EMI sources from mode conversion in a Telco system high-speed SERDES

Prediction of signal integrity (SI) and electromagnetic interference (EMI) from multilayer circuit boards is an important problem which is of interest to high speed digital system designers. SI and EMI control should be treated as a design problem from the start and it should be given the necessary resources throughout the design process. This can only be accomplished with accurate and effective electrical modeling of the complex interconnection structures. In this paper, electrical modeling methods are described where both differential-mode and common-mode effects and their conversions are accurately captured. Common-mode imbalances degrade the signal integrity margins making the system more susceptible to failure with increased EMI radiation. Simulation models are developed to properly quantify mode conversion mechanisms. A high speed blade system is chosen as a test bench to demonstrate the characterization methodology. The blade chassis is a compact, modular, rack mount scalable server system with support for twelve single-wide server blades. All high-speed communication occurs through a backplane using four sets of redundant fabrics, yielding a total of eight network connections per blade. This architecture allows for communication between any blade server and the chassis management units, the I/O switch modules, the highspeed switch modules, and the I/O bridge modules [1]. In the high speed differential serial channel design process it is important to model differential to common mode conversion. In order to minimize the differential insertion loss, increase the signal to noise ratio at the receiver to improve bit-error rate (BER) and minimize EMI, careful analysis of sources of loss, imbalance, and skew within the channel is needed. In this work, we demonstrate the importance of common-mode noise to radiation from a PCB. We then focus on mode conversion due to proximity of transmission lines to plane edges, impedance mismatch, current return path variation, t- - race loading to due coupling, and AC coupling components. Measured results were shown that highlight the significant variability in mode conversion that can result from the various mechanisms simulated in this study.