Standard Cell Characterization Considering Lithography Induced Variations Enable Lithoaware Static T

1
Standard Cell Characterization Considering
Lithography Induced Variations - Enable
Litho-aware Static Timing Analysis

• Ke Cao()(), Sorin Dobre(), Jiang Hu()
• Dept of Electrical Engineering, Texas AM
  University
• Qualcomm, Inc.

2
Acknowledgements

• We would like to thank the following people for
  their contribution for this work.
• Foua Vang standard cell library support.
• Kasim Mahmood Hspice model support.
• Dan Perry CAD support.
• Riko Radojcic technical suggestions and
  discussion.
• Matt Nowak management support.
• Bill Graupp tool support.

3
Outline

• Motivation.
• Methodology details.
• Current approaches.
• Test pattern.
• Dummy poly insertion.
• Modeling poly gate shapes.
• Proposed flow.
• Experimental results.
• Conclusion and future work.

4
Motivation

• Shape deviation due to Litho effects are not
  negligible.
• Litho effects are systematic.
• Current process corner methodology is pessimistic.

Shape contours generated by Mentor Graphics
LFD.
5
Current Approach

• Issues
• It is intrusive to the current flow.
• Litho simulation is expensive on block/chip
  level.
• Standard cell needs to be characterized for
  almost every instantiation.
• Design optimization is not straightforward.
• It is not clear how this flow can be scaled to
  SSTA.

6
Reasonable assumptions

• Interconnect timing variation.
• Still dominated by thickness variation due to CMP
  in 65nm.
• May not be true going forward.
• Litho-induced standard cell timing variation is
  poly gate-length dominated.
• Gate width and diffusion variations impacts on
  timing are small.
• Can be easily considered in our litho flow.
• Litho interaction for poly gate length in a row
  based standard cell block comes from the
  neighboring cells in the same row.
• Space for poly gates are large for gates in
  different rows.
• Poly routing on top and bottom of the cells are
  being phased out.
• Poly extension over diffusion provides enough
  space for a reasonably good OPC algorithm to
  recover.

7
Litho variation test structures

• Typical Litho interaction range is 0.6um-1um in
  65nm node.
• The closest neighboring shapes have the most
  influences.
• Shapes with the fixed closest neighbors are not
  sensitive to proximity topology.

8
Dummy poly insertion in standard cells

• Impact
• Providing the fixed closest neighbor for poly
  gate at boundary.
• Fixed litho aware timing model for standard
  cells.
• Cost
• Restriction field poly for cells should not be
  at cell boundary.
• No other DRC violation.
• No LVS error dummy poly is not a device.
• No area penalty.
• Gap between diffusions of different cells is
  usually greater than minimum.
• Cells have to be on grid, which provides
  additional white space.

9
Shape analysis

• Calculate Ids for Ion and Ioff.
• Match Ion and Ioff in lookup table with same W to
  get Leff for timing and leakage.
• Back annotate Leff_timing

10
Litho-aware standard cell characterization
Lookup table for Ion and Ioff
Calculate Ids for Ion and Ioff
Generate effective L for timing and leakage
Litho Simulation
Cell Characterization
Back annotate into cell netlist
.lib and .db for standard cell lib

• Look up table can be generated separately.
• Two netlists are generated for each standard
  cell,one for timing characterization and one for
  leakage characterization.

11
Proposed flow

• Advantage
• Design flow does not change. Litho simulations
  are not visible to designers. No issue with
  design optimization.
• Litho simulation happens off-line, in standard
  cell characterization phase. Each cell is
  characterized ONCE.
• Can afford to run litho simulation on standard
  cells in the most accurate mode.
• Portable to SSTA flow.

12
Experimentstandard cell timing and leakage
variability
13
Experiment Block level timing variability

• 400 most timing critical paths in 65nm 250MHz
  block.
• Variability of paths timing decreased by an
  averageof 320ps.

14
Conclusion

• Simulation data shows litho-aware timing analysis
  helps to peel out the litho-induced systematic
  timing variations from the corner models to
  reduce timing uncertainty.
• The litho-aware timing analysis flow does not
  have to impact the traditional flow, even though
  litho effects are proximity dependent.
• The scheduling impact of enabling the litho-aware
  timing analysis is minimal even if we use the
  most sophisticated litho simulation tool for most
  accurate results.
• Future works include interconnect litho
  variation timing impact, litho aware SSTA.

15
Thank you!