Adaptive FPGA placement by natural optimisation

Placement is an NP-complete problem, thus probabilistic algorithms are usually applied to its resolution. Simulated annealing (SA) is one of the most successful methods available to treat placement problems. In spite of the wide field of application of SA, experimentation with new cost functions requires costly studies for fine-tuning the algorithm parameters. In this paper, we present natural optimisation (NO), a new self-tuning combinatorial optimisation method. We suggest a slight variation on the SA method to convert it from a user-adaptive method into a self-adaptive one. The initial temperature is automatically determined, and its cooling is adjusted in a natural way to its fair value during the whole annealing process. We have compared the NO algorithm with a fine-tuned SA-based placement tool. As a result, NO appears to be a very promising combinatorial optimisation method, since it simplifies the hard task of parameter adjustment while maintaining the high-quality results provided by SA.