Modeling of single-ended interconnects using surrogate model and design of experiments techniques

Abstract

A methodology is defined to select the variables with the greatest impact on the response of the system to be used in the design of single-ended low speed interconnects. A point-to-point topology was studied in this work, consisting of a master and a slave buffer, package, and two transmission lines with a series resistor on the PCB. The considered response for the system was the flight time of digital signals coming out of the buffers. To lessen computational resources during optimization processes, surrogate models were obtained using design of experiments. To choose the active factors for the response, screening experiments were done. Then, to build the surrogate model face-centered and rotatable central composite designs for both the master and slave signals were compared in terms of model fit, with a 15% maximum error. The surrogate models were used to represent a Serial Peripheral Interface, which is a synchronous bus, and requires a timing analysis to assess the maximum allowed length of the routing. A Nelder-Mead algorithm with penalty functions was used to optimize the transmission lines lengths, while maintaining the timing margins equal to or larger than zero. Optimal lengths were validated using SPICE. The methodology proposed in this work can help in reducing design times for single-ended low-speed singles, as there would not be need to simulate the whole topology each time the value of a component needs to be changed.