4th Annual SFR Workshop, Nov. 14, 2001

Transcription

1 4th Annual SFR Workshop, Nov. 14, :30 9:00 Research and Educational Objectives / Spanos 9:00 9:45 CMP / Doyle, Dornfeld, Talbot, Spanos 9:45 10:30 Plasma & Diffusion / Graves, Lieberman, Cheung, Haller 10:30 10:45 break 10:45 12:00 Poster Session / Education, CMP, Plasma, Diffusion 12:00 1:00 lunch, CITRIS Presentation / King 1:00 1:45 Lithography / Spanos, Neureuther, Bokor 1:45 2:30 Sensors & Controls /Aydil, Poolla, Smith, Dunn, Cheung, Spanos 2:30 2:45 Break 2:45 4:30 Poster Session / all subjects 3:30 4:30 Steering Committee Meeting in room 373 Soda 4:30 5:30 Feedback Session

2 2 Chemical Mechanical Planarization SFR Workshop & Review David Dornfeld, Fiona Doyle, Kameshwar Poolla, Costas Spanos, Jan Talbot Berkeley, CA

3 3 CMP Milestones September 30 th, 2001 Build integrated CMP model for basic mechanical and chemical elements. Develop periodic grating metrology (Dornfeld, Doyle, Spanos,Talbot). Model Outline Progressing- initial chemical and mechanical modules in development; Simulation software prototype; Matlab-based model in development. September 30 th, 2002 Integrate initial chemical models into basic CMP model. Validate predicted pattern development. Integrated comprehensive process simulation with reduced parametric-metrology tuned MatLabbased model (Dornfeld, Doyle, Poola, Spanos,Talbot). September 30 th, 2003 Develop comprehensive chemical and mechanical model. Perform experimental and metrological validation. (Dornfeld, Doyle, Spanos, Talbot)

4 4 Focus of this presentation Review progress on development of reduced parameter model development of the integrated model simulation tools demonstration (time permitting) understanding the role of chemistry in CMP Process monitoring/scratch detection activity Full-profile metrology for CMP modeling Details of these and other key areas in posters!

5 5 CMP Modeling Roadmap Objectives from Industrial Viewpoint - VMIC 2001 Models are not reliable enough to be used as verification of process Usefulness of modeling is the ability to give feedback for what-if scenarios (predicting polishability of new mask designs) in lieu of time-consuming DOE tests Models should give some performance prediction for realistic, heterogeneous pattern effects Models should predict not only wafer scale phenomena but also have some capability to capture feature/chip scale interaction

6 6 Interaction Scale Global Wafer-Level Chip/Die-Level Feature/ Particle- Level Current Status of Modeling Efforts Industrial Viewpoint - VMIC 2001 Modeling Aspect Environmental Issues Material Removal Slurry/Abrasive/Pad Material Removal Chemistry Kinematics Pad Stress Dynamics Hydrodynamics Pattern Density Effects Oxide Pattern Density Effects Copper Pattern Density Effects STI Abrasive Particle Interaction Estimated Completion < 5 % 50% 50% 90% < 5% 30% 90% 20% 40% 50%

7 7 Roadblocks for Modeling Industrial Viewpoint - VMIC 2001 Multi-scale (wafer-, die-, feature-level) interactions must be integrated for global CMP modeling to be useful Linkage of models to upstream (deposition, etc.) and downstream (lithography,etc.) processes Models need to address defectivity New materials, consumables (pad, slurry, etc.) modeling and characterization

8 8 Modeling Strategy Two level approach: process mechanics-based & time-based reduced parameter set simulation Synthesis of approaches into Comprehensive Simulation Tool for CMP processes Link models to upstream (deposition, etc.) and downstream (lithography,etc.) processes as well as cross-dependencies in the process Models need to address defectivity New materials, consumables (pad, slurry, etc.) modeling and characterization

9 9 Reduced-parameter model Motivation & Approach Objectives Build comprehensive CMP simulation tool Simulation tool should predict pad and slurry and time history of wafer topography The Challenge Complexity: large numbers of unmeasurable parameters in model need to be estimated Limited availability of metrology to tune model Approach Evaluate sub-system sensitivity to drive model reduction Confirm sensitivities experimentally

10 10 Comprehensive Simulation Tool for CMP processes Build MATLAB based CMP model Pad pressure Spin speeds Slurry comp Pad state Initial topography CMP Process Removal rate Slurry state Pad state Topography Parameter tuning & adaptation RTR Metrology Tools

11 11

12 12 Research Plan Initial focus on Copper process- mechanical model + copper chemistry model 8/01 basic computational engine in place (done) 1/02 sensitivity studies complete 8/02 data driven model validation

13 13 Mechanism-based model Architecture of model previously introduced Includes sensitivity to consumables (slurry/abrasive, pad, etc.) Simplified representation of chemical effect Java simulation software prototype

14 14 Interactions between Input Variables Four Interactions: Wafer-Pad Interaction; Pad-Abrasive Interaction; Wafer-Slurry Chemical Interaction; Wafer-Abrasive Interaction Velocity V Chemically Influenced Wafer Surface Wafer Vol Abrasive particles in Fluid (All inactive) Polishing pad Pad asperity Source: J. Luo and D. Dornfeld, IEEE Trans: Semiconductor Manufacturing,, 2001 Abrasive particles on Contact area with number N Active abrasives on Contact area

15 15 Java Implementation of CMP Optimization Software based on the Material Removal Model

16 16 Design of consumables - Pad Example Prototype surface, 20X Prototype surface, 55X Design software interface for prototype pad surface; geometry of individual elements, pitch and mechanical properties are Variable, courtesy of J. F. Luo, LMA, 2001

17 17 Conclusions A comprehensive model is developed to explain the material removal mechanism in CMP The roles and interactions of polishing pad, slurry and wafer are being identified using this comprehensive model MRR formulations considering the integrated effects of input variables are developed and verified Future Work Further experimental verification of the model needed Model-based process optimization (e. g. using Java) Process design capabilities (e.g. pad, abrasive, chemistry)

18 18 Chemical Aspects of CMP Role of Chemistry Chemical and electrochemical reactions between material (metal, glass) and constituents of the slurry (oxidizers, complexing agents, ph) Dissolution and passivation Solubility Adsorption of dissolved species on the abrasive particles Colloidal effects Change of mechanical properties by diffusion & reaction of surface

19 19 Cu T =10-5 ; L T =10-2 E, V vs. SHE AERATED Copper-H 2 O-Glycine System Cu ph E, mv vs. SHE CuL + Cu CuL 2 CuO Cu 2 O CuO2 2- E, m V vs. SHE ph= ph=12 ph= E-03 1E-02 1E-01 1E+00 1E+01 1E+02 1E+03 i, A/cm 2 ph=9 ph=10 ph=11 ph= E-08 1E-07 1E-06 1E-05 1E-04 1E-03 1E-02 1E-01 i, A/cm 2 DE-AERATED

20 20 In-situ Polarization Diagrams E, m V vs. SHE ph=4 1E-08 1E-07 1E-06 1E-05 1E-04 1E-03 1E-02 1E-01 i, A/cm 2 E, mv vs. SHE E, mv vs. SHE ph= E-08 1E-07 1E-06 1E-05 1E-04 1E-03 1E-02 1E-01 i, A/cm ph= E-08 1E-07 1E-06 1E-05 1E-04 1E-03 1E-02 1E-01 i, A/cm 2

21 M Glycine Solutions with H 2 O 2 nm/min Dissolution Rate Polish Rate ph=4 EOC, mv vs. SHE H2O2 wt% H2O2 wt% nm/min Dissolution Rate Polish Rate ph=9 EOC, V vs. SHE H2O2 wt% H2O2 wt%

22 22 Conclusions and Future Goals Polarization results are well correlated with potential-ph diagrams No significant changes in in-situ polarization for active behavior Mechanical components significantly affected in-situ polarization for active-passive behavior Glycine (complexing agents) may enhance the polishing efficiency The preliminary results points out the significant role of H 2 O 2 in the dissolution and passivation of copper in glycine The nature of the passivation at high H 2 O 2 should be investigated by considering the possible formation of higher copper oxides

23 23 End Point Detection Methods Methods Physics D,I G,L App Optical Reflectance, Absorption D L Cu/STI/ILD Thermal Temperature Sensing I L Cu/STI/ILD Electrical Motor Current I G Cu/STI Microphone Intensity/ Freq. Analysis D G Cu/STI/ILD Force Friction Force D/I G Cu/STI Acoustic Emission Acoustic Waves D G Cu/STI D: Direct, I: Indirect G: Global, L: Local

24 24 Experimental Setup DAQ System AE sensor Signal Conditioning A/D (2MHz) AE Force Sensor Signal Conditioning A/D (20KHz) COF CMP Tool Test Wafers Slurry Type Pad Type Polishing Conditions CMP Tester by CETR Cu(1500Å)/Ta(250Å)/Ox(5000Å) Alumina based 5003 with 2.5% of H 2 O 2 IC1000 Polyurethane Pad Down Force : 30 ~ 40N Table RPM: 60RPM

25 25 AE Data / Friction Data AE Signals(V) A B C D E COF A B Ta Ox Cu Ta Ox Cu C D E Time(sec) Time(sec) A B C D E Cu Ta Ox Great Correlation Bet. AE and Friction Data

26 26 Scratch Evolution during CMP Contamination from Pad Conditioner Agglomerated Slurry Particles Pad conditioner Wafer carrier Wafer Polishing Pad Polishing Pad Agglomerated slurry particles Embedded diamond grit Wafer Polishing Pad Diamond grits that are pulled out of the conditioning pad can embed themselves in the polishing pad and cause wafer scratching. Slurry particles can agglomerate and act as large contaminates that can scratch the wafer surface.

27 27 Silicon Wafer Scratch Evolution 2000 nm 0 nm t = 0 s t = 45 s Subsurface damage caused during the initial scratch can propagate during subsequent polishing. t = 240 s t = 135 s t = 90 s

28 28 Roughness & Depth Scratch Depth Scratch Depth (nm) Time (sec) Silicon Oxide Scratch Bottom Roughness (nm) Scratch Bottom Roughness (Ra) Time (sec) Silicon Oxide Top view of AFM image. Roughness measurements were made of the bottom (valley) of the scratches and monitored during the polishing.

29 29 Full Profile CMP Results x x x x 10 4 SEM AFM Scatterometry Extracted profiles match SEM pictures within 5nm Scatterometry is non-destructive, faster and more descriptive than competing methods.

30 30 Dishing and Erosion Effects Dishing Input Erosion Output Oxide erosion and Cu dishing in the Cu damascene process are limiting factors. HSPICE simulation shows that 100Å (~5%) Cu loss may degrade interconnect performance by ~4.6% for the 0.25µm technology.

31 31 Key Ideas on Copper CMP Modeling Calculate total copper loss by adding up losses from two ways Copper loss due to oxide erosion can be modeled using an improved effective pattern density model Local copper dishing is correlated with the metal line width, pad asperity, slurry chemistry, etc. This can be modeled by the new concept of dishing radius, which can be extracted from our design of experiments oxide pad Dishing Radius Cu substrate