CAD Tool Design Space Exploration via Bayesian Optimization Yuzhe Ma - - PowerPoint PPT Presentation

CAD Tool Design Space Exploration via Bayesian Optimization

Yuzhe Ma1, Ziyang Yu2, Bei Yu1

1Chinese University of Hong Kong 2University of Hong Kong

1 / 18

Design Flow

System Specification Architectural Design Functional Design and Logic Design (RTL) Logic Synthesis Physical Design Physical Verification and Signoff Fabrication Packaging and Testing Chip

module test input in[3]; … endmodule

AND OR

DRC LVS STA

◮ It’s really a long journey; ◮ Each step is more complicated as

the technology node advances;

◮ Huge effort is needed to achieve the

desired design quality.

2 / 18

Case Study: Adder Design

Binary Adder ◮ Primary building blocks in the datapath logic of a microprocessor. ◮ A fundamental problem in VLSI industry for last several decades.

Anything else we can do?

3 / 18

Gaps Between Design Stages

1 2 3 4 5 6 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 54 57 60 63

Level

◮ Logic synthesis v.s. physical synthesis ◮ Constraints mapping between two synthesis stages is difficult.

4 / 18

Design Space – Front-End

Parallel Prefix Adders → Flexible delay-power trade-off Regular Adders → Sub-optimal Custom Adders → High TAT

5 / 18

Design Space – Back-End

◮ Tool settings. Huge space for different options.

Source: ICC documentation 6 / 18

Design Space Exploration

6000 6500 7000 7500 8000 340 360 380 400 420 440 Power(µw) Critical Delay(ps)

Real PF

• Rep. Adder

◮ Search for the Pareto-optimal

designs;

◮ None of the objective metrics, such

as area, power or delay, can be improved without worsening at least

• ne of the others.

7 / 18

Bayesian Optimization

◮ Good candidate to optimize functions that take a long time to evaluate. ◮ Can tolerate stochastic noise in function evaluations.

Black-box model Surrogate model Acquisition fucntion Optimization model Simulated data Proposed data

◮ Acquisition function serves as a utility measurement to select the next point for

evaluation;

◮ Surrogate model is adaptively refined to approximate the latent function.

8 / 18

Gaussian Process Regression

◮ Gaussian process regression is a Bayesian statistical approach for modeling unknown

functions.

◮ Prior: f ∼ GP(0, k(·, ·)) ◮ Posterior:

• m(x) = k(x, X)⊤(k(X, X) + σ2I)−1Y,

σ2(x) = k(x, x) − k(x, X)⊤(k(X, X) + σ2I)−1k(x, X),

2 4 6 8 10 x 10 5 5 10 15 20 f(x)

Prediction 95% confidence interval

9 / 18

Acquisition Function

Lower Confidence Bound (LCB)

◮ LCB(x) = m(x) − βσ(x); ◮ m(x) indicates the “exploitation” and σ(x) indicates the “exploration”; ◮ β is a parameter that balances the

exploitation and exploration. Expected Improvement (EI)

◮ EI(x) = σ(x)(λΦ(λ) + φ(λ)), where λ = τ−ξ−µ(x)

σ(x)

◮ The expected improvement function

favors the optimal region with high probability and the promising area with high uncertainty estimation.

10 / 18

Overall Flow

Acquisition Function EDA tools GP Regression

S1: Sample S2: Evaluate S3: Model update

Area, Power, Delay Configuration Surrogate Model

11 / 18

Experimental Configurations

◮ Design:

• DesignWare library,
• Regular: Sklansky, Kogge-Stone,
• Synthesized prefix adder [Ma+, TCAD’2019].

◮ Flow:

• Cell library: 32nm SAED;
• Tools: DC 2014 & ICC 2017.

◮ Design space:

• Parameters in timing constraints, placement utilization, power options, etc.

12 / 18

BO vs. Industrial Setting

◮ DSE with single objective ◮ Baseline: a set of complete scripts for adder synthesis from industrial.

DesignWare Sklansky Kogge-Stone Synthesized

1,000 2,000

Area (µm2) Baseline BO

DesignWare Sklansky Kogge-Stone Synthesized

300 320 340 360

Delay (ps) Baseline BO

13 / 18

BO vs. Industrial Setting

◮ DSE with multiple objectives using scalarization

Area Energy Delay

0.95 1

(DesignWare)

Ratio Baseline BO

Area Energy Delay

0.85 0.9 0.95 1

(Sklansky)

Ratio

Area Energy Delay

0.95 1

(Kogge-Stone)

Area Energy Delay

0.85 0.9 0.95 1

(Synthesized) 14 / 18

BO vs. Heuristic Search

◮ Evolutionary algorithms are widely applied in black-box function optimization; ◮ Genetic algorithm (GA) is not as stable as Bayesian optimization. 10 20 30 1,780 1,800 1,820 1,840 1,860

Iteration Area (µm2) BO GA

10 20 30 360 380 400 420

Iteration Delay (ps) BO GA

15 / 18

Scaling Trick

Area Energy Delay

0.96 0.98 1 1.02 1.04

(DesignWare)

Ratio Scale-up Scale-down

Area Energy Delay

0.94 0.96 0.98 1

(Sklansky)

Ratio

Area Energy Delay

0.94 0.96 0.98 1

(Kogge-Stone)

Area Energy Delay

0.94 0.96 0.98 1

(Synthesized) 16 / 18

Discussion & Conclusion

Conclusion ◮ A machine learning approach for better design; ◮ Adapt BO for multi-objective optimization to simultaneously minimize PPA values; ◮ BO substantially outperforms typical evolutionary algorithms.

17 / 18

Discussion & Conclusion

Conclusion ◮ A machine learning approach for better design; ◮ Adapt BO for multi-objective optimization to simultaneously minimize PPA values; ◮ BO substantially outperforms typical evolutionary algorithms. Further Improvement ◮ A unified design space for exploration.

• Currently the design spaces of front-end and back-end are separated.

◮ A more elegant way to handle multi-objective optimization.

• Scalarization requires tuning effort and data processing tricks.

17 / 18

Thank You

18 / 18