What Makes A Design Difficult to Route · 2015-07-29 · Bill Gates. Oct. 1955. Steve Jobs. Feb....

Preview:

Citation preview

IBM Corp.

What Makes A Design Difficult to Route

Charles J. Alpert, Zhuo Li, Michael Moffitt, Gi-Joon Nam, Jarrod Roy, Gustavo Tellez

What Makes a Design Difficult to Route

2 2010 International Symposium on Physical Design March 15, 2010

What Does This Man and a Router Have in Common?

Young math whiz

Discovers computerprogramming

Drops out of Harvard

Starts Microsoft

Works really hard

Becomes very rich

What Makes a Design Difficult to Route

3 2010 International Symposium on Physical Design March 15, 2010

Bill’s OpportunitiesASR-33 Teletype

1968: 8th grade

What Makes a Design Difficult to Route

4 2010 International Symposium on Physical Design March 15, 2010

Some Math Whiz, Computer Programmer, Hard Working Guys

Bill GatesOct. 1955

Steve JobsFeb. 1955

Bill JoyNov. 1954

Scott McNealyNov. 1954

Eric SchmidtApr. 1955

What Makes a Design Difficult to Route

5 2010 International Symposium on Physical Design March 15, 2010

What’s the Difference Between These Guys?

What Makes a Design Difficult to Route

6 2010 International Symposium on Physical Design March 15, 2010

What Does a Router Have in Common with Bill Gates?

Take away the opportunity to succeed, and they won’t

Give them the opportunity to succeed, and they will

What Makes a Design Difficult to Route

7 2010 International Symposium on Physical Design March 15, 2010

Routing is From Venus, Congestion Analysis is From Mars

FloorplanningPlacement

Timing Closure

FloorplanningPlacement

Timing Closure

RoutingCongestion

Analysis

What Makes a Design Difficult to Route

8 2010 International Symposium on Physical Design March 15, 2010

Measuring Routing Quality (40-track per gcell)

What Makes a Design Difficult to Route

9 2010 International Symposium on Physical Design March 15, 2010

Routable gcell (75% full)

What Makes a Design Difficult to Route

10 2010 International Symposium on Physical Design March 15, 2010

Unroutable gcell (150% full) – 20 Nets of Overflow

What Makes a Design Difficult to Route

11 2010 International Symposium on Physical Design March 15, 2010

Example Congestion MapHorizontal Congestion Vertical Congestion

What Makes a Design Difficult to Route

12 2010 International Symposium on Physical Design March 15, 2010

Congestion Map with Better Physical SynthesisHorizontal Congestion Vertical Congestion

What Makes a Design Difficult to Route

13 2010 International Symposium on Physical Design March 15, 2010

Common Routing Metrics

Number of Overflow Nets =Number of Nets Over 100% Congested

Number of nets over 90% congested

Number of nets over 80% congested

Average congestion of 20% worst gcells

What Makes a Design Difficult to Route

14 2010 International Symposium on Physical Design March 15, 2010

Handling Routing Blockages

What Makes a Design Difficult to Route

15 2010 International Symposium on Physical Design March 15, 2010

50% Full or 95% Full?

What Makes a Design Difficult to Route

16 2010 International Symposium on Physical Design March 15, 2010

The Wire Stack

250nm 130nm 90nm 65nm 45nm32nm

What Makes a Design Difficult to Route

17 2010 International Symposium on Physical Design March 15, 2010

Resistance per mm

0

1

2

3

4

5

6

thin medium thick

65 nm45 nm32 nm

What Makes a Design Difficult to Route

18 2010 International Symposium on Physical Design March 15, 2010

Routing Tracks on 4X Metal

What Makes a Design Difficult to Route

19 2010 International Symposium on Physical Design March 15, 2010

Only 2 Tracks Left, but Not Congested

What Makes a Design Difficult to Route

20 2010 International Symposium on Physical Design March 15, 2010

Routing Tracks on 10X Metal

What Makes a Design Difficult to Route

21 2010 International Symposium on Physical Design March 15, 2010

Routing Tracks on 16X Metal

What Makes a Design Difficult to Route

22 2010 International Symposium on Physical Design March 15, 2010

Is This Net Congested?

Different routers can and will report different routing metrics for the same exact global routes!

What Makes a Design Difficult to Route

23 2010 International Symposium on Physical Design March 15, 2010

Cross section of a gcell

1x

2x

4x

What Makes a Design Difficult to Route

24 2010 International Symposium on Physical Design March 15, 2010

Timing Closure Constrains The Router

Layer assignment / wire sizingassigns constraints for the router

What Makes a Design Difficult to Route

25 2010 International Symposium on Physical Design March 15, 2010

Is This a Congestion Problem?

1x

2x

4x

What Makes a Design Difficult to Route

26 2010 International Symposium on Physical Design March 15, 2010

Is It Routable?

Avg 20% = 83.5

#nets > 100% = 532

#nets > 90% = 2785

Avg 20% = 89.1

#nets > 100% = 9532

#nets > 90% = 25785

Avg 20% = 78.2

#nets > 100% = 16

#nets > 90% = 753

Avg 20% = 81.7

#nets > 100% = 3942

#nets > 90% = 11880

What Makes a Design Difficult to Route

27 2010 International Symposium on Physical Design March 15, 2010

Congestion Analysis Versus Routing

Probabilistic Route Fast

Congestion Analysis

No Scenic

Completely Scenic

A Little Scenic

Over Blockages

Around Blockages

May violate layer constraints

Obey layer constraints

Route Carefully

Completely Scenic

Around Blockages

May violate layer constraints

What Makes a Design Difficult to Route

28 2010 International Symposium on Physical Design March 15, 2010

Minimizing Wire Generally Improves CongestionBad placement Good placement

What Makes a Design Difficult to Route

29 2010 International Symposium on Physical Design March 15, 2010

Spread it Uniformly Doesn’t Always Work

What Makes a Design Difficult to Route

30 2010 International Symposium on Physical Design March 15, 2010

CRISP Progression

Initial Placement

Iteration 5 Iteration 10 Iteration 15

Horizontal Congestion

Vertical Congestion

What Makes a Design Difficult to Route

31 2010 International Symposium on Physical Design March 15, 2010

Effect of Timing Driven-PlacementInitial placement Timing-driven placement

What Makes a Design Difficult to Route

32 2010 International Symposium on Physical Design March 15, 2010

Effect of Timing Driven-Placement

Initial placement Timing-driven placement

What Makes a Design Difficult to Route

33 2010 International Symposium on Physical Design March 15, 2010

Impact of More Efficient Area Usage

Avg 20%: 108.06 >90% nets: 37855 >100% nets: 21364

Avg 20%: 86.89>90% nets: 7307>100% nets: 1147

What Makes a Design Difficult to Route

34 2010 International Symposium on Physical Design March 15, 2010

“Sitting” in the Student Section

Density

Rou

ting

diffi

culty

What Makes a Design Difficult to Route

35 2010 International Symposium on Physical Design March 15, 2010

Buffering Along A Global Route

What Makes a Design Difficult to Route

36 2010 International Symposium on Physical Design March 15, 2010

Buffer Packing

What Makes a Design Difficult to Route

37 2010 International Symposium on Physical Design March 15, 2010

Blockage Avoidance Routing

What Makes a Design Difficult to Route

38 2010 International Symposium on Physical Design March 15, 2010

The Corona Effect

What Makes a Design Difficult to Route

39 2010 International Symposium on Physical Design March 15, 2010

Corona Effect + Too Much Thick Metal

What Makes a Design Difficult to Route

40 2010 International Symposium on Physical Design March 15, 2010

Hot Spots versus Opens and Shorts

What Makes a Design Difficult to Route

41 2010 International Symposium on Physical Design March 15, 2010

What Makes a Design Difficult to Route?

Bad floorplan

Over packing dense logic

Inefficient area minimization

Over weighting for timing-driven placement

Buffering too packed

Over-constraining router

Not capturing local routing isses

Even if successful, don’t mess up timing too much

What Makes a Design Difficult to Route

42 2010 International Symposium on Physical Design March 15, 2010

What Does a Child Have in Common with a Router ?

Just because opportunity is provided, they may not take advantage.

We want to create opportunities to succeed.

We don’t really know what the best opportunities are.

Recommended