International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056 Volume: 08 Issue: 12 | Dec 2021 www.irjet.net p ISSN: 2395 0072 © 2021, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page1368
Key Words: Clustering,Semiconductor,WaferFabrication,LoadLock,CycleTime,Simulation
Abstract
This paper describes the concept of clustering which plays a significant role in semiconductor manufacturing plants. In this project, cluster tools have been used for wafer fabrication. Cluster tool is like a machine with number of processing chambers, a handler and number of Load locks. A lot is a box filled with wafers and it is to be loaded in the load lock, ready for processing.
1
Mostmodernclustertoolshave2loadlocks.Eachloadlockcanbeloadedwiththelot.Thecontainerthathaswafersisalot. Then,theclustertoolprocessesthelot.Thewaferinthelotisscheduledinsidetheclustertool.Theschedulerofthetoolcarries outthiswork.Thehandlersareliketransferrobotthatareusedformovingthewafersbetweenthechambersandloadlocks. Thechambersaremachinestoprocessthewafers.Sincetheprocessingofthewafersispipelinedthecycletimeofthelotin clustertoolisreducedwhencomparedwiththecycletimeofalotusingtheconsecutivemachines. Theinconveniencewithclustertoolsisthattheirbehaviorismorecomplexthanthebehaviorofsimplermachines.Sincethe clustertoolisoperatedinparallelmode,itisabletoprocesslotsinparallelthatsharethesameresources.Thecycletimeofa lot is not constant but depends on the situation inside the cluster tool during the processing of the lot. When a machine processesonlyonelot,thecycletimeissimplydeterminedbyaconstantorbyasinglerandomvariable.Whenaclustertool processeslotsinparallel,thisismorecomplex.Eachlotoverlapswithotherlots.Duringtheoverlaps,thelotssharethesame resources and the lot cycle time depends considerably on the lot combinations inside the cluster tool. The shared use of resourcesslowsdowntheprocessingofthelots. Nooverlap LotA LotB PC3 PC2 PC5PC1 PC4 LL2 LL1
Cluster tools can be operated in parallel mode. Load lock consists of lots. When lots are undergoing a process in processing chambers different kinds of lot combinations may be possible. There is a dramatic reduction in speed when lots are processed in parallel. The fact that the lots are utilizing the same resources does not help the cause either. In order to overcome this problem, lots are to be scheduled in such a way that scheduled lot combinations do not affect each other. The cycle time of the individual lots and lot combinations would be computed. Since cycle time cannot be computed during the time of process, simulation is adopted.
Usage of Genetic algorithms plays a key role in determining the optimal sequence of lots
Clustertoolsaremachinesthatcombinevariousprocessingsequenceinonemachine.Theyconsistofloadlocks,handlersand othermachinestoprocessthewafersofthelotsintheloadlocks.
Figure 1. Cluster tool model
1.INTRODUCTION
Improvement of Cluster Tool Performance Using Simulation Khalid Ahmed Ibrahim 1Karary University, Associate Professor, Faculty of Computer Science and Information Technology *** -
1.1. Single Mode
Definition2.Cycletime of the lot Thetimerequiredbythelottocompletetheprocess.
Figure 2. Various overlapping scenarios
International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056 Volume: 08 Issue: 12 | Dec 2021 www.irjet.net p ISSN: 2395 0072 © 2021, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page1369 Partial LotA overlap LotB Complete LotA overlap LotB
Theslow downfactoroflotAwhileprocessedinparallelwithlotBisdefinedas SDF(A,A+B)= CT(A,A+B)CT(A) where CT(A,A+B) isthecycletimeoflotAwhenitisprocessedtogetherwithlotBand CT(A) isthecycletimeoflotAwhenitis processedalone(singlemode).Forsimplificationofthedefinition,anassumptioncanbemadethatlotBislargeenoughsothat lotAisprocessedinparallelwithlotBforitscompletetoolcycletime.Whentheslow downfactorsforalllotcombinationsis computed.Thenthosevaluesandthetoolcycletimesinsinglemodeisusedtoapproximatethetoolcycletimesofthelots.To computethetoolcycletimes,startwiththefirstoverlapandusetheslow downfactorsofthelotstodeterminethe lengthof theoverlapandtheworkthatisdoneforeachlot.Thisprocessisrepeateduntilalllotsarecompleted.
Mostclustertoolscanoperateinsinglemodeasitofferslowesttoolcycletimes.Accordingtothismode,itdoesnotsupportthe processingofmorethanonelotatatime.
Definition1. Slow down Factor
3. Algorithm INPUT:Queuewithlotsintheordergivenbytheschedule,Time WHILEQueue.notEmpty()andClustertool.notEmpty() DOFORallemptyloadlocksDO IF THENQueue.nextLotReady(Time)AddQ.next()toloadlockandset itsstarttime.
1.2. Parallel Mode
ENDFORENDIF
Inparallelmode,aclustertoolislikeasetofparallelmachines.Eachloadlockisidenticaltoonemachine.Thedifferencewith realparallelmachinesisthatthese“pseudo parallel”machinessharethesameresources.Thus,thereisaninfluenceofone machineontheotherandviceversa.Thefactisthatthetoolcycletimesoflotsarecorrelated.Asaconsequence,alloverlapsof lotsinsidetheclustertoolcontributetothetoolcycletime. So,it isnot possible tocompute toolcycletimesin parallel mode directlywithout computingthetoolcycletimesfor the completescenario.Thiscanbedoneviasimulation.Thesimulationapproachisusedtospeedupthecomputationofthetool cycletimewithanapproximation.
schedules
number of process chambers, load locks and handlers. The
thatcanbe
International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056 Volume: 08 Issue: 12 | Dec 2021 www.irjet.net p ISSN: 2395 0072 © 2021, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page1370 LengthOfOverlap=TimeToNextEvent(all lotsin clustertool); Time=Time+LengthOfOverlap FORalllotsinclustertoolDO Determinetheamountofwork doneforlotduringtheoverlap. IFlotiscompleted THENRemovelotfromloadlockandset itscompletiontime. ENDWHILEENDFOR 4. Cluster Tool Simulation Simulationisusedfortwo purposes.Thefirstoneisto determineslow downfactorsforallcombinationsoflots.Theslow downfactormayvarydependingon variousfactorslikethemannerinwhichlotsoverlapandtheorderofprocessingoflots
Thesecondpurposeofsimulationistoevaluatethe computedbythedifferentoptimization Simulation Engine
arelistedinTable1. Table 1 Model Parameters Clustertools NumberofProcesschambers NumberofLoadlocks Numberofhandlers Handlers Movetimeforeveryorigin destinationpair Sequences Numberofprocesssteps Processsteps Processchambersqualifiedforastep Processtimeinthesechambers Lots Numberofwafersperlot Wafers Processsequence
approaches 5.
Asimulationenginecanbedevelopedforclustertools.Thesimulationmodelcanconsistsofanarbitrarynumberofcluster tools, each of which can have an arbitrary most important parameters specifiedinthesimulationmodel
International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056 Volume: 08 Issue: 12 | Dec 2021 www.irjet.net p ISSN: 2395 0072 © 2021, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page1371
7. Optimization of Sequences of Lots using Genetic Algorithm
Anapproachispresentedthatusesasimulationforfindingcycletimeofthelots.Ithasbeenfoundthatthereisavastamount ofreductionintimeascomparedtoconventionalapproaches. Ageneticalgorithmtofindoptimalsequencesoflotsatthe clustertool.Insteadoftestingallsequences,theGAneedtotestonlyselectedsequencestofinditsbestsolution.Finally,the numberofsequencesgeneratedtofindtheoptimumandtheruntimeoftheGAcanbedisplayed.
8. Conclusions
Thesimulationenginecomputesanumberofoutputstatisticsafterasimulationrun,suchascycletimesofwafersandutilization ofthecomponentsofaclustertool.
Toidentifycombinationsoftwolotsthatleadtolowlotcycletimeswhenprocessedinparallel,simulationexperimentcanbe done,usingtheclustertoolmodel.
Aheuristicmethodbasedonageneticalgorithm(GA)hasbeenproposed togeneratethelotsequences.Thebasicideabehinda GAistoimitateanevolutionaryprocess:Thebestindividualsor genomes ofagenerationsurviveandreproducetopasson theirgeneticmaterialtothenextgeneration.Roughlyspoken,theGAneedsthreepiecesofinputdata:Adatastructureto represent the genomes, operators on this data structure that allow the genetic algorithm to create new solutions and an objectivefunctiontoevaluatethe fitness ofagenome.
Startingwithaninitiallyemptyclustertool,putalotofprocesssequencei {1,….,4}inloadlock1andalotofprocesssequence j {1,….,4}inloadlock2simultaneously.Forallthepossiblecombinationofsequences,measurethecycletimeforbothlots.For everycombinationi,j,theratios Ti,j Tiiscomputed,whereTi,jisthecycletimeofalotofsequenceiwhenprocessedtogetherwithalotofsequencej.Tiisthe cycletimeofalotofsequenceiwhenprocessedexclusively.
6. Combination of the Lots
Sampletestresult Obviously, there will be combinations that lead to more favorable cycle times for both lots than other combinations. For example,combiningsequence2and3leadstoanincreaseincycletimeof29%forsequence2andof38%forsequence3. Hence,thisisamorefavorablecombinationthan,e.g.,sequence1and3,whichleadstoanincreaseincycletimeof85%for sequence1andof50%forsequence3.Combiningprocesssequences2and3leadstoshortercycletimes.
Table 2. Cycle Time ratios of lot combination Table2canbeusedasaguidelineforoperatorstochoosecombinationsofprocesssequencesthatleadtoshortcycletimes. However,whenalargenumberoflotshastobesequencedonseveralclustertools,thistaskcannolongerbeperformed manually.
ThedefaultoperatorscanbeusedonlistsandarraysthatareimplementedanddocumentedintheGAliblibrarytogenerate newgenomes.Astheobjectivefunction,thetimerequiredforprocessingalllotsaccordingtothesequenceisusedastheGA suggests.Thistimeisderivedusingthesimulationmodelofthecluster tools.TheGAusesthisobjectivefunctiontoevaluatea genomeandtodecidewhetheritis"fit"enoughtosurviveandreproduce.
SEQUENCES j=1 j=2 j=3 j=4 i =1 1.86 1.85 1.85 1.35 i =2 1.66 1.78 1.29 1.34 i =3 1.50 1.38 1.88 1.39 i =4 1.25 1.53 1.76 1.72
Sequencescanbere optimizedinonlyafewminutesifthesetoflotswaitingforprocessingchanges.Anotheradvantageofthe geneticalgorithmisthatitisan anytime algorithm.Thismeans,thatifaresultisneededbeforethegeneticalgorithmhas terminated,thecomputationcanbestoppedandthegeneticalgorithmwillrespondwiththebestcurrentsolution.When appliedontherealmanufacturingfloor,theproposedalgorithmcanleadtoasignificantreductionincycletimes.
[3] H. Niedermayer, and O. Rose, “A simulation based analysis of the cycle time of cluster tools in semiconductor manufacturing”,Inproceedingsof15th Europeansimulationsymposium,Delft,Netherlands,2003.
[6] M.Dummler,“UsingsimulationandGeneticalgorithmtoimproveclustertoolperformance”,Inproceedingsofthe1999 wintersimulationConference,875 879
[1] Goldberg,andEncore,IntroductiontoGeneticAlgorithm,1998and1997 [2] http://lancet.mit.edu/ga
International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056 Volume: 08 Issue: 12 | Dec 2021 www.irjet.net p ISSN: 2395 0072 © 2021, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page1372
[7] S.Oechsner,andO.Rose,“Afilteredbeamsearchapproachtoschedulingclustertoolsin semiconductormanufacturing”, inproceedingsofIERC’05,Atlanta,USA,May14 182005.
REFERENCES
[5] M.Dummler,“Modelingandoptimizationofclustertoolsinsemiconductormanufacturing”,Ph.Dthesis,Departmentof computerscience,Universityof Wurzburg,Germany,2004.
[4] H. Niedermayer, and O. Rose, “Approximation of cycle time of cluster tools in semiconductor manufacturing”, In proceedingsofannualIIEEngineeringResearchConference,Texas,2004.
[8] S.Oechsner,andO.Rose,“Schedulingclustertoolsusingfilteredbeamsearchandrecipecomparison”,Inproceedingsof 200wintersimulationconference,2005