Multi-Objective Simulation-based Optimization of Analog Transistor Sizing
Abstract
In this paper, the simulated annealing (SA) and genetic algorithm (GA) are explored for automating the design of an analog circuit. The transistor sizing methodology performs optimization with simulation-based feedback. The sizes of the transistors of an analog circuit are determined by the optimization algorithm and are applied to a schematic representation of the circuit for characterization through SPICE simulation. The characterized response of the circuit is then provided as input to the optimization algorithm for further tuning. The proposed transistor sizing methodology is applied to an active inductor based voltage controlled oscillator (VCO) for multiple target frequencies and amplitudes. The SA algorithm produced no more than a 6 percent error in frequency from target values, while the GA produced transistor sizes that resulted in no more than 6 percent error in frequency and 13 percent error in amplitude. In addition, the GA provided superior results in determining transistor sizes for higher target frequencies, as determined through calculation of the cost function. However, the improved results came at a considerably greater computational cost. A hybrid optimization technique is proposed, where GA performs coarse search space exploration to determine the initial conditions for the SA, and SA is then executed for fine optimization of the transistor sizes. The hybrid optimization technique not only resulted in 2.6x faster computational time as compared to GA but also resulted in a significantly lower cost function as compared to both the GA and SA optimization techniques. The results indicate that the proposed methodology provides a quick and accurate means of determining transistor sizes for target analog circuit specifications, which significantly reduces the cost and the design time of analog circuits.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 17, 2020
- Accession Number
- AD1102760
Entities
People
- Ioannis Savidis
- Vaibhav V. Rao
Organizations
- Drexel University