Inficon Holding Ag · SIX:IFCN
Inficon — Technical Presentation (February 2001)
33 slides from Inficon's technical presentation, February 2001
Slide 1: A Comprehensive Analysis of Statistical and Model-Based Overlay Lot Disposition Methods
A Comprehensive Analysis of Statistical and Model-Based Overlay Lot Disposition Methods David Crow, Ken Flugaur (Cypress Semiconductor) Joe Pellegrini, Etienne Joubert (New Vision Systems)
Slide 2: How to do a better job of figuring out overlay errors
How to do a better job of figuring out whether or not the overlay errors on a recently developed wafer are bad enough to justify the cost and general inconvenience of rework.
Slide 3: How I Learned to Love the Winter in Bloomington, Minnesota
How I Learned to Love the Winter in Bloomington, Minnesota Trivia Question: What is special about the temperature “-40 degrees”?
Slide 4: Presentation Overview
Presentation Overview ● Lot Dispositioning Background ● The Folly of Common Statistical Methods ● Model Based Overlay Dispositioning ● Experiment Design ● Experiment Results ● Conclusions
Slide 5: Overview of the Overlay Lot Disposition Decision
Overview of the Overlay Lot Disposition Decision MONO-LITH RT ? PASS REWORK FAIL
Slide 6: The Ideal Lot Disposition Algorithm
The Ideal Lot Disposition Algorithm ? EV(Reworked Lot) – EV(Non-reworked Lot) > Cost of Rework
Slide 7: The Yield Threshold Simplification
The Yield Threshold Simplification ? EV(Reworked Lot) – EV(Non-reworked Lot) > Cost of Rework Projected Maximum Error > Yield Loss Threshold
Slide 8: What is Projected Maximum Error?
What is “Projected Maximum Error”?
Slide 9: Methods for Calculating Projected Maximum Error
Methods for Calculating Projected Maximum Error ● Mean + 3 Sigma ● Maximum Absolute Value ● Model Based
Slide 10: The Problem with Mean + 3 Sigma
The Problem with Mean + 3 Sigma Scale_X = +1ppm Mean = 0 σ = SQRT((X1^2+X2^2)/N) σ = 75nm -75nm σ = 225nm Mean + 3σ +75nm Max. Observed Error = 75nm x1 = -75nm Max. Possible Error = Scale_X x Radius x2 = +75nm Max. Possible Error = 1ppm x 100 mm Max. Possible Error = 100nm
Slide 11: The Problem with Absolute Maximum Value
The Problem with Absolute Maximum Value Metrology Noise Contribution 12 10 8 6 4 2 0 0 10 20 30 40 50 Effective Samples
Slide 12: Model Based Algorithm
Model Based Algorithm Effective Effective Wafer Field Radius Radius Grid + + Intrafield Errors Errors + Asystematic + TOTAL MODELED Errors ERROR
Slide 13: Theory Predicted Results
Theory Predicted Results Model Based Maximum Absolute Mean + 3σ Error
Slide 14: Experimental Design: Goals
Experimental Design ● Goals 1. Validate Model Based theory with real data Mean + 3 sigma > Model Based > Max Vector 2. Predict the benefit of Model Based Disposition Rework reduction opportunity
Slide 15: Experimental Design: Tools
Experimental Design ● Metrology Tool ● KLA 5200 ● Modeling Software ● MONO-LITH Version 5.2 ● Wafer Track / Exposure Tool ● DNS80B & ASML /100 ● DNS80B & ASML /200 ● DNS80B & ASML /300 ● DNS200W & ASML /700
Slide 16: Experimental Design: Matching Schemes
Experimental Design I-Line DUV A-B Matching A-A Matching First Half of Loosely Controlled Tightly Controlled Manufacturing Process Second Half of Special Case I Special Case II Manufacturing Process
Slide 17: Experimental Design: Sampling Plan
Experimental Design ● Sampling plan ● Product type: high volume SRAM ● Overlay measurements: ● 3 wafers / lot ● 9 fields / wafer ● 5 sites / field 135 total measurements per lot
Slide 18: Experimental Results: Output Metrics
Experimental Results ● Output metrics ● Overlay performance X OVP = * 100 % Overlay Spec Limit ● Rework Rate # X > Overlay Spec Limit R ework = * 100 % n
Slide 19: Experimental Results: I-line, A-B Matching
Experimental Results: I-line, A-B Matching Process Layer n Metric Mean + 3σ Model Max 156 OVP 62% 54% 51% First half Rework 2% 1% 0% OVP 72% 70% 65% Second half 104 Rework 8% 2% 0%
Slide 20: I-line, A-B Matching OVP: First Half Layer
Experimental Results: I-line, A-B Matching First Half Layer OVP 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Cumulative Probability Mean+3sigma Model Error Max Vector spec
Slide 21: I-line Layer, A-B Matching: Rework Reduction
Experimental Results: I-line Layer, A-B Matching Median Comparison Rework Reduction 14% Gain 1% Benefit 40% 50% 60% 90 % 95% 10 0% Loosely Controlled Process: Model Based method shows significant improvement on layer where systematic correctable terms are not tightly controlled.
Slide 22: I-line, A-B Matching OVP: Second Half Layer
Experimental Results: I-line, A-B Matching Second Half Layer OVP 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Cumulative Probability Mean+3Sigma Model Error Max Vector spec
Slide 23: I-line, A-B Matching: Rework Reduction
Experimental Results: I-line, A-B Matching Median Comparison Rework Reduction 6% Benefit 4% Gain 40% 50% 60% 90% 95% 100% Special Case I: Reference layer can not be reworked resulting in higher variation of systematic correctable terms at the target layer.
Slide 24: Experimental Results: DUV, A-A Matching
Experimental Results: DUV, A-A Matching Process Layer n Metric Mean + 3σ Model Max OVP 47% 46% 46% First half 174 Rework <1% <1% <1% OVP 67% 61% 60% Second half 153 Rework 3% 1% 3%
Slide 25: DUV, A-A Matching OVP: First Half Layer
Experimental Results: DUV, A-A Matching First Half Layer OVP 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Cumulative Probability Mean+3Sigma Model Error Max Vector Spec Limit
Slide 26: DUV, A-A Matching: Rework Reduction
Experimental Results: DUV, A-A Matching Median Comparison Rework Reduction 3% Gain 40% 50% 60% 95% 100% Tightly Controlled Process: Systematic correctable terms are well controlled, resulting in small benefits observed with Model Based dispositioning
Slide 27: DUV, A-A Matching: Second Half Layer
Experimental Results: DUV, A-A Matching Second Half Layer OVP 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Cumulative Probability Mean+3Sigma Model Error Max Vector Spec Limit
Slide 28: DUV, A-A Matching: Rework Reduction
Experimental Results: DUV, A-A Matching Median Comparison Rework Reduction 8% Gain 40% 50% 60% 90% 95% 100% Special Case II: Known edge effect causes results that do not agree with theory ! Max Error > Model Based
Slide 29: Identification of Edge Wafer Effect
Identification of Edge Wafer Effect
Slide 30: Rework Reduction Opportunity
Experimental Results: Rework Reduction Opportunity ● Predictions based on normalized overlay results ● Since product flow conditions will vary and the incidence of rework for some cases is very low, our projections must be characterized using pessimistic and optimistic assumptions Type of Estimate O
Slide 31: Rework Reduction Opportunities
Experimental Results: Rework Reduction Opportunity Rework Reduction Opportunities OVP
Slide 32: Conclusions
Conclusions ● Product data supports the benefit of Model Based Lot dispositioning ● The amount benefit varies between layers ● Greatest benefit observed with A-B matching. This matching scheme has more opportunity for systematic correctable errors to exist. ● Smaller benefit observed with A-A matchi
Slide 33: Thank You! Questions?
Thank You! Questions? S&S CYPRESS
Inficon Holding — Technical Presentation (February 2001) on 6ix