Hardware Support for Simulated Annealing and Tabu Search

In this paper, we present a concept of a CPU kernel with hardware support for local-search based optimization algorithms like Simulated Annealing (SA) and Tabu-Search (TS). The special hardware modules are: (i) A linked-list memory representing the problem space. (ii) CPU instruction set extensions supporting fast moves within the neighborhood of a solution. (iii) Support for the generation of moves for both algorithms, SA and TS. (iv) A solution mover managing several solution memories according to the optimization progress. (v) Hardware addressing support for the calculation of cost functions. (vi) Support for nonlinear functions in the acceptance procedure of SA. (vii) A status module providing on-line information about the solution quality. (v) An acceptance prediction module supporting parallel SA algorithms. Simulations of a VHDL implementation show a speedup of up to 260 in comparison to an existing implementation without hardware support.