Evolving the Ultimate Processor

Bojan Antonovic. Evolving the Ultimate Processor
Diploma Thesis DA-2003.31 Summer Term 2003 Supervisor: Prof. Dr. Lothar Thiele
Institut fuŸr Technische Informatik und Kommunikationsnetze Computer Engineering and Networks LaboratoryThis diploma thesis is about optimizing processors by the criteria performance, area and power consumption. Because this is a multi criterion optimization problem, the created results do not represent a single optimal or best fraction of optimal solutions, they represent a bunch of non-dominated elements, called the Pareto front. To get this Pareto front, a multi criterion optimization algorithm, based on the principle of Evolutionary Algorithms (EA), SPEA2, is used. Tools are used to compute the objectives where possible. The estimation of the area
consumption is showed how it's made with tools and complementary information like pictures of processor cores. EA aspects like creation of random elements and variation operators like mutation and cross-over are discussed. A model is shown which support this all. The protocol PISA is used to connect the optimization algorithm with the optimization problem, together with its variators. A PISA implementation in Java, JavaPISA, is presented. At the end optimization results are presented and discussed.1 Introduction
2 Processor optimization
3 Evolutionary algorithm components
4 Realization and Implementation
5 Optimization results
6 Conclusion
A Processor con guration
B Installation