Towards embedded model predictive control for System-on-a-Chip applications

We propose a framework for embedding model predictive control for Systems-on-a-Chip applications. In order to allow the implementation of such a computationally expensive controller on chip, we propose reducing the precision of the microprocessor to the minimum while maintaining near optimal control performance. Taking advantage of the low precision, a logarithmic number system based microprocessor architecture is used, that allows the design of a reduced size processor, providing further energy and computational cost savings. The design parameters for this high-performance embedded controller are chosen using a combination of finite element method simulations and bit-accurate hardware emulations in a number of parametric tests. We provide the methodology for choosing the design parameters for two particular control problems; the temperature regulation in a wafer cross-section geometry, and the control of temperature in a non-isothermal fluid flow problem in a microdevice. Finally, we provide the microprocessor architecture details and estimates for the performance of the resulting embedded model predictive controller.