statistical machine control

Upload: cmfowler06

Post on 03-Jun-2018

222 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/12/2019 Statistical Machine Control

    1/7

    Statistical Machine Control: A Practical Approachto Total Productive Maintenance of Semiconductor Equipment

    Frank LavaUart, Equipment and SMC Engineer - Wafer Fab Operations.Ned Cooper, Manager of E,quipment Engineering and Maintenance - Wafer Fab Operations.

    Medtronic Micro-Re1 - 2343 W. 10th Place Tempe, Arizona 85281fiank lavallart@medtronic corri or (602) 929 54 13, [email protected] or (602) 92 1 6489

    AbstractStatistical Machine Control (SMC) encompasses

    the total maintenance effect on equipment. Thisapproach steps far beyond the ix it attitude thathas often been the understood job of a maintenancetechnician. SMC incorporates Total QualityManagement (TQM) practices, the predictive andpreventive maintenance programs. The goal of theprogram is to d e ptime while continuing toreduce variation. Variation has an effect on product

    yields, and equipment can be a major cause of thetotal variation in a factory. SMC addresses theseconcerns, and it requires a change in mind set thattakes time, training and a lot of ene:rgy to bring tofruition. SMC works if the organization iscommitted to overcome the stumbling blocks alongthe way.

    This paper is illustrated with four practicalexamples.

    Background:

    Several years ago, it became evident tomanagement that the Statistical Process Controleffort at Micro-Rel, while al encompassing byaccepted definition, was not providing a deepenough focus to minimize potenfial sources ofvariation within the equipment. The process controlcharts usually included the equipment variation, butthat variation was not in a derivative form that couldbe looked at separately fiom the process variation.In other words, the variation often included both theprocess and equipment related variations. Micro-Re1

    management felt that the Equipment Engineeringand Maintenance group could do a lot more toreduce variation within the equipment itself. Thiseffort was even more important in a custom orientedfab such as Micro-Rels fab where the processes arechanged several times a day. This makes equipmentvariation control an even more important elementsince equipment does not run a single process all the

    time making process control charts more difficult tointerpret.

    Micro-Re1 had spent a lot of time on SPCtraining throughout the factory, and that includedthe Equipment Maintenance Technicians, but thetechniques were not being utilized in maintenance toanywhere near the levels that management felt werewarranted. It became more and more apparent thatit was necessary to begin a program to emphasizethe importance of minimizing variation on the

    important equipment parameters that affect theprocess. This would allow process engineering tospend more time on the materials, time,temperature, etc. aspects of the process, havingconfidence that we (Equipment Engineering andMaintenance) had adequate controls on theequipment to minimize variation. This seemed like anoble effort and we recognized that this effortwould be a long difficult task, but even then, weunderestimated the effort required. In the firstplace, it is difficult to take a group of individuals

    whose primary training is to fix problem and tellthem that is not their primary job at all. Ourorganization, like many others, have been able to getover that hurdle with the efforts that go into a TotalProductive Maintenance program where PreventiveMaintenance (PM) and Predictive Maintenanceprogram have been put into place. The PM andPredictive Maintenance programs still fit fairly wellinto the accepted maintenance activities andpractices. The steps toward a Statistical MachineControl have been much more difficult to introduce

    as main stream maintenance. We found a lot oflistening occurred but it was tough to get theRubber to the Road as his process was outside ofthe normal comfort zone of a MaintenanceTechnician. We began to realize that it would takethe dedication and concentrated efforts of a fidl timeSMC/ equipment engineer to train, coach andmentor the program to make the conversion fiom

    0-7803-3929-0/97 8.00 01997 IEEE 81 1997 IEEWCPMTIntlElectronics ManufacturingTechnology Symposium

    mailto:[email protected]:[email protected]
  • 8/12/2019 Statistical Machine Control

    2/7

    fix it technician to SMC Technician. We presentedrogram to our director and the position was

    eventually open for that job. We then hired a anindividual to get the process started and the projectwas underway. After a couple of years, with thecompetitive environment in Phoenix, we lost theoriginal engineer to another company and he wasreplaced by Frank Lavallart. This turn out to be anexcellent choice and Frank has taken the program tothe next level as we continue to improve the SMCprogram.

    a~ntenance erving its customers:C activities form a practical response to aapproach in an equipment maintenance

    nt. A Maintenance department isdedicated to the needs of the Production andEngineering departments. Whenever equipment isbeing repaired, maintained, upgraded, etc., thetechnician delivering the service must satisfl theneeds of the people manufacturing products orimplementing processes (the customers9).

    Based on these premises, Micro-Re1 requestedthat any Continuous Improvement of themaintenance h c t io n start with a short survey of thecustomer needs. Each piece of equipmentselected for an SMC project was initially the objectof a rapid collection of information with the

    duction and Engineering users of the equipment.

    have interviewed the process engineer(s),operators, and maintenance technicians who havedealt with the equipment.

    We discovered that in some instances theproblem raised by the interviewee was actually aperceived issue on the equipment. Properinformation fiom the person who has revealed theperceived issue is quite often enough to alleviate thewhole problem. We trained the technicians not tounderestimate those issues, as they may generatefkustration without actual cause.

    However, most issues have a real foundation. Inthe event that the equipment does not influence theissue, the information is forwarded to the processengineer. In many of those cases, traditionalStatistical Process Control will allow the engineer tomonitor and put the issue under control.

    Whenever the equipment is involved in the issueraised by Production or Engineering, the

    Maintenance department deploys all the traditionalTQM tools known to circumvent the reoccmenceof the problem: pure problem solving techniques,and statistical tools. The particular application ofstatistical tools to address issues influenced orcaused by equipment is what we call StatisticalMachine Control.

    Maintena nce as a customer:Successfbl implementation of any Continuous

    Improvement effort on a piece of semi-conductormanufacturing equipment requires the knowledge ofthe equipment vendor. From the start of anyproject, we inform the vendor about the intent andthe scope of our SMC activities. For most vendors,we establish partnership meetings where we gainmore knowledge on the equipment, and offer inreturn the actual data on the equipmentperformance. The mutual benefits generated bySMC allows the technician(s) to accurately monitorthe equipment.

    The goal of SMC:The goal of our Statistical Machine Control

    program at Micro-Re1 is to reduce variation on theprocess output, and maximize uptime.

    Any equipment parameter that has a significantinfluence on the process output and that cannot becontrolled by the process recipe, must be under

    control. For instance, if a process c h b e r exhaustflow impacts the heat dispensed on a wafer, the flowmust be controlled. A close-loop system regulatingthe flow may offer a satisfactory control of thevariation.

    However, other parameters are not controlled aseasily. For instance, Mass Flow Controllers get

    flow f ... Process a mo mda tesfor degraded performance.

    I AI \ time

    I MFC perf. degrades...contaminated by process gases that flow through.When significant contamination has occurred, one

    1997 IEEUCPMT Intl ElectronicsManufacturingTechnology Symposium2

  • 8/12/2019 Statistical Machine Control

    3/7

    type of response consists of accommodating for thedegraded performance of the MFC by modifjing therecipe. Eventually, Maintenance intervenes or theMFC fails and gets overhauled. The process all of a

    ...process is out offlow A Control.

    Maintenanceoverhauls MFC ...

    time

    sudden gets out-of-control, without apparentreason.

    A carefid study of the impact of the MFC on theprocess allows the maintenance technician toperform calibration exactly when it is needed. Therecipe is never modified to accommodate anyequipment parameter, and ultimately variation of theprocess is reduced.

    Maximizing uptime is another goal of the SMCprogram, Tracking of equipment performance mustbe established. It is measured wth the number ofMures, the amount of time wiih the workingequipment, the Mean Time Between Failure, theMean Time To Repair, etc. Doing so encowagesthe technician to act upon declining trends. Aposting of those metrics allows the maintenance

    department to concentrate on the equipment thatOFF IdN EVENT PRODUCTIVE TlMEi, 1996ll997~

    5 E

    demands the most help.Moreover, the charts monitoriug equipment

    performance constitute a factual and efficient wayfor discussions with the equipment vendor. Timespent with the vendor can be dedicated to theresearch of a solution.

    Also, at any given time, one piece of equipment,or one type of equipment in a wafer fab constitutes

    bottleneck of the fab. If the productive time ofthat specific piece of equipment can be augmentedby 1%, the whole fab capacity goes up by 1% I .There are obvious benefits to knowing the level ofperformance of the critical equipment.

    A generic SMC project:To provide some guidance to the techniciansstarting an SMC project, we have defined the majormilestones of a project.*Collect the existing data and issues about the

    selected piece of equipment.*Define and prioritize among the issues (e.g. Pareto

    chart); identi@ he largest source of variation.*Meet with all the concerned parties (operators,

    process engineers, vendors, maintenancetechnician).Implement the Action Item List

    *Repeat the previous two steps until significantimprovement is recorded.

    *Act to limit variation and to avoid reoccurrence ofthe issue by utilizing SMC charts.Document the whole SMC process.Celebrate.

    We feel that any project needs to follow thesesteps, to insure quality and to allow othertechnicians to join the effort.

    Example #1: uniformity variation delivered by aRapid T hermal Annealer.

    The Rapid Thermal Processing (RTP) consists ofgrowing a thin oxide layer on the wafer by exposingthe wafer to a controlled and short duration heatingcycle.

    The RTP equipment used in this example wasAG Associates Heatpulse model 2146, installed in1989. This equipment contains dual bank parallel

    83 1997 IEEUCPMTInti Electronics ManufacturingTechnolosy symposium

  • 8/12/2019 Statistical Machine Control

    4/7

    lamps in contrast to currently available cross-lampmodels which have much higher within waferuniformity specifications.

    The equipment manufacturer AG Associates wasinvolved fiom the beginning of the SMC effort. Arapid survey of the users, Production and ProcessEngineering, and the vendor showed that uniformityof the oxide grown is the main issue.

    Oxide thickness variation occurs across thesurface of a wafer, fiom wafer to wafer within a lot,and fiom lot to lot. It appeared that the largestvariation occurred across the wafer surface.Thickness was varying by 12% or more until March1996 (well beyond the equipment capability).

    To bring the whole team together, the variationaddressed in this project was defined as The 700degrees Celsius, 1.5 microns silicide recipe does notgenerate sheet rho below 4 ohms per square at 3standard deviation.

    A cause and effect diagram was built, andsummarized the fact that the effect is caused by

    factors falling into six categories:Machine (i.e. the Rapid Thermal Annealer), Method(process or recipe), Material (the wafer itself),Measurement (the tools used to track the processoutcome), Environment (facilities parameters), andPeople (operators can have an effect on the

    .Condition of the processing quartz chamber(cleanliness, scratches, optical characteristics suchas haziness, etc.).

    a Wafer location in the chamber during processing.Gas flow rates.

    Processing chamber water cooling.Thermocouple calibration wafer.

    *Calibration method.

    Wafer supporting tray.

    Eventually, guidance ftom the equipmentmanufacturer and a careful examination of theprocess outcome, i.e. the oxide thickness controlchart, demonstrate that the pyrometer coolingwater, as well as the lamps age and placement havethe largest influence. The temperature of the

    E3

    1

    0

    RTP TOx film thicknessI

    was then controlled within 0.1 degree ofthe nominal temperature. Moreover,additional h e tuning of the heatgeneration lamps was performed by AGAssociates. As result, the variation inoxide thickness a wafer was further

    3.5 to less than 2.5 .As the temperature of the water isstaying very stable over a long period of

    m a

    /mim s

    process). Group Nominal Techniques (under the

    alleged cause of the largest variation in oxidethickness .

    The parameters suspected to have the largestinfluence are:

    t h e , installing limits with b as foundform of a vote) allowed the team to Prioritize the practical than a control ofthe temperature.

    0 Age and placement of heat generation lamps.

    pyrometer cooling water was controlledwithin 1 degree Celsius of the nomina1temperature until it was found to have a

    large influence on the measurementr delivered by the pyrometer. The water

    1997 IEEWCPMTIntlElectronics ManufacturingTechnology Symposium

  • 8/12/2019 Statistical Machine Control

    5/7

    Lamps are now replaced and tuned for p m ~ m a k e EYBOLDInspected byconsistent heat delivery on a yearly basis (until ;g;zdelBOA Date

    PROMiSC Zthen, lamps were very seldom modified). Asthe performance of each lamp degrades

    De C m dimension Replainconsistently over time, the process rankma R e mp ~ w m ' ~ Actual AciccedJ 1st pl

    uniformity degrades accordingly, but in aknown and controlled manner.

    labor and interruptions. The pump (;an continue toM f i U its h c t i o n satisfactorily until1 it reaches the

    From the beginning of this SMC project, causes

    pPoverhaul threshold .

    The remaining parameters appeared well incontrol. In some cases, they have insignificantor no influence on the process outcome. Thisallows the maintenance department to abandonunnecessary maintenance and cost.Example 2: cost-effective overhaul ofvacuum pumps.

    Vacuum pumps used to evacuateprocessing chambers are subject to the normalwear due to their technological limitation, aswell as the contamination and corrosioncaused by process gases.

    As a consequence, regular overhaul arenecessary, to replace the worn or corrodedparts, and remove any trace of con1ta nts. ~ n o s l o n ~ L i ~ c o n p o n e n b m a t o w c M m s l o n

    32 I

    3 4Due to the extent of the work, overhauls are 3 3

    maintenance of spares to replace a suddenly

    5 1

    disruptive, costly, and ck l the :5 L& Ohem M ,C 0 en son, %&

    hiling pump.

    the overhaul, and augments considerably the risk of the overhaul dimension or i the dimension causingcatastrophic failure. The cost of ownership of suchpieces of equipment rise once the overhaulthreshold has been passed.

    1 5 1 I I

    as new dim - current dimas new dim overhaul dim

    Current ranking = 5 x

    Late OVerhads Of the Pumps increase the Of ifthe considered current dimension has not reached

    the is not known, and

    /f performance degradation were classified into the 5

    three categories: contamination, corrosion, andwear.

    Wear, corrosion, and contamination are trackedon a scale fiom 0 to 10, O'being as new condition,overhaul threshold, and 10 condition causing failureof the whole pump.

    part is measured for wear, and inspected for

    0

    asn ew current over' l failure

    I

    Whenever a pump is overhauled, each critical dim dim dim dim

    contamination and corrosion. Data is entered in aspreadsheet. Raw dimensional data is entered, and a

    overhaul dim - urrent dimoverhaul dim - failure dim

    urrent ranking = 5 5 x

    ranking is automatically computed as:

    8 1997 IEEWCPMTInt l Electronics ManufacturingTechnology Symposium

  • 8/12/2019 Statistical Machine Control

    6/7

    if the considered current dimension has gone beyondthe overhaul dimension and the failure dimension isknown.

    Surface roughness is evaluated with a physicalsurface roughness template, and a measurement inmicroinches. Ranking is computed in the samemanner as the above mentioned dimensions.

    Corrosion and contamination are visuallyinspected, and ranked using a document withdescriptions of typical conditions.

    A general ranking is then given to the pump,defined as the maximum value among all therankings.

    An empirical relation based on type of process,model of pumps, and time elapsed determines theranking, and the optimal time to overhaul.

    xample #3: timely sc eduling of an etcher

    Etcher equipment requires maintenance based onusage. Performing maintenance on an etcherinterrupts production for several hours, as pumpingthe chamber down to a high level of vacuumrequires several hours.

    Maintenance is scheduled by performing a linearregression on the last ten wafer count. A tentativedate is computed, and gains accuracy as the taskgets closer.

    rku I

    libration of Mass Flowequipment.

    Plasma Enhanced Chemical Vapor Deposition(PECVD) equipment processes wafers with theusage of gases. The delivery of the exact quantityof gas onto the wafer affects the thickness of thesurface layer. The Mass Flow Controllers (MFC)

    86

    are the key instruments to measure the proper flow

    The PECVD equipment has a built-in procedureto self calibrate the MFC. The processing chamberbeing evacuated is monitored by the equipment forpressure versus time elapsed.volume is fixed and known, the flow of gas can berelated to the pressure, and time. Eventually, a flowcan be computed.

    The result of the measurement ought to bevalidated for Repeatability (Reproducibility is not anissue: automatic procedure without any operatorintervention).

    Concurrent calibration of the MFCs with agolden standard allows the maintenance technicianto evaluate the offset existing on the built-inmethod. Classic gage Repeatability andReproducibility shows the accuracy of themeasurement, and allows the technician to takeaction only for relevant measurements.

    The results of this autocalibration procedure areentered into a Control Chart X bar R chart).Control limits are calculated. Subsequent

    of gas.

    PECVD PROCESS GAS MFC AVERAGE

    247t4624544

    /11/97 3/31/97 WO197 5/10/97 5/30197 6/19/97

    PECVD PROCESS GAS MFC RANGE

    8

    i

    3/11/97 3/31/97 WOB7 5/10/97 5/30/97 6/19/97

    calibration results are added to the chart.Traditional SPC rules are applied to decide when itis relevant to remove the MFC for overhaul andcalibration.

    1997 IEEUCPMTIntlElectronics ManufacturingTechnology Symposium

  • 8/12/2019 Statistical Machine Control

    7/7

    Status of the SMC program, as o f June 1997Early in the program, Preventive Maintenance

    was the focus to establish control of the equipment.Predictive maintenance then followed.

    Resolution of variation issues (such as example1 above) with the use of classic TQM tools such as

    Pareto charts has been in place for two years and isnow a mature process. The statistical tracking ofequipment parameters is at an early stage, but isgaining momentum. We feel however that this keyelement in the SMC program will mature over thenext 1 or 2 years as it spreads to all the keyequipment.

    References1. Goldratt, Eliyahu M., The Goal : A Process of

    2. The Memory Jogger, second ed., GoaVQPCOngoing Improvement, North River Press

    (Methuen, MA, 1988), pp.70-71

    ConclusionsStatistical Machine Control at Micro-Re1

    encompasses a wide variety of techniques all relyingon the proactive analysis of equipment performance.As this program gains seniority and spreads in theorganization, each key piece of equipment shouldeventually be monitored with a set of ControlCharts, tracking the two or three most criticalparameters influencing process outcome and uptime.This augmented vision of the equipmentperformance is the determining factor in improvingequipment throughput and reducing semiconductorgeometries.

    This is the future as we envision it at Micro-Rel.

    Efforts to use Statistical Machine Control to reducevariation in equipment will continue: to be expandeduntil all equipment is covered by the SMC process.

    AcknowledgmentsWe would like to recognize the work of Bob

    Hanson on the Rapid Thermal Processor, Karl Piferon the vacuum pumps, Robert Scott on the etchers,and Chuck Hall on the MFC calibration. The hardwork of those individuals, as; well as the

    collaboration of the rest of the maintenance teamhas rendered the implementation of the SMCprogram possible.

    We would like to thank AG Associates, Inc. fortheir valuable insight on Rapid Thermal Annealingequipment.

    1997 IEEUCPMTlnri Electronics Manufacturing Technology Symposium7