Reliable production of steep edge interference filters

Transcription

1 Relable producton of steep edge nterference flters Marc Lappsches a, Peter Pfefer a, Uwe Schallenberg a, Henrk Ehlers b, Detlev Rstau b a mso jena Mkroschchtoptk GmbH, Carl Zess Promenade 10, Jena, Germany b Laser Zentrum Hannover, Hollerthallee 8, Hannover, Germany ABSTRACT Interference flters for spectroscopc purposes or sensor applcatons are characterzed by strctly specfed spectral blockng and transmttng regons wth ntermedate steep edges. These steep edges must be postoned wthn nanometer accuracy whle the coatng may consst of more than one hundred non-quarterwave layers. Though modern on asssted deposton processes n conjuncton wth quartz crystal control are well suted for the producton of complex flters, an optcal montorng devce seems to be necessary to fulfll the demandng spectral requrements. Broad band optcal montorng (BBM) drectly on the calotte has been employed to control the producton of ths type of band stop flters. For a large number of also dfferent types of these coatngs the BBM-technque demonstrated ts capablty to mprove the relablty and flexblty n ndustral producton. Wthn a stable well-characterzed deposton process error selfcompensaton effects allow for a fast realzaton of varous desgns wthn specfed tolerances. Nevertheless, optcal broad band montorng could not be appled to all types of these steep edge flters because error propagaton leads to unreachable solutons of the thckness tracng algorthm for specfc cases. The gven examples of complex steep flters and the correspondng post analyss of stored onlne spectra as well as the smulaton of the montorng process reveal the nfluence of the desgn tself to ths occurrence. A suggeston for an dentfcaton of crtcal thckness values wthn the layer sequence s dscussed and solutons to the problems are presented. Keywords: optcal nterference coatngs, broadband optcal montorng, band stop flters, precson optcs 1. INTRODUCTION The realzaton of complex oxde delectrc nterference flters n automated manufacturng and ndustral producton envronment usually requres nsenstve and robust desgns. The electron beam evaporaton process, n conjuncton wth plasma on assstance, offers the capablty to produce stable, dense and shft-free flms wth hghly reproducble refractve ndces and low absorpton. In many cases the thckness determnaton va quartz crystal montorng at the calotte centre poston yelds suffcent precson. A certan error budget has to be taken nto account whch consders several effects lke dfferng growth condtons or a varyng partcle dstrbuton due to a changng shape of the evaporaton plume. Flters for specal sensor applcatons or spectroscopc purposes,.e. laser fluorescence mcroscopy, are types of long wave, shortwave, band pass or multple band pass flters wth very strngent specfcatons. The requred spectral blockng regons and occasonally several hgh transmttng bands have to be separated by a transton zone whch defnes the wdth or steepness and the spectral poston of the edges. These zones may be smaller than 1% of the desgn wavelength, whle the optcal densty (OD = -log(t)) of the rejecton bands may exceed OD5 and more. In many cases, these specfcatons are no longer accessble by pure quartz crystal control n a relable way. Especally when the coatng s only one processng step wthn a cycle of already prepared components relablty s an mportant fact. To meet these demands t seems to be necessary to apply optcal montorng. Respondng on the changng spectral characterstc t allows for takng advantage of error self-compensaton effects. Thus ths method s essentally capable to keep a certan spectral characterstc fxed on the wavelength scale. Concernng the above stated flter types, many desgn solutons are accessble by thn flm systems consstng of quarter wave optcal thckness (QWOT) layers [1]. In ths case optcal sngle wavelength montorng would be preferable. But ths technque s affected by defcences n precson caused by nstable geometrcal factors, especally, when the montorng wtness glass s placed n the centre of the calotte. Addtonally most applcatons requre desgns wth a stop bandwdth, whch may not be realzed by a pure QWOT-system. These more sophstcated desgns would need an

2 elaboraton of a sutable montorng strategy frst. And ths means a loss n flexblty to get a fast and practcable soluton. One large advantage of the optcal broadband montor setup nstalled s ts measurement poston among the substrates of nterest. Effects of an nhomogeneous thckness dstrbuton along the radal poston of the calotte are drastcally reduced. The mplemented tracng algorthm for determnng the actual thckness has been demonstrated to be a sutable control mechansm for the realzaton of complex delectrc flters [2]. Dependng on the process stablty concernng the reproducblty of the materal dsperson data, many desgns can be accomplshed by just loadng the scrpt and press the start button. But certanly there are several condtons concevable for whch ths stuaton may not be gven. One extreme example may be represented by a flter, whch blocks the montorng wavelength range completely. Here surely precson has to be cut due to the lower dynamc range of the CCD-devce nstalled n the spectrometer system compared for nstance to a photomultpler detector. The present paper tres to fgure out whch crcumstances lead to a falure of ths type of montorng. Propagaton of thckness errors may lead to a msnterpretaton of the measured transmttance data causng severe devatons n thckness determnaton [3]. An understandng of ths behavor can be utlzed to nstall a smple test tool for checkng desgn sutablty. The presented coatng examples show a desgn soluton frst whch produces a non-quarter wave band pass flter n the straghtforward manner mentoned above. Ths may be a good canddate to be the postve example. On the other hand a broad band-blockng flter seems to be sutable to be the negatve example at the frst glance. Calculaton of the ntegrated gradual change n transmttance gves evdence of a manageable controllablty by broadband-montorng. The analyss of such a plot versus growng thckness gves a hnt of the necessty to change the layer sequence of the desgn. 2.1 Expermental 2. METHODOLOGY Frst the expermental set-up wll be brefly explaned. The prncple algorthm that nterprets the transmttance data s based on the well-known matrx formalsm. The embedded dscrepancy functon can by vsualzed, whch may gve nsght nto the methods ablty to determne correct thckness values. The basc prncple of the optcal broadband-montorng method n thn flm technology has been already publshed n the 1980 th for example n [4]. Further development towards the employed commercally avalable product consdered n the present study has been done by the LZH 1. The optcal broadband-montorng devce nstalled conssts of a fbercoupled CCD-array spectrometer that s lnked to the plant by a fber, whch collects lght from a halogen lamp nstalled nsde the chamber, and by a trgger-unt responsble for synchronzaton of the measurement cycles wth the calotte rotaton. The adaptaton to the Leybold Syrus Pro plant has been realzed wthn an mplementaton nto the general steerng software. The APS technology s used for on assstance of the evaporaton process. It s mportant to menton that stable process parameters are one requrement for a proper functon of ths optcal control method. Partcularly, the dsperson and absorpton behavor of all materals nvolved have to be characterzed very carefully. The montorng system comples calbrated transmttance data for every revoluton of the calotte. Each measurement wll be utlzed for calculatng the thckness of the growng layer. The already deposted layer stack underneath wll be assumed as correctly realzed accordng to the desgn. Hereby an teratve algorthm fnds the local mnmum regardng the sum of the devatons modulus from the theoretcal (T THEO. ) and measured (T MEAS. ) transmttance at every n avalable wavelength poston. Thus the basc equaton to be mnmzed s: MSE n T n MEAS 2.( ) TTHEO.(, d). (1) Here MSE stands for mean square error, and the varable parameter s the thckness d of the actual growng layer. An extrapolaton towards the layer termnaton pont s possble by takng several calculated thckness values nto account. From a practcable pont of vew the advantage of ths method s ts ndependence from any further nformaton. But at a certan level of coatng complexty addtonal nformaton for reasons of stablzaton may be helpful [5]. 1 Laser Zentrum Hannover e.v., Department of Laser Components, Hollerthallee 8, Hannover, Germany

3 2.2 Analyss and Interpretaton The determnaton of a mnmum poston by an teratve algorthm s easer when the mnmzaton functon shows a large gradent. Thus t can be recognzed that the devaton of T wth respect to d can by dentfed as a measure of relablty of the actual thckness determnaton. It follows from the equaton (1): n1 0 1 T n 2 n1 THEO 1. THEO. THEO. 0 n ) (, d) d 2 T (, d) T (, d d d T T The quantty T depcts the amount of change n transmttance from one traced thckness step to the next one wth dfference d. Thus a larger value of ths sum means that the algorthm may easer dstngush subsequent thckness ncrements whle comparng the onlne-measurements wth theoretcal values. Fgure 1 llustrates such a change n transmttance schematcally at two exemplary wavelengths showng that the dfferental quotent s the decsve quantty. (2) T T 2 T 1 ΔT ΔT j d Δd λ Δd Fg. 1. T s summed up by the nfntesmal changes n transmttance (from curve T 1 towards curve T 2 ) at two selected wavelength postons and j due to the change n thckness d. The nterpretaton of the resultng values of T may be explaned best wthn the framework of examples. Fgure 2 demonstrates the T -plot of a growng sngle hgh ndex layer of tantala on a B270 substrate. The step wdth for the thckness d s chosen to be a value of 0.6nm accordng to a typcal deposton rate and tme step between measurements. The montorng range was 400nm to 1000nm. As expected, the begnnng of the growng layer shows nearly no change n transmttance resultng n T -values near zero. Ths behavor s consstent wth the experence, when the coatng process just started and the measured transmttance of the bare substrate does not seem to change at frst. λ j λ

4 ΔT Increasng thckness leads to a sgnfcant rse caused by a growng number of detaled spectral features of the sngle layer. The frst mnmum corresponds to the appearance of the frst spectral turnng pont wthn the chosen wavelength range. In ths case the mnmum corresponds to 1QWOT@525nm. At ths thckness value the spectral change remans nearly statc, and after ths pont almost the complete spectrum seems to slowly rse agan. It s magnable that at ths poston the determnaton of the thckness s more error senstve than at other thckness values. Ths fact clearly marks a crucal dsadvantage compared to the sngle wavelength montorng concept. The followng maxmum arses when the growng thckness passes 1QWOT@750nm where the transmttance reaches ts second extreme poston. It has also to be consdered that the absolute value of T depends on the magntude of the materals refractve ndex. No sgnfcant dfferences towards the substrate refractve ndex wll cause only small changes n transmttance tantala sngle layer QWOT@750nm 1QWOT@525nm thckness [nm] Fg. 2. T (d)-plot of a growng hgh ndex sngle layer on a B270 substrate wth step wdth 0.6 nm. The magntude of the quantty depends on the materals refractve ndces and ts course s also effected by the chosen wavelength regon (here: nm). The marked extreme postons correspond to 1QWOT at 525nm and at 750nm respectvely. Further examples wll be gven n the next secton n conjuncton wth results of more complex coatngs. 3. RESULTS As frst example a band pass flter at 656nm s chosen. It s realzed by a near QWOT-system wth detuned cavtes. The adaptaton to the spectral requrements, especally the wdth and the smoothng of the pass regon, led to a complete nonquarter wave desgn. T rses up agan n the The llustraton of the spectral changes of ths 67 layer Ta 2 O 5 /SO 2 -system s gven n Fgure 3. frst layer, and after the second one t does not fall short off a certan lmt untl the end of the complete coatng. Ths level may be compared wth values taken durng the frst layer. The spectral changes are at least as useful for a relable thckness determnaton as they are after a growth of 6nm. And ths thckness of a hgh ndex materal starts to be well suted for dentfcaton by the broad band montorng system.

5 transmttance [%] Fg. 3. T (d)-plot of the growng band pass flter. Strong dfferences between the materals are observable caused by the dfferng refractve ndces. Stll the graph does not fall short of a certan level. Thus the coatng should be controllable durng the complete run and layer swtchng should be possble wthn suffcent precson. Ths statement may hold although nether measurement errors nor thckness errors due to statstcal or cumulatve effects are ncluded n the representaton. As suggested the flter could be manufactured successfully wthout any problems as demonstrated n Fgure specfcaton theory sample#1 sample# wavelength [nm] Fg. 4. Transmttance of two samples of the deposted band pass flter compared to the theoretcal desgn target spectrum. The forecast of good reproducblty s fulflled.

6 The second example does also deal wth a band pass flter. But ths one s specfed at the wavelength 905nm whch s near the edge of accessble wavelengths of the montorng system. Addtonally the complete vsble and near ultra-volet spectral regon has to be blocked, leadng agan to a complete non-quarterwave desgn. Fg. 5. Comparson of two T (d)-plots for two dfferent TO 2 /SO 2 -desgns of a 900nm band pass flter wth a wde blockng regon. At certan ndvdual overall thckness values, very small T values ndcate that a thckness determnaton s not possble n a relable way anymore. In Fgure 5 two dfferent desgns are analyzed, whch produce quas the same spectral band pass characterstcs wth also nearly dentcal overall thckness. The frst desgn shows very small T values already before reachng the half of the complete flter and fall short of a reasonable lmt already n an early stadum of the developng flter. For example the level of s reached at 3500nm of 8200nm total thckness, whch corresponds to the 32 th of 89 layers, respectvely. In fact, t has been almost mpossble to get a satsfactory result utlzng broad band montorng n practce. T that s stll After restructurng the desgn, at least the last two mcrons of the coatng show a low level of manageable. Essentally the blockng from the green to the red spectral regon has been shfted towards the end of the layer sequence. Ths acton dd result n a delayed suppresson of sgnfcant spectral features. The new desgn fulflled the specfed requrements wth 81 layers. The correspondng transmttance dagram of the fnally realzed desgn no.2 s depcted n Fgure 6.

7 transmttance (%) specfed average theoretcal desgn measurement wavelength (nm) Fg. 6. Transmttance of the adapted 900nm band pass flter desgn compared to the theoretcal desgn target spectrum. After analyzng a frst desgn approach a necessary restructurng could be dentfed. The agreement of the measurement wth the target theory curve demonstrates that thckness errors are almost neglgble. As a consequence of suffcent changes of the transmttance durng the coatng and especally at the end of each layer the algorthm s capable to determne the layer termnaton ponts wth hgh precson. 4. CONCLUSIONS The advantages of optcal broadband montorng drectly on the rotatng calotte are ts measurement poston tself, the ablty to control the optcal thckness, and the flexblty to realze a knd of rapd prototypng. By analyzng the mode of operaton of the algorthm, whch determnes the layer thckness of a growng nterference flter, explanatons for a dfferng desgn dependent precson n thckness control could by llustrated. The preparaton of the so called T (d)- plot allows for an estmaton of the desgn sutablty for a control by broadband montorng. Senstve layers, whch may be dffcult to control, or requred desgn recalculatons can potentally be dentfed. The analyss of a layer sequence by ths tool makes a successful coatng more predctable. The authors thank the German Federal Mnstry of Economcs and Labour (BMWA) for the fnancal support wthn the framework of the cooperatve research project Optsche Hochlestungs-Flter für de Fluoreszenzmkroskope wthn the program Pro Inno II. REFERENCES [1] Thelen, A., Desgn of Optcal Interference Coatngs, McGraw-Hll, New York, (1988)

8 [2] [3] [4] [5] Rstau, D., Ehlers, H., Groß, T., Lappsches, M., Optcal broadband montorng of conventonal and on processes, Appled Optcs, Vol. 45, pp (2006) Lappsches, M., Groß, T., Ehlers, H., Rstau, D., Broadband optcal montorng for the deposton of complex coatngs, Proc. SPIE, Vol. 5250, pp (2003) Tlsch, M., Scheuer, V., Staub, J., Tschud, T.T., Drect optcal montorng nstrument wth a double detecton system for the control of multlayer systems from the vsble to the near nfrared, Proc. SPIE, Vol. 2253, pp (1994) Tkhonravov, A. V., Trubetskov, M. K., Stablzaton of computatonal algorthms for the characterzaton of thn flm coatngs, Numercal Methods and Programmng, Vol. 6, pp (2005)