1. QoR impact after logic synthesis

In this study, Focus was used to generate timing exceptions from RTL. These exceptions were verified using a property checking tool. The exceptions were read into a synthesis tool. QoR (WNS, total number of violations and instance count) was compared after logic synthesis with and without Focus exceptions. Run time data is for a Linux platform.
 

Design 1: 10K instances; 2 clocks

Exception Generation  

Description

False Paths

Multicycle Paths

FishTail Focus

Generated

5

114

Focus Run Time

2 hours

6 hours 26 minutes

Logic Synthesis Data      

  Without Exceptions   

  With Exceptions   

Worst Negative Slack (WNS)

- 0.57 ns

-0.26 ns

No. of Setup Violations

688

272

Cell Area (1 nd202as = 2 C.U.A)

31,122

31,001

Total Instance Count

10,796

10,613

Compile Time

186 seconds

143 seconds

Design 2: 20K instances; 2 clocks

Exception Generation

Description

False Paths

Multicycle Paths

FishTail Focus

Generated

39

29

Focus Run Time

3 minutes

7 hours 22 minutes

Assertion Verification

Verified

39

29

Failed

0

0

Timed-out

0

0

Logic Synthesis Data

Without Exceptions

With Exceptions

Worst Negative Slack (WNS)

-2.75 ns

+0.02 ns

No. of Setup Violations

729

0

Cell Area (1 nd202as = 2 C.U.A)

203,401

201,816

Total Instance Count

19,771

17,265

Design Compiler Run Time

4 minutes

2 minutes

Design 3: 50K instances; 1 Clock

Exception Generation

Description

False Paths

Multicycle Paths

FishTail Focus

Generated

1059

93

Focus Run Time

2 hours 21 minutes

2 hours 48 minutes

Assertion Verification

Verified

1059

93

Failed

0

0

Timed-out

0

0

Logic Synthesis Data

Without Exceptions

With Exceptions

Worst Negative Slack (WNS)

-0.22 ns

0.0 ns

No. of Setup Violations

64

0

Cell Area (1 nd202as = 2 C.U.A)

245,873

246,772

Total Instance Count

49,966

50,532

Compile Time

11 minutes

25 minutes

Design 4: 120K instances; 4 Clocks

Exception Generation

Description

False Paths

Multicycle Paths

FishTail Focus

Generated

0

527

Focus Run Time

5 minutes

12 hours

Assertion Verification

Verified

-NA-

95

Failed

-NA-

0

Timed-out

-NA-

432

Logic Synthesis Data

Without Exceptions

With Exceptions

Worst Negative Slack (WNS)

-0.91 ns

-0.78 ns

No. of Setup Violations

469

447

Cell Area (1 nd202as = 2 C.U.A)

261,982

262,667

Total Instance Count

1,18,142

1,18,247

Compile Time

27 minutes

31 minutes


 

2. QoR impact after physical synthesis

In this study, Focus was used to generate timing exceptions from RTL for the critical endpoints on a design. The critical endpoints were obtained after doing a quick synthesis of the RTL and seeing which endpoints did not meet timing. The exceptions were verified using Solidify and then read into Design Compiler. The netlist and SDC generated by Design Compiler were then fed into Physical Compiler. QoR data was compared for the same flow with and without Focus exceptions.

Design 5: 200K instances; 4 Clocks

Exception Generation

Description

False Paths

FishTail Focus

Generated

6266

Focus Run Time

3 hours 57 minutes

Physical Synthesis Data

Without Exceptions

With Exceptions

Worst Negative Slack (WNS)

-0.82 ns

-0.03 ns

No. of Setup Violations

682

19

Cell Area (1 nd202as = 2 C.U.A)

4,267,509

4,246,323

Total Instance Count

193,147

192,123

Net Length

11,841,259

10,634,302


 

3. QoR impact on static timing

In this study, the coverage of Focus generated exceptions was measured by taking taped-out designs and removing the timing exceptions that Focus was expected to catch. The number of timing violations and TNS with the timing exceptions removed was reported using PrimeTime. The endpoints that no longer met timing because timing exceptions were removed were written out. Exceptions for these endpoints were generated using Focus and then imported into PrimeTime. Now, the timing violations and TNS with Focus generated exceptions was reported. If the number of timing violations and TNS dropped to 0 with Focus exceptions then it demonstrated that we had caught all the timing exceptions that we were expected to catch. Runtime data for this study is on Solaris.

Design 6: 250K instances; 7 clocks

Exception Generation

Description

False Paths

Multicycle Paths

FishTail Focus

Generated

4395

36

Focus Run Time

2 hours 9 minutes

9 hours 48 minutes

Static Timing Data

Without Exceptions

With Exceptions

Total Negative Slack (TNS)

413 ns

18 ns

No. of Setup Violations

814

105

Percentage Reduction in TNS

-

96%

Percentage Reduction in Violations

-

87%

Design 7: 50K instances; 2 clocks

Exception Generation

Description

False Paths

Multicycle Paths

FishTail Focus

Generated

79

6299

Focus Run Time

15 minutes

51 minutes

Static Timing Data

Without Exceptions

With Exceptions

Total Negative Slack (TNS)

125993 ns

8166 ns

No. of Setup Violations

63126

5966

Percentage Reduction in TNS

-

94%

Percentage Reduction in Violations

-

91%



Copyright 2006 FishTail Design Automation, All rights reserved