Using GEM to measure equipment utilization
06/01/1998
Using GEM to measure equipment utilization
Jerry Secrest, Secrest Research, Portola Valley, California
Relatively few equipment engineers know the advantages of the "process state model" in Semi`s generic equipment model (GEM). In all equipment shipped with this compliance, GEM provides a straightforward method for measuring equipment utilization with little more than a spreadsheet. An engineer who fully understands his equipment`s process state model, and its relationship to Semi`s E10 reliability, availability, and maintainability definitions, can use the method described here to hone the throughput of wafer fab equipment and modules.
Monitoring equipment utilization is crucially important because it can measure bottlenecks that limit factory output. Opening these bottlenecks means more product can be made from the same equipment base. This defers the purchase of additional equipment and increases sales from the present factory. Since the majority of semiconductor product cost is in equipment amortization, any increase in product output from the same equipment base results in higher profitability. While this scenario is rooted in equipment utilization, many equipment engineers often perceive its accurate measurement as difficult and costly.
Fortunately, today`s level of semiconductor equipment communication enables a straightforward method to provide a robust and accurate measure of equipment utilization, using a cell or area host-computer spreadsheet and signals defined by Semi equipment communications standards, specifically the GEM, Semi Standard E30. This GEM-to-spreadsheet method is applicable to both front-end and back-end equipment installations.
GEM is widely known and implemented, but the emphasis has focused on getting the right recipe to equipment and on collecting metrology data. Many process and equipment engineers simply do not know about the GEM "process state model" and the messages GEM sends about equipment activity. Of course, complex factory-wide manufacturing execution system software packages can include modules that monitor and control operations related to equipment utilization. But the method discussed here is inherent in GEM-compatible equipment as it is shipped.
 |
Figure 1. GEM-standard equipment data can input to a host-computer spreadsheet to provide valuable information on equipment utilization.
When this method is properly implemented and used, equipment engineers have a clean output of equipment utilization. For example, a simple pie chart (Fig. 1) can summarize data from one piece of equipment over an extended production period or data from many pieces of equipment of the same class, such as one model of wire bonder. (Note that the host-computer spreadsheet will need to be configured and separately set up to interface with different equipment types, such as ion implanters from different suppliers.)
Engineers can use spreadsheet output to determine equipment utilization in percent - emphasizing the percent of time spent in nonproductive equipment states. This determines where process or factory engineering work can be shifted for a higher percentage of productive or process-execute time. While pie charts are intuitively helpful, results could also be the output to a Pareto analysis to determine the sources and priorities of problems, or a stacked bar chart that shows how equipment utilization changes or improves over time.
Measurement tools
The equipment interface (EI) on a factory computer system provides communications and logic functions to communicate from the host computer to a particular piece of equipment. It also parses Semiconductor Equipment Communications Standard (SECS) messages to extract data and acts based on message content coming from equipment.
Semi standard E10, titled "Reliability, Availability, and Maintainability," provides definitions for measuring equipment utilization; think of these definitions as "time bins."
Simply described (Fig. 2), the signals from GEM and definitions from E10 describe the many events of a given process sequence that are picked up by a host computer. When logged into a spreadsheet, equipment event times provide an automatic measure of equipment utilization. In addition, with this method, automatic logging of alarms provided by the Semi equipment software standards can be used to help diagnose sources of equipment utilization loss.
 |
Figure 2. Simple schematic of a semiconductor manufacturing equipment interface necessary to yield equipment utilization data via GEM.
GEM requires a "process state model" for each piece of manufacturing equipment; this model describes a specific system`s activities, such as idle, set-up, ready, etc. (Fig. 3a). Further, GEM outlines the sequence of "event messages" (i.e., numbers 1, 2, 3, etc. in Fig. 3a) that occur while processing a single wafer, a lot of wafers, a tray of leadframes, etc. Within GEM, the event message sequence titled "Stream 6 Function 11" can be tapped to provide inputs to the host computer`s spreadsheet, thus providing times for transitions between GEM process states (see table). GEM also supplies "alarm messages" that can contain information on why equipment pauses or aborts processing.
"Terminal services" messages in GEM, which are initiated at equipment by an operator, can be used to tell the factory computer system to send and receive text messages. As such, terminal services may be useful when the operator or the factory computer system knows the condition or use of the equipment or the equipment needs to be put in a down state.
The E10 standard defines metrics for six basic equipment time bins: productive, standby, engineering, scheduled downtime, unscheduled downtime, and nonscheduled time. The equipment itself, via GEM activity messages, only knows whether it is in standby, productive, or unscheduled downtime because of a fault (Fig. 3b). For the other time bins, the equipment does not know how to define its state or condition. In these cases, either the equipment operator or the factory host computer must define the correct time bin to be filled.
The host computer spreadsheet provides a format for logging event and time data during a manufacturing period defined by an equipment engineer. Setting up spreadsheet input is the key to success in using the automatic GEM-generated method of equipment utilization logging. Basically, the spreadsheet must be configured to match the process state model of a specific piece of equipment being measured, or a group of equipment in the same class, such as bonders in assembly or integrated lithography systems in wafer fabrication.
Further, by writing a spreadsheet macro, an equipment engineer can determine the amount of activity to store defined by start and stop times, for example, a specified amount of production, a shift, a day, a week. For example, a macro could add a row to a spreadsheet at each transition from execute to idle, terminating data collection at a given time or date.
With the stored data, the spreadsheet calculates equipment utilization and time spent in nonproductive equipment activities, such as idle and set-up. As noted above, the calculation results can be displayed in pie charts, stacked bar charts, Pareto analyses, or other metrics of interest.
Using the tools
The objective in measuring equipment utilization is to determine how much time the equipment is spending in each process state. To do this, the host computer uses event messages coming from the equipment to signal times going into and out of process states. For example, in Figure 3a, the transition into set-up is message one, and the transition out of set-up into ready is message two. The time difference between message one and two is the time in the set-up process state.
E10 provides various definitions:
Productive equipment time is the difference between messages one and four.
Unscheduled downtime includes activities of unplanned maintenance and aborts.
Standby is equipment waiting for work to enter the productive state.
Idle is equipment waiting for work and recipe instructions.
 |
Figure 3. The relationships between a) GEM`s "process states" and b) E10`s "time bins."
 |
Set-up of the spreadsheet within the host computer depends on the process state model of the equipment, the detail of the data collection, and user preference, which requires a number of considerations:
Different equipment classes can have different process state models and states that are finely or grossly defined. For example, a cluster tool doing a plasma clean followed by a CVD step may have a process state model that describes parallel wafer processing steps. Or, a process state model for a die bonder may describe activities that relate to individual die sites in a leadframe.
When setting up a spreadsheet for tracking equipment utilization, equipment engineers may be interested in only one portion of equipment activity - one section of the defined GEM process state model - for example, when measuring handling time of leadframes in a die bonder or when tracking time spent locating defects in an inspection tool.
Other measurement projects can be initiated to look at change over time with equipment, such as implanters, used for several processing steps in the overall fab process. Similarly, the interest could be processing rate differences caused by product variation at chemical mechanical polish.
The quantity of lots used to record data in the spreadsheet can vary. For example, a spreadsheet user could decide to make the spreadsheet longer than the expected number of lots to be processed in the period that the spreadsheet is open. This allows a simple spreadsheet with no decisions to be made during data collection or calculation. Time in a GEM process state is found by summing a column from the top of the spreadsheet to the bottom. An alternate approach is to write macros that expand the rows in the spreadsheet as more lot activity is recorded and to detect column length for addition.
The start and end times of a data collection period need to be defined for the spreadsheet. The start and end could be based on a production shift, a number of lots, a week, or other manufacturing signals.
Columns can be added to the spreadsheet to collect information related to equipment operation, for example, equipment number and location, operator identification, and process program identification. (Major sections in an equipment utilization spreadsheet are the time logs, bin time, and related data.) The related information provides a means to look for relationships that affect utilization, and answer questions such as, "Is utilization a function of where the equipment is relative to the WIP stocker?"
GEM equipment process states and E10 equipment-use definitions do not exactly correlate to one another. However, the E10 "productive" time, for example, is about equal to the sum of time spent in "set-up," "ready," and "execute." Time spent in the GEM pause state is a part of unscheduled downtime. And time spent in the GEM idle state is similar to the time spent in E10`s standby bin.
The caution here is that some but not all of E10 timekeeping can be done by tracking process-state-model activity. In addition, the equipment could be doing an engineering function and still send out productive event messages with time stamps that are not actually productive time. Accordingly, a user could set up two spreadsheets: one for measuring utilization, the other for measuring E10 time bins.
The utilization measurement
Utilization is the time a piece of equipment spends in productive states (i.e., doing its intended function) divided by the total time in a period. For example, if the time period is 24 hr, the utilization is the sum of the time spent in set-up, ready, and execute states divided by 24. For a spreadsheet set-up to take in activity over a 24-hr period, the utilization for that period is the sum of the productive time divided by 24.
Calculation of utilization based on time derived from event messages is straightforward if there are no pauses caused by operator assists, alarms, or interrupts. As can be seen in Fig. 3, if the machine pauses during execute, and execute time is calculated based on time stamps three and four, the time in pause must be subtracted from execute time by computing the time difference between events six and five.
Multiple pauses can occur during the processing of a lot. This can be handled by computing the time in and out of pause each time an event six is seen and adding that to the pause time in the computed time section of the spreadsheet. When an event four is seen, the correct time in execute can be computed from the time in execute minus the sum of the pause times.
The sum of pause times may be added to the tabulation of unscheduled downtime on more complex spreadsheets used to collect E10 times in parallel. In addition, the operator can enter data into the equipment terminal to track the time the equipment is in unscheduled downtime.
The information in an alarm message is more valuable for diagnostic purposes if the alarm message can be associated with an equipment state, which can be accomplished by logging alarms with state data. This information can be put in another area of the spreadsheet or a separate spreadsheet.
Opening and closing data collection
To keep data clean in the spreadsheet, it may be necessary to exclude some data. For example, when equipment communications comes on-line, the equipment may be in any state. This can be handled by only initiating a new row in the spreadsheet when event four is seen.
Set-up must also consider starting and ending the spreadsheet data set. For example, suppose processing of a wafer lot bridges the closing of one spreadsheet and the beginning of another. A simple set-up here is to delete any row that has incomplete data.
 |
Figure 4. A comparison of a) E10 time bins and b) ARAMS` greater details of defined states and substates.
Practical application: Modular equipment utilization
"Bottleneck," the location in the manufacturing flow with minimum capacity, and "parallel processing," where equipment or modules divide up work on the same process step, are two interesting areas where utilization measurement can be used to examine equipment utilization.
Die bond through molding where a transport system connects several pieces of equipment is a common equipment modular set-up in the back-end of semiconductor manufacturing. Conceivably, such a module could be purposely configured such that die bonding is a bottleneck; this insures that queues of work in progress do not build in the system due to restricted downstream capacity. Downstream, wire bonders are operated in parallel and are most likely the next most highly utilized, compared to die bonders, in a modular assembly system. An equipment engineer can use equipment utilization measurement to study the possibility of increasing wire bonder utilization. Then the die bonders` rate can be increased to improve utilization of the whole system.
The spreadsheet timekeeper and time computer can be used to track wire bonders by dividing the sum of calculated time in a process state for all the wire bonders operated in parallel, and dividing by the number of wire bonders. This gives average utilization and other times that an equipment engineer can use to determine what to work on to improve the system. In addition, an engineer can sort the spreadsheet by equipment identification to examine the utilization of each bonder.
The same approach to measuring parallel tools can be applied to resist-track and cluster tools, such as thin-film deposition systems.
It is typical to design fabs with lithography as the bottleneck control on capacity. A utilization measuring system on lithography that continually produces reports and supplies loss-cause information could assist in increasing fab output. The leverage provided by such utilization tracking can be quite significant; consider that a 5% increase in output from a billion dollar wafer fabrication facility might mean $50-100 million in yearly semiconductor sales.
Limitations
Spreadsheet data storage and calculation can run out of capability when the amount of data and the rules governing the data become large enough that the user becomes unable to develop and maintain them. This can be avoided by dividing tasks into different segments. For example, one such division method could preprocess incoming data by separating all normal state sequences from pause, abort, or interrupt process-state sequences. Then, a properly set-up spreadsheet would have separate rules for handling the two separate groups of sequences.
The future with ARAMS
While Semi`s Automated Reliability and Maintainability Standard (ARAMS, Semi Standard E58) is not yet widely implemented on semiconductor equipment (see "Pros and cons of ARAMS" on p. 200), in the future it may enable collection of a higher level of detail on equipment activities and conditions. ARAMS provides methods for collecting equipment symptoms and defines equipment E10 time bins in greater detail (so-called substates, Fig. 4). For example, E10`s productive state is redefined into substates of regular production, work for third parties, rework, or engineering runs.
When monitoring equipment utilization, ARAMS will supply a structure for adding more detail about equipment activity and conditions. In addition to E10 substates, it provides for:
operator or factory control of states and substates,
operator prompts for symptoms in states (i.e., an operator`s observation),
accumulators onboard equipment for recording time, and
transfer of accumulator data in table format.
While ARAMS will add a dimension to measuring equipment utilization, the level of detail available will require an expanded spreadsheet.
Conclusion
GEM`s mapping of equipment process states and E10 definitions provide a low cost method for measuring equipment utilization on a host computer`s spreadsheet. This method is readily available on wafer fabrication and assembly equipment that is GEM and SECS compliant. While this spreadsheet method is straightforward, its implementation requires knowledge of the process states defined for each specific model of an equipment set. As a user becomes competent with the spreadsheet method, he or she can add more details on equipment activity that assist in diagnosing sources of lost production time. n
Acknowledgments
The author thanks Bill Park for discussions on approaches to using spreadsheets, and Cynthia Neufdoerffer of the Peer Group for material on the host equipment interface.
Jerry Secrest received his BA degree in physics. He provides technical consulting services under the name of Secrest Research, including education in SECS and GEM, design of GEM implementation for factories, and data-based modeling of semiconductor manufacturing. Prior to consulting, he worked in manufacturing and engineering of ICs; he has 32 years of experience in the semiconductor industry. Secrest Research, 250 Willowbrook Dr., Portola Valley, CA; ph 415/851-8142, fax 415/851-8142.