driven ECO Subramanyam Sripada Song Chen Synopsys Inc. Mar 16, - - PowerPoint PPT Presentation

driven eco
SMART_READER_LITE
LIVE PREVIEW

driven ECO Subramanyam Sripada Song Chen Synopsys Inc. Mar 16, - - PowerPoint PPT Presentation

Efficient Incremental Flow for signoff- driven ECO Subramanyam Sripada Song Chen Synopsys Inc. Mar 16, 2017 Agenda Background Motivation Approach Results Design/Complexity Projections SoC Transistor Count Trend Scenario


slide-1
SLIDE 1

Efficient Incremental Flow for signoff- driven ECO

Subramanyam Sripada Song Chen Synopsys Inc. Mar 16, 2017

slide-2
SLIDE 2

Agenda

  • Background
  • Motivation
  • Approach
  • Results
slide-3
SLIDE 3

Design/Complexity Projections

1 2 3 4 5 6 7 8 9 10 2013 2014 2015 2016 2017 2018 2019 2020

Billion Transistors

SoC Transistor Count Trend (Logic + SRAM)

Source: ITRS 2013

3X Growth 2015-2020

5 9 10 12 15 7 11 15 24 27

65nm 40nm 28nm 20nm FinFET

Scenario Growth at Advanced Nodes

Modes Corners

Source: Synopsys Customer & Partner Data

slide-4
SLIDE 4

Signoff-driven timing closure

Implementation Timing-driven optimization for critical scenarios hierarchically Signoff-driven ECO Signoff-accurate, physically-aware, all-scenario ECO Place and Route

slide-5
SLIDE 5

Implementation ECO vs Signoff ECO

  • Hierarchical vs Flat full-chip

– Implementation ECO is usually done block-level and top-level whereas signoff ECO is typically performed flat but MIM-aware, physically-aware

  • Timing scenarios

– Number of timing scenarios is signoff ECO is full combination of modes and corners

  • Constraints

– Timing constraints can be different between implementation and signoff

slide-6
SLIDE 6

Multiply Instantiated Modules

  • MIM is very common:

– Implementation ECO typically honors MIM as it sees single context – Signoff ECO needs to provide options to honor MIM or break away to meet final timing

  • Flat analysis eliminates MIM pessimism
slide-7
SLIDE 7

Motivation

  • How to perform signoff ECO with all timing

scenarios on huge (several hundred million instance) designs with high accuracy, efficient runtime, computational resources?

  • Efficient incremental flow [this presentation]
  • Violation driven circuit reduction [next

presentation]

slide-8
SLIDE 8

Traditional signoff ECO

Parasitic extraction (hierarchical) Signoff STA/ECO (full chip) Place and Route, ECO (hierarchical)

slide-9
SLIDE 9

Previous Attempts

  • Previous Attempts:

– All functionality in single process – Perform post-route-ECO with a signoff timer accuracy

  • Limited success due to flat analysis on huge

designs

– Additional final flat signoff run

slide-10
SLIDE 10

Incremental signoff ECO

Incremental parasitic extraction (StarRC) Incremental signoff STA and ECO (PrimeTime) Incremental Place and Route (IC Compiler II) 3 1 2

slide-11
SLIDE 11

Incremental signoff ECO

  • PrimeTime runs full chip signoff STA
  • For each iteration,
  • 1. PrimeTime performs physical ECO on all scenarios
  • Provides ECO changelist for P&R for each block [2]
  • 2. IC Compiler II performs incremental hierarchical P&R
  • Provides incremental hierarchical database to StarRC
  • 3. StarRC performs incremental hierarchical parasitic

extraction

  • Provides incremental SPEF to PrimeTime
  • 4. PrimeTime performs incremental full chip signoff

STA

slide-12
SLIDE 12

Incremental signoff ECO

  • All communication is done via binary database

– Ascii versions are available for advanced users and for mix-and-match – Incremental SPEF has asymmetric coupling

  • All communication ensures data integrity via

signature mechanism

slide-13
SLIDE 13

Results

Design Number

  • f

endpoints Outliers > 2ps Outliers > 1% of capture clock period A 22142 39 B 5552 1 C 322364 D 401534 13 E 6240 2 F 2071132 26

3X performance improvement over full STA analysis

slide-14
SLIDE 14

Summary

  • There is a great need for optimizing the signoff

ECO process

– With all scenarios, signoff constraints and full-chip

  • Proposed a fully incremental signoff ECO

system

– Presented approach and results – Results have improved quite a bit since the paper