At a Glance | | |
|
| ||
| |||
The industry goal for improving productivity, according to SEMATECH, is a 25-30% reduction in cost per chip function per year. Reducing design rules, increasing wafer size and improving yields all contribute to this goal, though wafer size and yield improvement are expected to have a decreasing impact (Fig. 1). "Other equipment productivity" is the area that will have to provide an increasing contribution.
One of the most commonly used metrics for this effectiveness is overall equipment effectiveness (OEE), the ratio of good wafers out to the theoretical maximum. It is a calculated percentage based on a tool's availability, performance efficiency and quality of work. Currently, OEE is estimated to be 30% to 40%, on average, in the semiconductor industry. The goal for 300 mm fabs is to reach 60% OEE, up significantly from the current average.
Another metric for productivity is overall factory effectiveness (OFE). 1 Many consider this to be equivalent to the OEE of the "bottleneck tool." This could be true for a fab running at high volume on only one or two products. For a fab that offers multiple products and build-to-order runs, the concept of a bottleneck tool is not necessarily valid. Methods for quantifying OFE have not been fully developed, but a number of people believe that optimizing OFE is a more effective approach to improving productivity than optimizing OEEs. In fact, optimizing OFE for a fab should automatically improve the OEEs of individual tools. A CIM system can contribute greatly to improving a fab's OFE.
Also important for a CIM system is the use of data to keep yield under control. The MES may not keep track of all the data, but it certainly collects and distributes it. Automated data collection is considered to be a must, not only because of the vast amount of data, but also because it maintains the integrity of the data. This is particularly important, since wafer processing is becoming extremely dependent on metrology data. Future uses of this information will also include advanced process control (APC) applications such as run-to-run control and fault detection
|
| Fig. 1. "Other equipment productivity" will have to provide an increasing contribution to reducing manufacturing cost. Well-constructed CIM systems can support this increase. (Source: SEMATECH) |
MES functional models
There is a degree of variation in how an MES is defined. Some essentially define it as all parts of the CIM system that lie above tool control; others essentially define it as those parts of the CIM system that interface with tool controls. Some refer to the components of a CIM system as MES and peer level applications. The MESA International (Pittsburg, Pa.) model for an MES is designed to encompass all these functions (Fig. 2).2 The model shows the system's components and how it interacts with external systems.
The CIM Framework developed by SEMATECH is a distributed object model for an MES in a fab. Its modular construction is intended to simplify MES implementations, simplify integration of software components from different suppliers and make customization oriented more toward a company's needs rather than to the equipment. It is intended to keep technology from being a hurdle in the development and integration of MESs. What is built on that framework depends on the MES vendors, but it also depends on what a particular fab or company needs. It has reached version 2.0, and it is the initial standards balloting process at SEMI.
|
| Fig. 2. An MES, or an MES with peer level components, has many functions and is connected to many functions outside the fab. (Source: MESA International) |
Continuous and flexible processing
One of the goals in the standards development for 300 mm fabs is continuous processing (Fig. 3).3 I300I, International SEMATECH and J300 are all championing standards to make it a reality. Many business models also call for a great deal of flexibility in the way lots are handled. The model of producing large volumes of only one product will still remain, but the model of producing multiple products in quantities tuned to the market need will gain more prominence. Balancing these two goals will be a challenge, because the needs of continuous processing tend to drive WIP levels up, and the needs of flexible processing tend to drive it down.
The CIM system in the fab must support the continuous and flexible processing goals. A wafer lot must be delivered to a tool and ready for processing before the previous lot is finished processing, and this must be done continuously. This is a significant problem for a scheduler when events such as hot orders and unscheduled downtime for a critical tool occur.
Scheduling is primarily done using dispatching and queuing models that have grown in sophistication with the industry's needs. Some believe that repair-based methods are better for taking the whole fab into account. (See sidebar on "Scheduling Can be Better Without Following the Rules.")
Vigilant systems
|
| Fig. 3. Continuous processing at each tool can improve factory throughput. (Source: I300I) |
There are events that can occur in the pursuit of continuous processing, like turbopump crashes, that can destroy the wafer being processed. The cost of a 300 mm wafers, however, is prohibitive for crash-to-crash operation. Forward looking approaches, or vigilance, are needed to route product wafers away from tools that may be on the verge of a crash.
On the concept of a vigilant system, Jon Golovin of Consilium said, "The concept of a vigilant system is that it looks at all five or 10 thousand things (that affect fab operations) because it has the computing power to do so today, and instead of giving you a report with 10,000 data elements, it says there are 25 things you need to worry about. The other 9,975 are OK." Vigilance requires a great deal of not only data handling, but also data filtering.
Golovin continued, "The fab today is overwhelmed by data. There is no one who isn't overwhelmed by data, most of which says, 'It's OK.'" An important shift in focus in the way CIM systems are implemented will likely be from collecting and communicating data to using data and only communicating useful data.
Supply chain management
Technically speaking, supply chain management is a separate function from that performed by an MES. In the semiconductor industry, however, manufacturing is typically performed in two facilities: the front-end fab and the back-end packaging house. A significant amount of miscommunication, yield loss and lack of traceability has occurred in this hand-off. This will no longer be allowable.
A number of solutions are cropping up to solve these problems. For example, ink dot marking is expected to give way to wafer map files. These files can be loaded directly into automated pick-and-place equipment, which then only uses the good devices on the wafer. This transfer of data with the wafer between facilities is an extension of an MES's responsibilities. Yield and automation levels in back-end facilities will soon have to catch up to that in front-end fabs, and having the hand-off problems solved ahead of time will ease the transition.
The Semiconductor Manufacturing Data Exchange format (SMDX), created by N-Able Group International (Fremont, Calif.), is one of the things currently in use that is designed to solve the data hand-off problem. Currently, much of the information exchange between fabs and packaging houses is done by custom arrangements. Standardization of this data communication will allow fabs and packaging facilities to pass the required data without regard to the specific CIM systems used in each facility. SMDX also enables communication of data between fabs and customers' or headquarters' functions.
A concept that is well developed in the supply chain arena, the build-to-order (BTO) supply chain, is also quite applicable in semiconductor manufacturing. Some of the main benefits of the BTO concept are reduced inventory, higher production speed, reduced depreciation exposure, reduced exposure to demand volatility and faster product transitions. Skills required to implement a BTO supply chain include the ability to plan under uncertainty, the ability to plan to a high degree of granularity and accuracy, the ability to conduct planning and implement changes quickly, the ability to plan around events rather than a fixed schedule and the ability to globally monitor activities in the chain. Principles like these will have to be applied to the fab floor to improve productivity.
The RHYTHM suite from i2 Technologies (Irving, Texas) performs a number of supply chain and enterprise resources planning functions, as well as detailed factory planning and scheduling. Manugistics (Rockville, Md.), a supply chain solutions provider, recently made moves to extend its reach into the execution arena. Manugistics recently acquired TYECIN Systems (Los Altos, Calif.), a supplier of scheduling, planning and analysis products for wafer fab and assembly and test operations.
Packaging, assembly and test facilities
The level of automation used in packaging, assembly and test facilities is starting to catch up to that in front-end facilities. It started with improved tool and intertool transport automation, and MESs are now becoming more prevalent. Many of these facilities are contract houses that handle many products and customers, which is the kind of environment an MES is designed to handle.
Abpac recently purchased FACTORYworks from FASTech (Lincoln, Mass.) for its Phoenix, Ariz., facility. Amkor Technology recently chose MESA from Camstar Systems (Campbell, Calif.) to control operations in all of its packaging facilities in the Philippines, and ChipPAC is using it for its Shanghai, Malaysia, facility.
MES implementations
In a particular MES implementation, there are two aspects that must be differentiated. One is the architecture and computer technology on which it is based, and the other is the set of methods and functions it is designed to perform. Since the amount of data to be collected and number of decisions to be made with it will continue to grow exponentially, it should use the most effective architecture and computer technology available. An open, distributed object architecture is gaining preference, because it is the most flexible and easiest to upgrade. The CIM Framework is based on a distributed object architecture that uses the CORBA object request broker (ORB). This has raised some controversy now that it is in the standards balloting process. (See sidebar on "COM and CORBA.")
MESA, the MES suite from Camstar Systems (Campbell, Calif.), is built on a client/server architecture with Windows 95/NT clients and the IBM AS/400 as a transaction and database server. It includes Windows-based applications for access to cell controls and enterprise applications.
Consilium (Mountain View, Calif.), supplier of WorkStream DFS, is introducing its Fab300 MES. Fab300 will be on a Windows NT platform, and it will offer fab vigilance and virtual fab features. Its vigilance feature is intended to help anticipate problems in the fab and work around them. The virtual fab concept is intended to include multiple facilities in its material control and process control tasks.
FASTech Integration (Lincoln, Mass.) has both MES and cell control offerings. Its FACTORYworks MES and its CELLworks cell control products are designed to provide bottom-up, object-oriented, distributed transaction processing for factory control. Its architecture has a high degree of compliance with the CIM Framework. FASTech recently acquired Midas Software (San Jose, Calif.) to improve equipment maintenance capabilities.
Adventa Control Technologies (Dallas, Texas), formerly the WORKS development organization at Texas Instruments (TI, Dallas, Texas), has developed ProcessWORKS, a model-based control application for advanced process control. TI was intimately involved in the CIM Framework development effort, and the WORKS MES suite has a high degree of compliance with the CIM Framework as a result.
SiView, the newest offering from IBM Semiconductor Manufacturing Solutions (Middleburg, Va.), is an MES and equipment automation product designed to be compatible with the CIM Framework and with Object Management Group (OMG, Framingham, Mass.) standards. It runs on AIX platforms for servers and Windows NT for clients, and it can operate with any CORBA v2.0 compliant ORB. Its automation interface supports connection with a number of commercially available cell control products.
Object Automation (Santa Ana, Calif.) recently introduced its OAenterprise 98 product. The company characterizes its framework as "object-based," rather than object oriented, emphasizing its goal of making configuration possible with no additional programming.
Encore! from Promis Systems (Toronto, Ontario, Canada) is an open architecture MES whose integration framework allows the user to use Windows NT clients for reporting and analysis functions and continue to use existing UNIX, OpenVMS or NT servers. It is integratable with many leading enterprise and cell control products. Its Process Composer also supports the concept of a virtual factory to account for subcontractors, vendors and sister facilities.
The AutoSched productivity family from AutoSimulations (Bountiful, Utah) provides reporting, analysis, simulation and scheduling capabilities for fab operations and controls as well as what-if analyses. It is fully integratable with the most widely used MESs.
ObjectSpace (Dallas, Texas) provides advanced process control products and services. Its products are highly compliant with SECS/GEM, APC/CIM Framework and CORBA/IIOP standards and can be scaled for use in tools or the overall fab. Along with Advanced Micro Devices (AMD, Sunnyvale, Calif.), the company was recently awarded a NIST grant to develop advanced process control technology.
PRI Automation (Billerica, Mass.) recently acquired Interval Logic (Menlo Park, Calif.). Interval Logic's Leverage suite includes Planning product, which is designed to provide fabwide scheduling quickly enough to perform what-if analyses and event driven management of fab operations.
Integration
Integration issues have been a major source of problems in MES implementations. It has historically centered on simply getting things to work, rather than tuning it to meet production and business goals. Part of the reason is that some process tools are not adequately prepared to communicate. Full generic equipment model (GEM) communication capability will be required for 300 mm equipment (Fig. 4). Another part is that integrating software components from multiple vendors has, until now, been an unwieldy task. Open systems and alliances between various vendors are starting to simplify the integration task.
|
| Fig. 4. Equipment communications have gone through awkward phases. Full SECS-II and GEM communications capability will be required for 300 mm fabs. (Source: SEMATECH) |
Conclusion
MES and CIM systems will have to provide more than just passive tracking and recommendations to make 300 mm fabs successful. They will have to provide active and automatic control of the fab, using enterprise and supply chain information as well as internal factory information.
Both fab management and software producing cultures will have to adjust to the needs of increased productivity. The MES must be treated with the same importance, if not more, as tools receive in order for it to become capable enough to help meet the goal of significantly improving OEE. The CIM system in general will have to be viewed as a system that requires continual maintenance, enhancement and investment rather than simply something that is just installed and used. They will be key in the drive to improve both productivity and flexibility.
Software and communication capability for tools must improve to serve the information needs of the fab. Full SECS-II and GEM communications capability should be considered a bare-bones minimum for 300 mm tools.
|
Concurrent Development Speeds Time to Market Bill Reaves, Cliff Smith, Intel, Folsom, Calif.
The technology required to place two bits of memory on one transistor called for significant improvements in three technology areas at the same time. Precise control of the number of electrons collected on the floating gate during a programming operation is required to keep from overshooting the desired voltage state. Stable charge storage is required to maintain the 300 mV distribution between voltage states over time. Precise charge sensing is required to distinguish between voltage states during a read operation, since the device uses ~5000 electrons to distinguish between states compared to the typical ~20,000 used in single bit per cell devices. Development and realization of the technology in high-volume manufacturing in a timely manner required the use of a concurrent development model. Traditional product development models align functional engineering groups to a particular phase of product development. For example, technology development and design engineering groups are tied to the research and development phase, while manufacturing, reliability and test engineering groups are tied to the factory integration phase. Best known methods and incremental innovation are used in this model to bring new products to market within 18 months of conception. The drawbacks of this approach are very little interaction between the different engineering functions and poor communication of the needs of "downstream" functions to the engineers working on the "upstream" phases. This "pass the baton" approach usually leads to manufacturing delays when introducing new technologies into new products.
For the StrataFlash product, a cross functional team was assembled to ensure that issues with the downstream phases (silicon validation, qualification and high-volume manufacturing) were identified and addressed in the upstream phases (R&D and product design) of product development (Fig. a). Although many of the traditional one bit per cell development methods and tools could be leveraged, new methods and tools had to be developed to debug the two bit per cell technical requirements within the confines of the schedule. By bringing factory integration issues into the development process at the beginning, the end product and the manufacturing process did not have to be reworked in the downstream phases to meet customer reliability requirements. The final product hit the rigorous qualification schedule with no retrofits to the product design or the manufacturing process. |
In 1992,
Intel engineers started a development program, called
StrataFlash memory, to help drive memory to higher densities
and into new markets like cellular phones and digital cameras
at an accelerated pace. The development of StrataFlash memory
was a technically challenging task that required a new product
development approach.
|
Scheduling Can Be Better Without Following the Rules Donald Rosenthal, Interval Logic Corp., Menlo Park, Calif.
Most fabs today operate with dispatching, rather than scheduling programs. Simulation tools are used to help analyze the fab and test possible dispatch rules. Simulation is a very useful tool for designing the layout of a fab and for predicting the behavior of process tools. However, using rules-based dispatch for managing fab operations is a bit like trying to play chess while looking ahead only a single move. What may appear to be the best move if the look-ahead is limited to a single turn, might actually turn out to be a trap if the player looked far enough in advance to see the full consequences of making that move. In the same way, using rules-based dispatch to sort lots in queue might give the appearance that it is correctly prioritizing lots at a process tool, but it does not look far enough in advance. The decision-making capability is limited by an inability to review data for the full fab and its inability to understand the follow-on consequences of the choice being made. To schedule and optimize a fab a full shift or more into the future, manufacturers need a scheduler that utilizes a comprehensive view of the entire fab, generates equipment reservations at specific times for specific tasks and creates full time-tagged itineraries for each lot for the entire manufacturing process. This type of scheduler would start by placing all the tasks onto a timeline, temporarily ignoring constraints and then "repairing" away constraint violations to generate an optimized schedule. Since this approach can generate a schedule in a short time, the schedule can be redone whenever a change, such as the change in status of a tool or a lot, occurs in factory operations. By optimizing the schedule for the fab as a whole, there are no surprises at one point because of an inappropriate choice at another point; manufacturers do not fall into a "trap." Decisions are not made based on data that are either localized in place (only one tool) or in time (only what is currently in queue). By focusing on the entire fab rather than the individual tools, productivity at both the fab and tool levels can be optimized. |
As the
semiconductor industry transitions to 300 mm processing, it
will be important for manufacturers to utilize process tools
more efficiently and make fab operations more flexible.
Meeting these conflicting requirements will require a new,
comprehensive approach to scheduling and managing fab
operations that includes the ability to schedule, in advance,
down to individual tool load and unload operations.
|
The Computer Integrated Manufacturing (CIM) Framework developed by SEMATECH has gone to blue ballot at SEMI. There is a push to get CIM standards through yellow ballot and in place before 300 mm fab implementation. One of the potential snags in the process is that the CIM Framework specifies the CORBA object request broker (ORB) as the backbone for fab CIM systems, and some would prefer to use the COM/DCOM object manager that comes with Windows NT. They would prefer that the standards be more general about the ORB, and not require CORBA. CORBA stands for Common Object Request Broker Architecture, and it has been developed by the Object Management Group (OMG, Framingham, Mass.), a consortium of more than 800 companies. One of the reasons that CORBA was chosen by SEMATECH is cross platform interoperability. Interfaces between it and other systems is done using an interface definition language (IDL). Currently defined interfaces with CORBA include COM, OLE Automation, C, C++, Smalltalk, COBOL, Ada and Java. On the subject of SEMATECH specifying CORBA, Richard Soley, CEO of OMG, added, "SEMATECH just wants something that runs on everything, because while obviously lots of people are going to NT, and I would be the last one to try to stop them, they want it to run on other things as well. CORBA gives them the interoperability across platforms." Soley also noted that Windows NT is the No. 1 deployment platform for CORBA. Windows NT is not the only system that MES suppliers use. Camstar Systems (Campbell, Calif.), for example, bases MESA on IBM'sAS/400 system. Consilium's (Mountain View, Calif.) WorkStream DFS uses a UNIX based system, but its plans for Fab300 are to put it on Windows NT. The main reasons for not wanting to specify CORBA, though, are the short-term cost savings and convenience of using DCOM. As for concerns about the maturity of CORBA, work on it has been done for seven years. In fact, CORBA's interface with COM was defined before DCOM's beta release. According to Soley, "I know it sounds trite, but it's actually just plain old true: CORBA runs on more Microsoft platforms than COM. To be more specific, it runs on all current Microsoft platforms." It also runs on most major mainframe platforms, all UNIX platforms, midrange platforms such as AS/400 and VAX VMS and on several real time platforms. CORBA is currently deployed in about 50,000 to 60,000 applications, and there are 100 success stories about it available from the web site www.corba.org. The stories include point of entry control in the UK and CNN Interactive. TYECIN Systems (Los Altos, Calif.) is using a CORBA implementation from Expersoft (San Diego, Calif.). AutoSoft's (Salt Lake City, Utah) iMaven product is also CORBA compliant. National Semiconductor's fab in Greenock, Scotland, is using Orbix from IONA Technologies (Dublin, Ireland) to tie its legacy systems together. Cross-platform capability may not be necessary for a new installation. However, integration with legacy systems may require a very reliable and standard architecture designed for interoperability |
References
2. MES Explained: A High Level Vision, MESA International White Paper Number 6, September 1997.
3. I300I Factory Guidelines: Version 2.0, International 300 mm Initiative, 1997.
