Towards a Dynamic Evolutionary Approach to FPGA Temperature Management

Peter R. Lewis, Walter C. Chibamu and Xin Yao
In Proceedings of the 2nd International Workshop on Self-awareness in Reconfigurable Computing Systems (SRCS), pp 22-27. 2013.

Partially reconfigurable computing systems, such as FPGAs, provide great promise for on-the-fly adaptive computation. Such systems can be modified while in use, to deal with newly arriving computational tasks. However, how to adapt, and with respect to what objectives, is not yet well established. Typically, one might be interested to achieve an allocation of tasks, as they arrive, to differently sized regions of the FPGA, in order to provide the most efficient computation. At the same time, it is important to also maintain healthy on-chip temperature. In this paper, we propose a novel description of this dynamic FPGA temperature management problem. We formulate the problem as a dynamic multi-objective optimisation problem, with three objectives and a time-linkage characteristic. We propose the use of evolutionary algorithms as a method to tackle the problem.