Correctng Image Placement Errors Usng Regstraton Control (RegC ) Technology In The Photomask Perphery Av Cohen 1, Falk Lange 2 Guy Ben-Zv 1, Erez Gratzer 1, Dmtrev Vladmr 1 1. Carl Zess SMS Ltd., Karmel Israel 2. Advanced Mask Technology Center, Dresden, Germany Correspondng author: Av Cohen 1 (a.cohen@smt.zess.com) 1. ABSTRACT The ITRS roadmap specfes wafer overlay control as one of the major tasks for the sub 40 nm nodes n addton to CD control and defect control. Wafer overlay s strongly dependent on mask mage placement error (regstraton errors or Reg errors) 1. The specfcatons for regstraton or mask placement accuracy are sgnfcantly tghter n some of the double patternng technques (DPT). Ths puts a heavy challenge on mask manufacturers (mask shops) to comply wth advanced node regstraton specfcatons. The conventonal methods of feedng back the systematc regstraton error to the E-beam wrter and re-wrtng the mask are becomng dffcult, expensve and not suffcent for the advanced nodes especally for double patterng technologes. Sx producton masks were measured on a standard regstraton metrology tool and the regstraton errors were calculated and plotted. Specally developed algorthm along wth the RegC Wzard (dedcated software) was used to compute a correcton lateral stran feld that would mnmze the regstraton errors. Ths stran feld was then mplemented n the photomask bulk materal usng an ultra short pulse laser based system. Fnally the post process regstraton error maps were measured and the resultng resdual regstraton error feld wth and wthout scale and orthogonal errors removal was calculated. In ths paper we present a robust process flow n the mask shop whch leads up to 32% regstraton 3sgma mprovement, brngng some out-of-spec masks nto spec, utlzng the RegC process n the photomask perphery whle leavng the exposure feld optcally unaffected. Key words: Image Placement, Regstraton, Wafer Overlay, Photomask, Laser, RegC, RegC Wzard. Photomask Technology 2012, edted by Frank E. Abboud, Thomas B. Faure, Proc. of SPIE Vol. 8522, 85220D 2012 SPIE CCC code: 0277-786/12/$18 do: 10.1117/12.980258 Proc. of SPIE Vol. 8522 85220D-1
2. INTRODUCTION For the advanced nodes and double patternng photomasks manufacturng s becomng more and more demandng. The E-beam wrters are pushed to ther lmts regardng mage placement performance. Currently, f a mask s rejected because of mage placement s out of specfcaton, there s no way to correct for t. The mask has to be scrapped and must be rewrtten agan Carl Zess SMS has developed a new technology named RegC whch enables the user (mask shop) correctng the global regstraton errors and mprovng the mage placement of a manufactured mask. The process s based on a fs laser technology smlar to the technology used n the CDC tools of Carl Zess for CD Unformty correcton 2,3. The RegC (Regstraton Control) process enables the mask maker to mprove the regstraton performance of a mask or to brng a mask whch s out of specfcaton nto specfcaton. As a result the mask manufacturng yeld s ncreased. Fgure 1 shows the basc Regstraton Control process flow n the mask shop Fgure 2 shows the RegC block dagram. The system has two man optcal sub systems. The frst sub system s used to generate the deformaton elements (pxels) utlzng an optcal setup that ncludes the followng man components: Pulse laser, beam delvery path, beam steerng devce and a focusng optcs. The second sub system s a metrology system that s used to measure and characterze the propertes of the generated deformaton element, so-called Mode Sgnture (MS). The Mode sgnature wll be then used as one of the nputs for the RegC job computaton by the dedcated software named RegC Wzard. Fgure 3 shows the basc nputs and outputs utlzng ths supportng software. In ths paper "fused slca", "quartz (Qz)" and mask "blank substrate" are used nterchangeably. / Mask E beam, wrter -> / Regstraton Metrology by PROVE J Regstraton Control In The mask Shop Process Flow \ Yes., Regstraton Metrology Pellcle through pellcle Mount PROVE 1 1 New Tool: Correct regstraton No new mask wrtng Benefts: Mask Shop No mask rewrte 4 manufacturng cost savng Correct for pellcle mpact 4 no mask rejects by wafer fab Wafer Fab Improved overlay 4 Better wafer yeld No f- New Tool: Correct regstraton Through Pellcle Fgure 1.RegC process flow Proc. of SPIE Vol. 8522 85220D-2
Metrology sub system Deformaton elements propertes determnaton (Rego process calbraton) Beam Scanner Deformaton elements generaton sub system Beam Delvery J Pulse Laser Deformaton elements top vew Beam Shaper Focusng Optcs Back surface Fgure 2. RegC block dagram of the tool man components The RegC wzard nputs & outputs Tool property Mode Sgnatures (MS) Retcle Property Intal Regstraton Problem RegC Wzard RegC tool job fles User Input Doman & Attenuaton lmts Statstcs and mages Fgure 3.The basc nputs/outputs of the supportng software - RegC wzard 3. THE REGISTRATION CONTROL (REGC) PRINCIPLES Intra volume laser wrtng at certan condtons creates a predctable deformaton element n the quartz (Qz) materal. Ths deformaton can be descrbed by a physcal-mathematcal model that well represents the deformaton caused by RegC element. The deformed zone nsde the Qz bulk s a 3 dmensonal volume of fused slca whch has a slghtly dfferent morphologcal organzaton of the atoms wth a slghtly less dense packng, or lower densty. The zone wth lower densty expands and pushes away the adjacent atoms and thus deforms the whole bulk of the Qz pece. Due to the elastc amorphous property of fused slca ths deformaton behaves almost truly elastcally wthout crtcal breakage (cracks). Proc. of SPIE Vol. 8522 85220D-3
In other words, when consderng very small deformatons n the order of ppb and even ppm, fused slca behaves practcally lke rubber, elastcally. The specal model that was developed to descrbe the accumulatve effect of multtude pxels generated nsde the Qz substrate takes nto account the physcal propertes of fused slca such as ts Young Modulus, ts Posson rato etc. The model has been verfed expermentally and provdes a laser-materal assocated parameter, called the Mode Sgnature (MS). The MS defnes the magntude and angle/drecton of the deformaton nduced by wrtng a laser pxel at gven condtons. The Mode Sgnature can be used frst to calculate and predct the deformaton and hence the affect on regstraton by wrtng a gven array of pxels. Second and relevant for the RegC process the MS can be used to calculate a set of pxels needed to compensate for a gven regstraton error map. The current RegC process can only nduce expanson pxels. Ths means that the average mask dmenson after the RegC process wll always be larger than before the process. Ths also means that the absolute value of regstraton after RegC wll typcally be hgher than the absolute regstraton error before the process, except for rare cases where the whole mask error was contracted relatve to the target. However ths s not a lmtaton snce the target of the RegC process s not to compensate for the absolute regstraton errors, but rather to remove only the non compensable errors as t s well known to the mask and ltho ndustry that the scanners have the ablty to compensate for all systematc lnear errors whch have rotatonal, orthogonal and scale components (n short "Scale and Ortho"). The man ssue wth regstraton errors of masks s the non compensable resduals, the regstraton errors whch are left over after the scanner has done ts job. These resduals are typcally 6-8 nm 3S n advanced 40nm nodes and below. However the specs at these nodes are 4-8 nm and n sub 20 nm nodes can go down to < 4nm, especally n double patternng technologes. Therefore the task of the RegC process s to decrease these non compensable resduals from ~8 nm to ~4 nm, or about 50% mprovement n the 2X and 1X nodes. Because of the mportance of the scanner ablty to compensate for scale and ortho, all regstraton metrology tools report n addton to raw regstraton errors also the scale and ortho (S/O) removed resdual errors. These are the values whch typcally nterest mask makers and ther fab customers who are nterested eventually at mask to mask overlay n the scanner. The capablty of the scanner to remove specfc regstraton errors s shown n Fgure 4 as an example. (4a) Regstraton error - Raw Example of regstraton compensaton capabltes by scanner (4b) Regstraton error - S1O XM;,,-11.96, XMax 2.00, X3s 7.25 YM;,, -0.99, YMax 1.48, Y3s 1.51 XM;,, -9.91, XMax 3.04, X3s 6.33 YM;n, -0.55, YMax 0.90, Y3s 1.01 Fgure 4a shows the raw regstraton error whle Fgure 4b shows the regstraton error after S\O removal. Note that the large vectors on the top left have been reduced n magntude but not sgnfcantly Proc. of SPIE Vol. 8522 85220D-4
The basc prncple of RegC s to take the regstraton error ("the problem ) shown n Fgure (5a) and apply the requred regstraton change shown n Fgure (5b).The vector summaton of those two wll result n a new state shown n Fgure (5c) that wll enable hgher capabltes of the scanner to remove resdual errors by applyng S\O as shown n Fgure (5d). The basc RegC'F prncple (5a) The ntal regstraton error (5b) The requred regstraton change m NNE NM. am NẈ. NNW MO MN ME, 1 + + -. r. + (5c) The ntal problem +the nduced change (5d) After S1O removal by scanner ; 3S X& < 0.3 nm Fgure 5a shows the ntal regstraton error (raw) whle Fgure 5b shows the requred regstraton change n order to brng all the errors to a correctable feld. The post process regstraton error shown n Fgure 5c s the vector summaton of the ntal error and the nduced change. Fgure 5d shows very low resdual error after S\O removal by scanner 4. EXPERIMENTAL RESULTS Sx OMOG (Opaque MoS On Glass, bnary photomask materal developed by Shn-Etsu, IBM and Toppan) producton plates (28nm) were measured by a regstraton tool as an nput for the RegC job computaton. The regstraton measurement error was estmated as 0.8 nm (long term and short term error components 4 ).Then a pre-calculated RegC process was appled and the plates were measured agan for "Post" process regstraton errors. In order to maxmze the process capabltes, the actual processng was dvded nto two steps; each step had ts own mode sgnature (deformaton propertes). Fgure 6 shows the generated RegC jobs for mask number 1 along wth schematc drawng of the assocated deformaton drecton due to the gven mode utlzaton. Referrng to the mask's Z drecton, all the sx masks were processed (creatng pxels) at the quartz plate center whle the spoken deformaton or regstraton change was pre-calculated and targeted to the mask absorber level. Proc. of SPIE Vol. 8522 85220D-5
(6a) RegC process area (6b) Appled pxels densty Mode # 1 (6c) Appled pxels densty Mode # 2 No process over the exposure feld Full mask area (152mmX152mm) RegC process area Actve area - not processed (128mmX105mm) s - Pxels In arbtrary unts ; Red - hgh densty Blue - low densty torr Pxels Fgure 6a shows the selected area of processng over the mask, Fgure 6b shows the frst mode lateral pxels densty dstrbuton along wth the assocated deformaton drecton, Fgure 6c shows the lateral pxels densty dstrbuton n the case of the second complementary mode utlzaton Ths experment examned two man aspects related to the RegC technology; the frst aspect s how accurate the physcal- mathematcal model s and can t predct the regstraton change pror to the mask processng? The second s how effcent wll the process be consderng the constrant of processng the mask utlzng less then 42% out of the quartz area, keepng the exposure feld optcally unaffected. It s mportant to menton that n ths test we were utlzng a system whch s not a dedcated RegC system for the actual processng step. As for the accuracy of the model, Fgure 7 shows on one plot the actual measured and the predcted change n regstraton due to the RegC process for mask number 1. Hgh agreement can be vsually seen and t's been quantfed by coeffcent of determnaton R^2 = 0.94. Moreover, Table 1 summarzes the 3 Sgma dfferences between the actual measured regstraton errors post process and the predcted ones by the RegC wzard where less then 0.75 nm devaton can be seen. Proc. of SPIE Vol. 8522 85220D-6
Measured and predcted regstraton change /l \ 4.4\ I -... -... t, r /, r t 1! f 4 Blue - measured Brown - predcted Very good agreement ; RA2 = 0.94 Fgure 7.The actual versus predcted regstraton change due to the RegC process 5nm I 3 Sgma values n nm after ScalelOrtho removal Actual Measured Post Predcted Post Dfference Mask Number X Y X Y X Y 1 7.20 10.80 7.45 10.64-0.25 0.16 2 5.90 11.30 5.76 10.56 0.14 0.74 3 6.40 11.60 6.64 10.88-0.24 0.72 4 6.60 8.10 6.49 7.83 0.11 0.27 5 5.30 5.50 5.04 4.76 0.26 0.74 Table 1: Dfferences between the measured and predcted regstraton error 3 Sgma's As for the process effcency, Fgure 8 shows mask number 1 regstraton errors before and after the RegC process. 32% mprovement n X axs and 14% mprovement n Y axs 3 sgma were observed, brngng an out of spec mask nto spec. Table 2 summarzes the mprovements acheved after the 5 masks processng. An 8% -32% mprovement was seen. Proc. of SPIE Vol. 8522 85220D-7
1 (8a) Regstraton error pre process (8b) Regstraton error post process Scale /Ortho removed Scale /Ortho removed 3 Sgma X : 10.55 nm 3 Sgma Y : 12.57 nm 4nm > 32% Imp> > 14% Imp> 4nm 3 Sgma X : 7.20 nm 3 Sgma Y : 10.80 nm Mask out of spec Mask n spec Fgure 8. Pre (8a) and post (8b) RegC process regstraton errors of mask number 1 3 Sgma values n nm after ScalelOrtho removal Pre RegC Actual Post RegC % Improvement Mask Number X Y X Y X Y 1 10.55 12.57 7.20 10.80 32 14 2 6.80 12.60 5.90 11.30 13 10 3 7.30 13.20 6.40 11.60 12 12 4 7.80 8.80 6.60 8.10 15 8 5 7.00 6.50 5.30 5.50 24 15 Table 2: Pre and post RegC regstraton error 3 sgma and the assocated percentage of mprovement 5. CONCLUSIONS It was proven that a regstraton correcton stran feld can be computed usng a specal algorthm and that a laser based correcton method can be used to effectvely reduce the regstraton error n the mask wthout affectng any other mask propertes. The above reported experments have shown that a mask whch was rejected based on ts regstraton problem can be saved and brought nto spec by treatng the non actve area. It s recognzed that a better mprovement n the order of 50% could be acheved by applyng the RegC process n the whole mask area. For ths purpose Carl Zess has developed a new process where the whole mask area s treated. In addton, more and more chp manufacturers are now specfyng not only the mask regstraton error but also mask to mask overlay error, whch adds even more challenge to the mask maker. Proc. of SPIE Vol. 8522 85220D-8
6. AKNOWLEDGEMENTS AMTC s a jont venture of GLOBALFOUNDRIES and TOPPAN Photomasks and gratefully acknowledges the fnancal support by the German Federal Mnstry of Educaton and Research (BMBF). 7. REFERENCES 1. Schultz, B, et al,"meetng overlay requrements for future technology nodes wth n-de overlay metrology", SPIE Mcrolthography, 6581-13 (2007). 2. Pforr, R,et al,"performance comparson of technques for ntra-feld CD control mprovement", BACUS, 6730-107 (2007). 3. Buttgeret, U, et al," Process Wndow mprovement on 45 nm technology Non Volatle Memory by CD unformty mprovement", BACUS, 7823-11 (2010). 4. Enkrch, C, et al., "Regstraton measurement capablty of VISTEC LMS IPRO4 wth focus on small features," Proc. of SPIE Vol. 7028, 70282Y (2008). Proc. of SPIE Vol. 8522 85220D-9