UC San Diego Computer Engineering VLSI CAD Laboratory UC San Diego Computer Engineering VLSI CAD Laboratory UC San Diego Computer Engineering VLSI CAD Laboratory UC San Diego Computer Engineering VLSI CAD Laboratory UC San Diego Computer Engineering VLSI CAD Laboratory UC San Diego Computer Engineering VLSI CAD Laboratory UC San Diego Computer Engineering VLSI CAD Laboratory  Our testcases: 2 large designs created by assembling smaller ones  “Mixed”: 2mm x 2mm, 756K cells  “OpenRisc8”: 2.8mm x 3mm, 423K cells + SRAM  Our testcases: 2 large designs created by assembling smaller ones  “Mixed”: 2mm x 2mm, 756K cells  “OpenRisc8”: 2.8mm x 3mm, 423K cells + SRAM  Imperfect STI CMP causes functional and parametric yield loss  Traditionally tile-based fill used with expensive reverse etchback to control post-CMP topography variability  Our fill insertion approach focuses on: (1) oxide density variation minimization, and (2) nitride density maximization  Large nitride fill features contribute to nitride and oxide densities, small ones to nitride only  shape fill to control both densities  Proposed max. nitride fill insertion with holes to control oxide density and achieve high nitride density  Results indicate significant decrease in oxide density variation and increase in nitride density over tile-based fill.  CMP simulation shows superior CMP characteristics, planarization window increases by 17%, and step height decreases by 9%.  Imperfect STI CMP causes functional and parametric yield loss  Traditionally tile-based fill used with expensive reverse etchback to control post-CMP topography variability  Our fill insertion approach focuses on: (1) oxide density variation minimization, and (2) nitride density maximization  Large nitride fill features contribute to nitride and oxide densities, small ones to nitride only  shape fill to control both densities  Proposed max. nitride fill insertion with holes to control oxide density and achieve high nitride density  Results indicate significant decrease in oxide density variation and increase in nitride density over tile-based fill.  CMP simulation shows superior CMP characteristics, planarization window increases by 17%, and step height decreases by 9%.  Shallow trench isolation (STI) mainstream inter-device electrical isolation technique used in all designs today  Chemical mechanical planarization (CMP) critical process step in STI to remove excess deposited oxide  Imperfect CMP  Loss of functional and parametric yield  Post-CMP topography variation  process (esp. defocus) variation  CMP is pattern dependent  fill can reduce post-CMP variability  Traditional fill tile based, used with expensive reverse etchback  Our goal: fill insertion for superior post-CMP topography characteristics without use of reverse etchback  Results show proposed method achieves significantly better density and superior post-CMP topography (as predicted by CMP simulation)  Shallow trench isolation (STI) mainstream inter-device electrical isolation technique used in all designs today  Chemical mechanical planarization (CMP) critical process step in STI to remove excess deposited oxide  Imperfect CMP  Loss of functional and parametric yield  Post-CMP topography variation  process (esp. defocus) variation  CMP is pattern dependent  fill can reduce post-CMP variability  Traditional fill tile based, used with expensive reverse etchback  Our goal: fill insertion for superior post-CMP topography characteristics without use of reverse etchback  Results show proposed method achieves significantly better density and superior post-CMP topography (as predicted by CMP simulation)  In STI, substrate trenches filled with oxide surround devices or group of devices that need to be isolated  Relevant process steps: 1.Diffusion (OD) regions covered with nitride 2.Trenches created where nitride absent and filled with oxide 3.Chemical Mechanical Polishing (CMP) to remove excess oxide over nitride (overburden oxide)  Unfortunately, CMP is not perfect  Planarization window: Time window in which CMP may be stopped. Stopping sooner leaves oxide over nitride, stopping later polishes silicon under nitride. Larger window desirable.  Step height: Oxide thickness variation; quantifies oxide dishing. Smaller step height desirable.  Deposition bias: Oxide over nitride deposited with slanted profile  Oxide features are “shrunk” nitride features  CMP is pattern dependent  Fill insertion improves planarization window and step height  In STI, substrate trenches filled with oxide surround devices or group of devices that need to be isolated  Relevant process steps: 1.Diffusion (OD) regions covered with nitride 2.Trenches created where nitride absent and filled with oxide 3.Chemical Mechanical Polishing (CMP) to remove excess oxide over nitride (overburden oxide)  Unfortunately, CMP is not perfect  Planarization window: Time window in which CMP may be stopped. Stopping sooner leaves oxide over nitride, stopping later polishes silicon under nitride. Larger window desirable.  Step height: Oxide thickness variation; quantifies oxide dishing. Smaller step height desirable.  Deposition bias: Oxide over nitride deposited with slanted profile  Oxide features are “shrunk” nitride features  CMP is pattern dependent  Fill insertion improves planarization window and step height Background CMP Fill for Reduced STI Variability Introduction Methodology Experimental Study Conclusions Student:Puneet Sharma Advisors:Andrew B. Kahng, Alex Zelikovsky Theme: System-Level Living Roadmap To appear in ICCAD’06 Si Oxide Nitride αα Oxide Nitride Shrinkage = α Top View Density Improvements for the Proposed Fill Insertion Method Testcase: MixedTestcase: OpenRisc8 Unfilled Tiled 0.5µ/0.5µTiled 1.0µ/0.5µTiled 1.0µ/1.0µ Proposed Tiling-based fill Fill with proposed approach Inserted fill Inserted fill Design features Design features Layout After Fill InsertionSTI CMP Simulation Results Final Max. Step Height (nm) 50.4Proposed 44.7Tiled 0.5µ/0.5µ 42.7UnfilledOpenRisc8 53.6Proposed 46.5Tiled 0.5µ/0.5µ 45.3UnfilledMixed Planarization Window (s) Fill ApproachTestcase Before CMP After Perfect CMP Top view of layout Nitride Oxide deposited over nitride Area available for fill insertion  Objectives of fill insertion:  Minimize oxide density variation  Overburden oxide uniformly removed from all regions  Enlarges planarization window as oxide clears simultaneously  Maximize nitride density  Enlarges planarization window as nitride polishes slowly  Shrinkage  Oxide density depends on nitride density  Insert fill (nitride features) to control nitride and oxide densities  Dual-objective problem formulation: Insert dummy fill  Given: STI regions where fill can be inserted, shrinkage α  Constraint: No DRC violations  Objectives: (1)min. oxide density variation, (2)max. nitride density  Minimize oxide density variation  Use previously proposed LP-based solution  Inputs: min. oxide density (|Oxide Min | and and max. oxide density (|Oxide Max |) per tile  Output: target oxide density (|Oxide Target |) per tile  For min. oxide density: shrink nitride features by α  For max. oxide density: insert max. fill, shrink nitride features by α  Nitride maximization problem formulation: Insert dummy fill  Given: STI regions where fill can be inserted, shrinkage α  Constraint: Target oxide density per tile  Objective: max. nitride density  Solution based on case analysis  Case 1: |Oxide Target | = |Oxide Max |  Insert max. fill  Case 2: |Oxide Target | = |Oxide Min |  Need to insert fill that does not increase oxide density  Naïve approach: insert fill rectangles of shorter side < α  Clever approach: perform max nitride fill then dig square holes of min. allowable side β  Gives higher nitride:oxide density ratio  No oxide density due to nitride in rounded square around a hole  Need to cover max. nitride fill with rounded squares  Our approach:  Approximate rounded squares with inscribed hexagons  Proposed optimal algorithm to cover rectilinear nitride fill regions with hexagons  Case 3: |Oxide Min | < |Oxide Target | < |Oxide Max |  Rounded squares approximated by circumscribed hexagons  Pack bulk nitride with hexagons or squares and dig holes at the center of approximated rounded square  Stop when:  |Oxide Target | = |Oxide Min |  Continue with Case 2  |Oxide Target | = |Oxide Max |  No more holes needed  Objectives of fill insertion:  Minimize oxide density variation  Overburden oxide uniformly removed from all regions  Enlarges planarization window as oxide clears simultaneously  Maximize nitride density  Enlarges planarization window as nitride polishes slowly  Shrinkage  Oxide density depends on nitride density  Insert fill (nitride features) to control nitride and oxide densities  Dual-objective problem formulation: Insert dummy fill  Given: STI regions where fill can be inserted, shrinkage α  Constraint: No DRC violations  Objectives: (1)min. oxide density variation, (2)max. nitride density  Minimize oxide density variation  Use previously proposed LP-based solution  Inputs: min. oxide density (|Oxide Min | and and max. oxide density (|Oxide Max |) per tile  Output: target oxide density (|Oxide Target |) per tile  For min. oxide density: shrink nitride features by α  For max. oxide density: insert max. fill, shrink nitride features by α  Nitride maximization problem formulation: Insert dummy fill  Given: STI regions where fill can be inserted, shrinkage α  Constraint: Target oxide density per tile  Objective: max. nitride density  Solution based on case analysis  Case 1: |Oxide Target | = |Oxide Max |  Insert max. fill  Case 2: |Oxide Target | = |Oxide Min |  Need to insert fill that does not increase oxide density  Naïve approach: insert fill rectangles of shorter side < α  Clever approach: perform max nitride fill then dig square holes of min. allowable side β  Gives higher nitride:oxide density ratio  No oxide density due to nitride in rounded square around a hole  Need to cover max. nitride fill with rounded squares  Our approach:  Approximate rounded squares with inscribed hexagons  Proposed optimal algorithm to cover rectilinear nitride fill regions with hexagons  Case 3: |Oxide Min | < |Oxide Target | < |Oxide Max |  Rounded squares approximated by circumscribed hexagons  Pack bulk nitride with hexagons or squares and dig holes at the center of approximated rounded square  Stop when:  |Oxide Target | = |Oxide Min |  Continue with Case 2  |Oxide Target | = |Oxide Max |  No more holes needed β α αNitride Hole No oxide in this region Top View Max. nitride fill (purple rectilinear region) Optimal hexagon cover (beige) Holes created in nitride at centers of hexagons ensure zero oxide contribution Failure to clear oxideNitride erosionOxide dishing Key Failures Caused by Imperfect CMP