Optimal RF design using Smart Evolutionary Algorithms

1
Optimal RF design using Smart Evolutionary
Algorithms

• Peter Vancorenland, Carl De Ranter, Michiel
  Steyaert, Georges Gielen
• MICAS - KULeuven
• Design Automation Conference
• Los Angeles
• 6-6-00

2
Overview

• Current Optimization Algorithms
• Smart Evolutionary Algorithm
• Combined evolutionary Algorithm
• Cost Function fitting
• The LNA example
• Specfications
• Models
• experiments

3
Current Optimization Algorithms

• Full-featured simulation based optimization
• Accurate
• Slow!!
• Approximating simulation based optimization
• Fast
• Limited to the approximation region
• Genetic algorithm with cost function fitting
• Fast
• Adaptive approximation region

4
Smart Evolutionary Algorithm
Fit Costfunction
Optimize Cost Function over X steps
Optimize fitted Costfunction over y steps
Restart the whole process
5
Total optimization
6
Cost function fitting

• Fit function
• Polynomial
• Easy
• No guaranteed good fit
• Hand Calculation Model
• More extensive
• Mostly better fit
• Fit Algorithm
• Stochastic
• Slow
• Non-Stochastic like Steepest Gradient
• Fast
• Greedy

7
Example

• Optimisation of an LNA

8
LNA parameters

• IIP3 gt IIP3spec
• S11 lt S11spec
• Gain gt Gainspec
• Noise Figure lt Nfspec
• Frequency Fspec
• Min. Power

9
Cost function specifications
10
Costfunction modeling

• Parasitic coil parameters as function of
  geometry
• Level 3 transistor model
• Mathematical input matching model

11
Optimization Process
12
Results Gain Power Trade-off
13
Conclusions

• Fast Optimization strategy
• Allows to put circuit knowledge into the
  optimization
• First step of Global Optimization on system level
• Example on RF LNA Optimization