Pareto Optimization of Analog Circuits Using Reinforcement Learning
Abstract
1 Introduction
2 Related Work
2.1 Evolutionary Algorithms
2.2 Bayesian Optimization
2.3 Reinforcement Learning
3 Our Contributions
4 Multi-Objective Optimization Via Reinforcement Learning
4.1 Problem Formulation
4.2 Goal Vector and Multi-Goal Reinforcement Learning Setup
4.3 The Bellman Operator for Multi Objective Reinforcement Learning
5 Multi-Objective Reinforcement Learning for Continuous Sizing Solutions in Analog Circuits
5.1 Actor Policy
5.2 Preference Policy
5.3 Training Phase
5.4 Inference Phase
5.5 Mini-Batch Optimization
6 Experimental Results
6.1 Metrics Used
Two-stage Differential Amplifier
Method | \(\mathcal {HV}\) | No. of samples | Simulation Time |
---|---|---|---|
Monte Carlo | 0.83 | 15,000 | \(\sim\) 4 hr |
NSGA-II | 0.67 | 2,500 | \(\sim\) 1 hr |
BO | 0.64 | 15,000 | \(\sim\) 10 hr |
RL-train | 0.62 | 2,500 | \(\sim\) 1 hr |
\(\mathcal {FC}\) | RL-train | NSGA-II | BO | Monte Carlo |
---|---|---|---|---|
RL-train | — | 0.6 | 0.76 | 1.0 |
NSGA-II | 0.4 | — | 0.81 | 0.98 |
BO | 0.24 | 0.19 | — | 1.0 |
Monte Carlo | 0.0 | 0.02 | 0.0 | — |
Folded Cascode Amplifier
Method | \(\mathcal {HV}\) | No. of samples | Simulation Time |
---|---|---|---|
Monte Carlo | 0.30 | 15,000 | \(\sim\) 4 hr |
NSGA-II | 0.25 | 2,500 | \(\sim\) 1 hr |
BO | 0.32 | 15,000 | \(\sim\) 10 hr |
RL-train | 0.26 | 2,500 | \(\sim\) 1 hr |
\(\mathcal {FC}\) | RL-train | NSGA-II | BO | Monte Carlo |
---|---|---|---|---|
RL-train | — | 0.55 | 0.79 | 0.97 |
NSGA-II | 0.45 | — | 0.71 | 0.94 |
BO | 0.21 | 0.29 | — | 0.7 |
Monte Carlo | 0.03 | 0.06 | 0.3 | — |
Hysteresis Comparator
6.2 Inference Phase
Method | \(\mathcal {HV}\) | No. of samples | Simulation Time |
---|---|---|---|
Monte Carlo | 1.91 | 15,000 | \(\sim\) 4 hr |
NSGA-II | 1.76 | 5,000 | \(\sim\) 2 hr |
BO | 1.66 | 15,000 | \(\sim\) 10 hr |
RL-train | 1.60 | 5,000 | \(\sim\) 2 hr |
\(\mathcal {FC}\) | RL-train | NSGA-II | BO | Monte Carlo |
---|---|---|---|---|
RL-train | — | 0.61 | 0.9 | 1.0 |
NSGA-II | 0.39 | — | 0.83 | 1.0 |
BO | 0.1 | 0.17 | — | 0.91 |
Monte Carlo | 0.0 | 0.0 | 0.09 | — |
Circuit | \(\eta\) |
---|---|
Two-stage Diff Amp. | 100% |
Folded Cascode Amp. | 98% |
Hysteresis Comp. | 91% |
7 Conclusion and Discussions
References
Index Terms
- Pareto Optimization of Analog Circuits Using Reinforcement Learning
Recommendations
Pareto-Based Multiobjective Machine Learning: An Overview and Case Studies
Machine learning is inherently a multiobjective task. Traditionally, however, either only one of the objectives is adopted as the cost function or multiple objectives are aggregated to a scalar cost function. This can be mainly attributed to the fact ...
Pareto optimization of analog circuits considering variability
ECCTD 2007The trade-off between competing design objectives is often decided at the beginning of the analog sizing process by assigning weights to the design objectives. Architectural design, however, requires the knowledge of all possible optimal trade-offs of a ...
Multi-objective Pareto optimization of axial compressors using genetic algorithms
ICCOMP'06: Proceedings of the 10th WSEAS international conference on ComputersMulti-objective genetic algorithm (GAs) (non-dominated sorting genetic algorithm, NSGA-II) with a new diversity preserving mechanism is used for Pareto optimization of axial compressor. The conflicting design objectives of axial compressor are, total ...
Comments
Information & Contributors
Information
Published In
Publisher
Association for Computing Machinery
New York, NY, United States
Journal Family
Publication History
Check for updates
Author Tags
Qualifiers
- Research-article
Funding Sources
- Semiconductor Research Corporation Task No. 2810.043 through UT Dallas’ Texas Analog Center of Excellence, the U.S. Department of Energy’s “Data-Driven Decision Control for Complex Systems (DnC2S)”
- National Science Foundation (NSF)
Contributors
Other Metrics
Bibliometrics & Citations
Bibliometrics
Article Metrics
- 0Total Citations
- 718Total Downloads
- Downloads (Last 12 months)718
- Downloads (Last 6 weeks)100
Other Metrics
Citations
View Options
Get Access
Login options
Check if you have access through your login credentials or your institution to get full access on this article.
Sign in