1 of 24

2022 ICCAD CAD Contest Problem B: 3D Placement with D2D Vertical Connections

Problem Description, Benchmarks, and Results

Kai-Shun Hu, I-Jye Lin, Yu-Hui Huang, Hao-Yu Chi, Yi-Hsuan Wu, and Chin-Fang Cindy Shen

November 2022

© 2022 Synopsys, Inc.

1

2 of 24

Outline

  • Problem Description
  • Evaluation Methodology
  • Benchmarks
  • Results

Synopsys Confidential Information

© 2022 Synopsys, Inc.

2

3 of 24

Outline

  • Problem Description
  • Evaluation Methodology
  • Benchmarks
  • Results

Synopsys Confidential Information

© 2022 Synopsys, Inc.

3

4 of 24

Problem Description

  • In the chiplet era, the benefits of splitting a large die into multiple small dies
    • Better yield
    • Better system performance
    • Better cost

​

  • Develop a partitioning & placer engine
    • Partition the netlist into 2 dies
    • Place the partitioned netlist in each die
    • Create inter-die connection terminals
    • Honor all given constraints
    • Optimize the give score metric

Synopsys Confidential Information

© 2022 Synopsys, Inc.

4

5 of 24

Problem Description

  • In this contest, we focus on 2-die face-to-face vertically stacked configuration

​

  • Given
    • Netlist
    • Die size of top & bottom dies
      • Would be the same size
    • Max placement u-rate for top & bottom dies
    • Placement rows of top & bottom dies
    • The std cell library for each die
      • Can be 2 different technologies
    • Inter-die terminal size & the spacing constraint
  • Output
    • Top die placement result
    • Bottom die placement result
    • Inter-die terminal creation result with net info

​

  • Challenges
    • Partitioning and placement fusion
    • #inter-die terminals is limited
    • Each die can have its own technology

​

  • Multi-threading is encouraged

Synopsys Confidential Information

© 2022 Synopsys, Inc.

5

6 of 24

HPWL Calculation with Inter-die Terminal

  • 2-pin net examples

​

​

​

​

​

​

  • 3-pin net example

Bottom die

Top die

Bottom die

Top die

One large die

Good

Not

Good

Bottom die

Top die

Good

One large die

Cell instance and pins

Inter-die terminal

HPWL calculation bbox

Synopsys Confidential Information

© 2022 Synopsys, Inc.

6

7 of 24

Outline

  • Problem Description
  • Evaluation Methodology
  • Benchmarks
  • Results

Synopsys Confidential Information

© 2022 Synopsys, Inc.

7

8 of 24

Evaluation Methodology

  • Evaluation score metric
    • All the given constraints need to be satisfied
    • Runtime limit is 1hr for each case on the evaluation machine

​

​

​

​

​

  • Evaluator is provided
    • Usage:

./evaluator <input.txt> <output.txt>

 

Synopsys Confidential Information

© 2022 Synopsys, Inc.

8

9 of 24

Outline

  • Problem Description
  • Evaluation Methodology
  • Benchmarks
  • Results

Synopsys Confidential Information

© 2022 Synopsys, Inc.

9

10 of 24

Benchmarks

  • Public cases

​

​

​

​

​

  • Hidden cases

​

case1

case2

case3

case4

Die size

30 x 30

10175 x 8151

19240 x 19192

53294 x 53255

#nets

6

2644

44360

220071

#cellInsts

8

2735

44764

220845

max #inter-die terminals

4

2000

36481

183612

max u-rate of top die

80

70

78

66

max u-rate of bottom die

90

75

78

70

diff tech?

Yes

Yes

No

Yes

​

case2_hidden

case3_hidden

case4_hidden

Die size

11670 x 9349

17599 x 17555

55988 x 55947

#nets

2644

44360

220071

#cellInsts

2735

44764

220845

max #inter-die terminals

2000

36100

178929

max u-rate of top die

79

68

66

max u-rate of bottom die

79

78

76

diff tech?

No

Yes

Yes

Synopsys Confidential Information

© 2022 Synopsys, Inc.

10

11 of 24

Outline

  • Problem Description
  • Evaluation Methodology
  • Benchmarks
  • Results

Synopsys Confidential Information

© 2022 Synopsys, Inc.

11

12 of 24

Results – Top X

  • Alpha

​

​

​

​

​

  • Beta

Alpha to Beta

Rank

case1

case2

case2_hidden

case3

case3_hidden

case4

case4_hidden

1

122

2,605,824

3,188,338

38,604,995

33,677,415

363,280,018

377,724,079

2

125

3,219,794

4,384,874

54,660,200

52,324,900

434,594,063

798,279,815

3

125

4,238,334

5,792,885

69,785,361

89,461,363

780,627,443

1,925,577,517

4

132

4,777,852

6,436,693

102,660,962

126,758,615

988,969,163

2,696,898,820

5

132

5,825,538

8,032,807

110,591,580

234,349,400

1,143,228,315

2,696,898,820

6

137

5,967,618

8,305,699

205,284,108

234,349,400

2,454,429,019

3,790,600,666

Rank

case1

case2

case2_hidden

case3

case3_hidden

case4

case4_hidden

1

100

2,189,151

2,958,953

36,191,154

31,911,473

328,483,651

361,787,452

2

115

2,299,926

3,077,797

36,392,614

32,021,465

342,887,221

369,635,153

3

121

2,307,539

3,131,276

36,900,132

32,362,919

351,522,988

378,026,892

4

122

2,502,830

3,148,305

37,083,115

33,165,721

352,500,029

385,194,065

5

123

2,615,658

3,179,571

39,074,388

34,209,591

356,928,663

396,335,554

6

125

2,685,059

3,524,680

41,191,210

34,963,557

381,683,342

455,519,945

Synopsys Confidential Information

© 2022 Synopsys, Inc.

12

13 of 24

Results – Top X

Beta to Final

  • Beta

​

​

​

​

​

  • Final

Rank

case1

case2

case2_hidden

case3

case3_hidden

case4

case4_hidden

1

100

2,189,151

2,958,953

36,191,154

31,911,473

328,483,651

361,787,452

2

115

2,299,926

3,077,797

36,392,614

32,021,465

342,887,221

369,635,153

3

121

2,307,539

3,131,276

36,900,132

32,362,919

351,522,988

378,026,892

4

122

2,502,830

3,148,305

37,083,115

33,165,721

352,500,029

385,194,065

5

123

2,615,658

3,179,571

39,074,388

34,209,591

356,928,663

396,335,554

6

125

2,685,059

3,524,680

41,191,210

34,963,557

381,683,342

455,519,945

Rank

case1

case2

case2_hidden

case3

case3_hidden

case4

case4_hidden

1

97

1,960,913

2,555,461

30,247,740

27,650,329

274,026,678

301,193,374

2

100

2,014,563

2,613,806

30,580,336

27,756,492

281,315,669

307,399,565

3

106

2,062,029

2,644,791

30,969,011

28,372,567

281,378,079

308,359,159

4

112

2,072,075

2,730,588

32,542,760

28,472,020

290,929,204

316,790,425

5

113

2,080,647

2,735,158

32,702,572

28,929,430

298,707,254

325,459,781

6

114

2,097,487

2,743,092

32,786,576

30,010,475

302,078,105

327,983,245

Synopsys Confidential Information

© 2022 Synopsys, Inc.

13

14 of 24

Extra Experiments of Different Terminal Pitch

  • We adjust the terminal pitch to test how the top teams can perform
    • Smaller pitch (0.5x) indicates more terminals can be placed in the same area
    • Larger pitch (1.5x) indicated less terminals can be placed

​

  • Result of ‘case3_hidden’

​

Synopsys Confidential Information

© 2022 Synopsys, Inc.

14

15 of 24

Extra Experiments of Different Terminal Pitch (Cont.)

  • We adjust the terminal pitch to test how the top teams can perform
    • Smaller pitch (0.5x) indicates more terminals can be placed in the same area
    • Larger pitch (1.5x) indicated less terminals can be placed

​

  • Result of ‘case4’

Synopsys Confidential Information

© 2022 Synopsys, Inc.

15

16 of 24

Extra Experiments of Different Utilization Rate

  • We adjust the U-rate for both dies
    • Higher U-rate provides more flexibility of placing the cells on either die
    • Interestingly, this experiments showed all 3 top teams did NOT perform better with higher U-rate

​

  • Results of ‘case3_hidden’ and ‘case4’

Synopsys Confidential Information

© 2022 Synopsys, Inc.

16

17 of 24

Detail of Extra Experiments

​

Teams

case3_hidden

case4

0.5x

1x

1.5x

0.5x

1x

1.5x

Orig.

Top 1

27,134,568

27,650,329

29,296,998

274,169,548

281,315,669

314,557,706

Top 2

27,723,498

27,756,492

28,510,944

277,205,164

274,026,678

316,409,897

Top 3

28,225,663

28,372,567

32,420,159

280,700,872

281,378,079

309,513,989

Mid

Top 1

27,741,967

28,378,047

29,386,056

287,826,200

293,815,102

316,249,057

Top 2

27,565,968

27,711,754

28,008,978

296,898,320

297,501,818

300,798,531

Top 3

27,785,575

27,920,944

28,513,311

305,153,056

311,949,757

322,763,231

100

Top 1

28,852,463

28,846,873

30,287,405

305,507,426

310,114,603

326,824,728

Top 2

30,346,130

30,182,531

30,125,795

329,000,686

326,529,846

325,724,725

Top 3

267,484,464

265,325,967

264,512,092

336,407,277

336,485,337

335,502,622

U-Rate

T-Pitch

Synopsys Confidential Information

© 2022 Synopsys, Inc.

17

18 of 24

Winners

Synopsys Confidential Information

© 2022 Synopsys, Inc.

18

19 of 24

Third Place

Team: cadb1021

​

The Chinese University of Hong Kong

Peking University

​

Peiyu Liao

Yuxuan Zhao

Dawei Guo

Shuo Yin

Prof. Bei Yu

Prof. Yibo Lin

Synopsys Confidential Information

© 2022 Synopsys, Inc.

19

20 of 24

Second Place

Team: cadb1015

​

Southeast University

​

Yangjie Mei

Fuheng Shen

Yuejian Shi

Hong Liu

Prof. Ziran Zhu

Prof. Jun Yang

Synopsys Confidential Information

© 2022 Synopsys, Inc.

20

21 of 24

First Place

Team: cadb1051

​

Chinese Academy of Sciences

Minnan Normal University

Guangdong University of Technology

​

Shijian Chen

Xueyan Zhao

Jiangkao Li

Yihang Qiu

Prof. Xingquan Li, Prof. Zhipeng Huang

Synopsys Confidential Information

© 2022 Synopsys, Inc.

21

22 of 24

Honorable Mention

Team ID: cadb1024

Fudan University

Peng Zou, Min Wei, Yuan Wen, Zhijie Cai, Xingyu Tong

Advisor: Prof. Jianli Chen, Prof. Jun Yu

​

​

Team ID: cadb1039

The Chinese University of Hong Kong

Bangqi Fu, Lixin Liu, Xinshi Zang, Fangzhou Wang

Advisor: Prof. Evangeline F.Y. Young, Prof. Martin Ding Fat Wong

​

​

​

Synopsys Confidential Information

© 2022 Synopsys, Inc.

22

23 of 24

Thank You

© 2022 Synopsys, Inc.

23

24 of 24

2022 ICCAD CAD Contest Problem B: Placement with D2D Vertical Connections

Problem Description, Benchmarks, and Results

Kai-Shun Hu, I-Jye Lin, Yu-Hui Huang, Hao-Yu Chi, Yi-Hsuan Wu, and Chin-Fang Cindy Shen

November 2022

© 2022 Synopsys, Inc.

24