Aurora Solar Technologies - Gordon Deans

PERC production yield improvement using I-V data and a Los Component approach 13 March 2019 Gordon Deans, CEO Aurora Solar Technolgies 2019 Aurora Solar Technolgies Aurora Solar Technologies 2019 Aurora Solar Technolgies 2 Mision –MEASURE, VISUALIZE, CONTROL Deliver superior results to the PV industry through measurement, understanding and control of critical cel manufacturing proceses. Solutions for Real-time measurement of critical-to-quality PV cel properties Analysis of the causal conections betwen proceses and I-V results to maximize yield and throughput Visualization of proces equipment behaviour for optimal control and management Proven with SMSL and high-efficiency solar cel manufacturers in China, Taiwan, Korea, and S.E. Asia Cel fabrication proces analysis and visualization for instant verification, diagnosis and control Real-time emitter, BSF, TCO measurement in c-Si cel production Previously at PV CelTech “Inline mid-stream wafer measurements to reveal process variation afecting PV cell performance and reliability” Introduction of a new method for using material and cell fabrication proces properties in proces analysi and control Benefits from using the method “Enhancing quality control in PV cel production by implementing Industry 4.0 design features” Further theory and applications of the above method to tie end- of-line results to cell fabrication process properties 2017 2018 3 2019 Aurora Solar Technolgies Today “Inline mid-stream wafer measurements to reveal process variation afecting PV cell performance and reliability” Introduction of a new method for using material and cell fabrication proces properties in proces analysi and control Benefits from using the method “Enhancing quality control in PV cel production by implementing Industry 4.0 design features” Further theory and applications of the above method to tie end- of-line results to cell fabrication process properties 2017 2018 Realization and production line examples 2019 Aurora Solar Technolgies 4 Basis –loss analysis approach IEEEJOURNALOFPHOTOVOLTAICS,VOL.5,NO.2,MARCH2015 619 ASystematicLossAnalysisMethodfor Rear-PassivatedSiliconSolarCells JohnsonWong,ShubhamDuttagupta,RolfStangl,BramHoex,andArminG.Aberle AbstractBycombiningcommonlyavailablesolarcellcharac- terizationmethodswitheasy-to-prepare teststructuresandpar- tiallyprocessedrear-passivatedsolarcellsfromtheproduction line,weshowthatvariouscelllossmechanismscanbe quanti- fiedinexquisitedetailtogenerateprocess-relateddiagnostics.An examplemonocrystallinesiliconlocalizedbacksurfacefieldso- larcelltypeis examinedusinga systematicroutinethatbreaks downthefactorslimitingopen-circuitvoltage,short-circuitcur- rent,andfillfactorFFtoidentifythecellstructure’sheadroomfor improvement. IndexTermsCharacterization,metrology,SiPVmodeling. I. INTRODUCTION L OCALIZEDbacksurfacefieldLBSFsolarcells,bifacial cells,andpassivatedemitterreartotallydiffusedsilicon wafersolarcelltypesrelyonpassivationlayersonboththefront andrearsurfacesofthedevicetomoreeffectivelyreduceminor- itycarrierrecombinationattherearsurface,comparedwiththe omnipresentaluminumbacksurfacefieldsiliconwafersolarcell [1],[2].Inthesecelltypes,thepassivationstructureiscompleted beforethemetallizationstep,makingitsensibletotrackthecell passivationqualityatthevariousstagesafterapplicationofthe passivationcoatings.Followingalineofpredecessorsolarcell lossanalysismethods[3],[4],weoutlineanopen-circuitvoltage V oc lossanalysisroutinethatenablestheextractionofthevar- iouscomponentsofcarrierrecombinationbythemeasurements ofexcesscarrierdensityineasy-to-prepareteststructuresand partiallyprocessedrear-passivatedsolarcellsfromtheproduc- tionline,usingcommonlyavailablecharacterizationmethods suchasradio-frequencyRFphotoconductance[5],photolu- minescencePLimaging[6],andshort-circuitcurrentdensity versusopen-circuitvoltageJ sc −V oc measurements[7].Com- binedwiththemoreestablishedmethodsofshort-circuitcurrent I sc andfillfactorFFlossanalysisonthefinishedcel,a comprehensivepictureofthefactorslimitingthecellefficiency canbeobtained.Thesetofmethodsdescribedinthispaperis relatedtoapreviouslypublishedsystemthatquantifiedseven powerlossmechanismsinasiliconwafersolarcellfrontmetal Manuscriptreceived September15,2014;revisedDecember2, 2014;ac- ceptedDecember8,2014.DateofpublicationJanuary16,2015;dateofcurrent versionFebruary18,2015.TheSolarEnergyResearchInstituteofSingapore SERISisaresearchinstituteattheNationalUniversityofSingaporeNUS. SERISissponsoredbyNUSandSingapore’s NationalResearchFoundation throughtheSingaporeEconomicDevelopmentBoard. TheauthorsarewiththeSolarEnergyResearchInstituteofSingapore,Sin- gapore117574e-mailJohnson.wongnus.edu.sg;shubham.duttaguptanus. edu.sg;rolf.stanglnus.edu.sg;Bram.Hoexnus.edu.sg;armin.aberlenus. edu.sg. Colorversionsofoneormoreofthefiguresinthispaperareavailableonline at http//ieeexplore.ieee.org. DigitalObjectIdentifier10.1109/JPHOTOV.2014.2388071 shading,frontsurfacereflectanceintheactivecellarea,front surfaceescapeoflight,seriesresistance,shuntconductance, nonperfectactive-areainternalquantumefficiencyIQE,and theforward-biascurrentatthe1-sunmaximumpowerpoint[3]. Themaindifferencesinthisstudyarethat1powerlossisre- placedbythequantificationoflossfactorsinthethree1-sun current–voltageparametersV oc , I sc ,andFF;2theV oc loss mechanismsarequantifiedintermsofthesaturationcurrent densitiesoriginatingfromthedifferentcomponentsofthesolar cell;and3thenonperfectactive-areaIQEisfurtherbroken downintoparasiticabsorptionandcollectionlosses,suchthat thecurrentlossesofopticaloriginparasiticabsorptioninthe rearreflectorcanbeclearlydiscernedfromthosearisingfrom recombinatione.g.,highrecombinationattherearsurface. Themoredetailedapproachinthisstudybetterenablesthe assessmentoftheimpactofdifferentprocessingstepsandlay- ers/structuresinthecellonitsperformanceandprovidesthe necessaryinputsforfurtheranalysisandpredictionsbysimula- tionmodels. II. EXPERIMENTALDETAILS Weillustratethelossanalysismethodsforascreen-printedAl localizedbacksurfacefieldAl-LBSFmonocrystallinesilicon wafersolarcell,buttheapproachisalsoapplicabletoothersolar celltypeswithrearpassivation[2],[8].Fig.1showsthesample andmeasurementplan,illustratingthestructureofthesamples indetail.Thefinishedcellswerefabricatedusing156-mmpseu- dosquarep-typemonocrystallineCzsiliconwafers.TableIlists therelevantmaterial,structural,andmeasured1-suncurrent– voltageI–Vparameters.TheV oc lossanalysisroutinerequires onesymmetricalpassivatedemitterstructureAandfourpar- tiallyprocessedsolarcellsB–E.Allfivesampleshavegone throughthefinalmetallizationcontactfiringstepintheproduc- tionlineinamultizoneconvectionfurnacetoensurethatthey experience similar thermal history as a completely processed cellandarethusrepresentativeofthedifferentstructuresinthe finaldevice.Thedegreeofequivalenceinthermalhistoryde- pendsontherelativeamountsofconvectiveandradiativeheat transferinthefiringfurnace.SamplesAandB,withoutfull- arearearmetallayers,wouldabsorbsignificantlymoreinfrared radiationincidentontherearsidesthanwouldsamplesCand E.Therefore,thesiliconnitrideSiN x layersinsampleAis probablyfiredatahighertemperaturethantheoneinthefully metallizedcell,leadingtoanuncertainlyintheemittersatu- rationcurrentdensityJ oe inthenextsection.Althoughthe dielectriclayersinsampleBshouldhave alsoexperienceda differentfiringtemperaturethantheonesinsamplesC–E,this isacceptablebecauseoneneedsnotassumethatthedielectric 2156-33812015IEEE.Personaluseispermitted,butrepublication/redistributionrequiresIEEEpermission. Seehttp//www.ieee.org/publicationsstandards/publications/rights/index.htmlformoreinformation. 622 IEEEJOURNALOFPHOTOVOLTAICS,VOL.5,NO.2,MARCH2015 TABLE II SATURATION CURRENTDENSITIESOFVARIOUSRECOMBINATIONCURRENTSOURCESUNDERTHETWO-DIODEMODEL SampleCfirstdiode J01,C Suns–PL 210fA/cm 2 SampleCseconddiode J02,C Suns–PL 14nA/cm 2 SampleDfirstdiode J01,D Suns–PL 290fA/cm 2 SampleDseconddiode J02,D Suns–PL 17nA/cm 2 SampleEfirstdiode J01,E Jsc−Voc 440fA/cm 2 SampleEseconddiode J02,E Jsc−Voc 22nA/cm 2 Passivatedemitterfirstdiode J0e Kane–Swansonmethod 93fA/cm 2 Bulkandrearpassivationfirstdiode J01,base,pass J01,C −J0e 117fA/cm 2 Rearcontactsfirstdiode J01,rear,met J01,D −J01,C 80fA/cm 2 Rearcontactsseconddiode J02,rear,met J02,D −J02,C 3nA/cm 2 Cellbasefirstdiode J01,base,cell J01,base,pass J01,rear,met 197fA/cm 2 Junctionseconddiode J02,junction J02,C 14nA/cm 2 Firstdiodeduetofrontmetallization J01,front,met J01,E −J01,D 150fA/cm 2 Seconddiodeduetofrontmetallization J02,front,met J02,E −J02,D 5nA/cm 2 Puttingittogether, Fig.4comparestheSuns–PL[11]curves ofsamplesC–E,aswellastheactualprobedV oc at different lightintensitiesJ sc −V oc ofthefinishedcellsampleE. TheSuns–PLandJ sc −V oc curves canbe analyzedusing thetwo-diodemodelforsolarcells.Underthismodel,inopen circuit,wehave J L J 01 bracketleftbigg exp parenleftbigg qV kT parenrightbigg −1 bracketrightbigg J 02 bracketleftbigg exp parenleftbigg qV 2kT parenrightbigg −1 bracketrightbigg 3 whereJ L isthelight-inducedcurrentdensity, andJ 01 andJ 02 arethefirstandseconddiodesaturationrecombinationcurrent densities,respectively.Notethatwehaveomittedanyshuntcur- renttermin3,thusassumingthattheshuntresistanceislarge enoughtocausenegligibleeffectsinallsamples.Basedoncell measurements,J L J sc 38.43mA/cm 2 forsampleE. For samplesCandD,withoutthe3metalbusbarand4finger shading,J L 38.43/1−f metal 41.32mA/cm 2 .TableII liststheJ 01 andJ 02 valuesof samplesC–Eat 25 C,whichare denotedbysubscriptscorrespondingtothesample.Italsolists, ona basisthatis normalizedtothewaferarea,thepassivated emitterJ 0e asmeasuredfromsampleAandcalculatestheJ 01 originatingfromthebulkandrearpassivationJ 01,base,pass , localizedrearpointcontactsJ 01,rear,met ,frontmetallization J 01,front,met ,andJ 02 originatingfromthejunctionunderthe passivatedemitterJ 02,junction ,andJ 02 duetofrontmetalliza- tionJ 02,front,met .ThesumofJ 01,base,pass andJ 01,rear,met is denotedJ 01,base,cell ,whichisthebasesaturationcurrentdenity ofthefinishedcell. At1-sunillumination,theimpactoftheJ 01 andJ 02 sources onthecellV oc canbeassessedbytherecombinationcurrent densitiesJ R they generate.Namely, J R J 01 expqV oc /kT andJ R J 02 expqV oc /2kTforfirstandseconddiodesources, respectively. TheseJ R ’s can then be divided by expqV oc /kT toderiveequivalentJ 01 values under 1-sunopen-circuit condi- tions,whicharesummarizedasapiechartinFig.5.Thechart givesa breakdownofthedifferentrecombinationsourcesona commonbasis. ThepiechartinFig.5givesacomprehensiveoverviewofthe recombinationlossmechanismsandoffersa basisofcompari- sontostate-of-the-artLBSFsolarcells.Theplasma-enhanced chemicalvapor depositionPECVDsiliconnitrideSiN x Fig.5. EquivalentfirstdiodesaturationcurrentdensityJ 01 valuesinfA/cm 2 atV oc arisingfromdifferentpartsoftheLBSFsolarcell. passivated emitteris excellent,having a J 0e of less than 100fA/cm 2 [19],[20].On the otherhand,the combination ofbulkrecombinationandrearrecombinationatthepassivated rear, consistingof a PECVDaluminumoxideandsiliconni- trideAlO x /SiN x stack,contributesabout117fA/cm 2 ,whichis manytimeshigherthantheJ 01,base,pass reportedforLBSFc-Si cellswithhighV oc andrearpassivationsurfacerecombination velocity of less than 10 cm/s [21]–[23]. The front-grid-related recombinationisalsounusuallylarge,pullingthejunctionvolt- agefrom655mVinsampleDto642.5mVinsampleE.There arethreeeffectsintheintroductionofthefrontgridcontributing tothisvoltagedrop1metalrecombinationbymetalcontact- inganderosionoftheemitter;2thegridconnectingthevast areaofthecelltolocalizedhigh-recombinationregionssuchas thewaferedgesviahighlyconductive paths;and3reduction ofJ sc bythemetalshadingfractionf metal 7in thiscase. Ofthesefactors,shadingcanonlylowertheV oc by roughly kTln1−f metal ,i.e.,by∼2mV.Asforedgerecombination, duetothecellplane’sfiniteconductance,theimpactonthecell V oc wouldbemorepronouncedatlowlightintensitiescompared withat1sun.Thus,edgerecombinationandotherlocalizedcur- rentsinks,regardlessofwhethertheyaren1orn2diodes innature,willinevitablycausetheJ sc −V oc curve toexhibit ahighidealityfactoratlowlightintensity.Onecan,therefore, relyontheJ sc −V oc characteristicsinFig.4toderiveanupper 624 IEEEJOURNALOFPHOTOVOLTAICS,VOL.5,NO.2,MARCH2015 Fig.7. Short-circuitcurrentdensityJ sc lossmechanismsandtheirmagni- tudesinmA/cm 2 . uponmultiplicationwiththeelementarychargeq,thecontribu- tionsascurrentdensities,asdisplayedinFig.7 asa piechart. Thetotallosscurrentdensityof7.94mA/cm 2 ,whenaddedt the cellJ sc of38.43mA/cm 2 ,yields46.37mA/cm 2 ,whichisthe maximumgenerationcurrentdensityavailablefromthephoton fluxintheAM1.5Gspectrumbetween300and1200nm. Thequantitative treatmentof currentlossesenablesoneto comparethecellunderstudywiththestateoftheart.Thecur- rentrecordLBSFcell,forexample,hasacurrentdensityJ sc of 39.8mA/cm 2 [19].Oneobviousdifferenceis inthefrontgrid metallizationfraction,being4forthestateoftheartcellcom- paredwith7forthecellofthisstudy. Thisdifferencealone contributesto346.37mA/cm 2 1.39mA/cm 2 inlostcur- rentdensity. Notwithstandingthe metalshading,in the non- shadedareas,thecurrentdensityisreportedtobe41.3mA/cm 2 forthestate-of-the-artLBSFcell[19],whichispracticallyiden- ticaltothenonshadedareacurrentdensitycalculatedforthecell in thisstudy. In reality, it is difficultto quantifyveryfinedif- ferencesincellJ sc ,especiallyfortwocellsthatweremeasured underdifferentsolarsimulators.It is,therefore,worthwhileto delve a littledeeperintotheJ 01,base,cell andthequantumeffi- ciencycurvetoestimatetheimpactofdifferentrearpassivation qualityonthebasecollectioncurrentloss.AccordingtoBasore [27],J 01,base,cell andtheeffectivebasediffusionlengthL eff are relatedvia J 01,base,cell qn 2 i D N A L eff 4 where D∼29.2 cm 2 /s is the minority carrier diffusion coefficient at N A 8.810 15 cm −3 .UsingJ 01,base,cell 197fA/cm 2 resultsinL eff 2mmforthecellofthisstudy. This effective diffusionlengthvalue is consistentwith the measuredIQEEQE full /[1−f metal 1−R active ]93 at 1000nm,whencheckedagainstPC1D[30]simulationsus- ingtheappropriateinternalreflectanceparametersandeffective diffusionlength.Incomparison,thestate-of-the-artLBSFcell hasconsistentlyIQE∼96at1000nm[19],[21]–[23],which impliesroughly0.2mA/cm 2 higherJ sc fromreductioninbase collectionlossaccordingtoPC1D.Asalowerbound,usingbulk diffusionlengthand rearsurface velocitiesreportedby Gatz Fig.8. FillfactorFFlossmechanismsandtheirmagnitudesinpercent. et al.fora Ga-dopedlight-stabilizedLBSFcell[23],thesim- ulationshowsthatthereisatleasta differenceof0.1mA/cm 2 . Optically,itisevidentthatthelight-trappingabilityofthestate- of-the-artcellisnotmuchdifferentfromthecellofthisstudy, asbothhavesimilarIQEvaluesandreflectanceat1100nmand beyond.ThereisalsonotasignificantdifferenceintheIQE reflectancein the300–500-nmrange.Nevertheless,any sma