1. The document discusses using image morphology and machine learning to define dose and focus settings for NSO tools based on FE maps printed using different NSO tools, layouts, corners, and masks.
2. Testing showed higher accuracy (67.7%) when defining dose and focus for the same corner versus different tools (43.3% accuracy) or masks (43.3% accuracy).
3. The conclusion recommends collecting more data with smaller dose and focus steps to confirm results, and establishing a baseline by retraining the software after any NSO hardware or software changes.
Software and Systems Engineering Standards: Verification and Validation of Sy...
LVTS - Dose&Focus recognition by image part2
1. Is it possible to make the life of Litho Engineer easier or how to trace DOSE and FOCUS by image Part2
Vladislav Kaplan, September 2008
2. Test results
Chosen NSO tools –NSO105 and NSO106
▫FE maps printed by two different NSO tools
▫Class 6z12
Chosen Layout
▫Two similar patterns for C, D, E, single pattern for A, 3 pattern for B
▫Rec1 –12,23063 corner B
▫Rec2 -7721,14517 corner A
▫Rec3 -11,23063 corner B
▫Rec4 --23,23062 corner B
▫Rec5 -1404,24196 corner E
▫Rec6 -1415,24196 corner E
▫Rec7 -14916,1128 corner D
▫Rec8 -14927,1128 corner D
▫Rec9 –13525,24196 corner C
▫Rec10 –13536,24196 corner C
A
B
C
D
E
3. Test results continued
Chosen tests
▫Tool variation
▫Corner variation
▫Sample plan variation
▫Mask variation
Tool variation
▫Training results tests for same MP between NSO106/NSO106
▫Corner variation
▫Training results tests for same corner NSO106/NSO106 (different MP)
Mask variation
▫Training results tests for different corners per NSO106/NSO106
Sample set variation
▫Training results tests for different size sets per same MP.
CDSEM tool variation
▫Training results tests for same MP same CDSEM (2 different runs)
4. Vector Set
▫Hydraulic radius
▫Convex Hull perimeter
▫Waddle disk diameter
▫Max Ferret diameter
▫Elongation
▫Compactness
▫Heywood circularity
▫Type factor
▫Area
▫Orientation
Hidraulic44.24.44.64.855.2147101316192225283134Pic Value Run1Run2Run3Hidraulic44.24.44.64.855.2147101316192225283134Pic Value Run1Run2Run3Waddle disk diameter2627282930313233341357911131517192123252729313335Pic Value Run1Run2Run3Waddle disk diameter2627282930313233341357911131517192123252729313335Pic Value Run1Run2Run3Elongation00.10.20.30.40.51357911131517192123252729313335Pic Value Run1Run2Run3Type factor00.10.20.30.40.50.61357911131517192123252729313335Pic Value Run1Run2Run3Area6006507007508008509001357911131517192123252729313335Pic Value Run1Run2Run3Orientation0204060801001201401601801357911131517192123252729313335Pic Value Run1Run2Run3
5. ResultsTool variation:
▫Week correlation between NSJ105 and NSJ106 –training set is tool specific. Mask variation:
▫Average probability of 43.3% to define Dose and Focus correctly. Corner variation:
▫Average probability of 67.7.3% to define Dose and Focus correctly
Accuracy of recognition00.10.20.30.40.50.60.70.80.9Trial Accuracy Tool variationCorner variation NSJ105Corner variation NSJ106Mask Variation NSJ105Mask Variation NSJ106Same MP - different toolsMix Different corners per toolMix of same corners per tool2B2B2B2C2E2B2D2CBEBCBCBCBCCDCDBDBBCCDDBC
6. ResultsCDSEM variation (CDD107 vsitself , run 1 and 2): High correlation -0.87, but apparently charge impact significant -13%. Sample set variationExpected reduction of accuracy. Combination of increase sample set and mask/corner variation gives significant drop in prediction Accuracy of recognition00.20.40.60.81Trial Accuracy Self Accuracy and Setsize accuracyBB'BBB'BBCBBB
7. Conclusion
▫There is technical possibility to define DoseFocussettings for NSO based on image morphology and MMSE classification.
▫Current development targeting offline Engineering tool, which could be activated as part of troubleshooting or casual stability check procedure
▫To confirm results on larger statistical sets need 5 FE per NSO tool with reduced Dose/Focus step size.
▫SW sustaining –all major HW/SW changes/PM on NSO should be followed by run of 3 relevant FE with defined stepping in Focus and Dose, followed by set training on my SW for baseline establishment.
▫CDSEM needs –per IDP add 5 image MP per corner for relevant statistics. Estimated tool time per recipe –additional 12 sec.
▫Decision should be taken by Eng GL’s for further steps.