An architecture for solving quadratic programs with the fast gradient method on a Field Programmable Gate Array

Abstract

In this paper an architecture for the implementation of gradient-based optimisation methods on a Field Programmable Gate Array (FPGA) is proposed. Combining the algorithmic advantages of gradient-based algorithms with the computational strengths of a tailored FPGA implementation allows to solve quadratic programs occurring, for example, in Model Predictive Control (MPC) applications in the microsecond range. The experimental comparisons show a computational advantage of the proposed FPGA implementation against parallel software versions ranging between one and two orders of magnitude. The proposed FPGA-based solution can broaden the applicability of MPC to problems that were considered out-of-reach till recent years.