Abstract
Nowadays, investigating new techniques to accelerate nonlinear systems’ simulations is a need because of their regular usage in industry. Systems’ simulators need to solve an enormous number of nonlinear equations. Software-based solutions cannot solve complex nonlinear equations in a reasonable time while maintaining scalability with the increase in the number of equations. The speedup of the process can be achieved by hardware accelerators. This research enhances a neural network architecture with a new hybrid-updating rule realized in hardware to solve nonlinear equations. A scalable hardware architecture is introduced to solve any number of nonlinear equations and achieve a high performance gain. Results show the increase in the performance gain between our proposed solution and software solvers. For example, our proposed architecture is able to solve 1000 sparse equations on Xilinx Virtex-7 with 400x speedup compared to other software methods.