Analysis and Optimization of Channel Equalization for High Speed Serial Links

Abstract

Serial links are becoming more vital in the past decade especially with the emerg¬ing needs of high-performance connectivity with the exponential increase of required data rates in applications like Automotive, Data centers and cloud computing. The growing demand for such high data rates in chip-to-chip links combined with bandwidth limited channels imposes severe equalization requirements on the serial link. Channel equalization could be realized in both transmitter and receiver in the same time. While the Feed Forward Equalizer (FFE) is the major contributor to equalization requirements in the transmitter, the Continuous Time Linear Equalizer (CTLE) and Decision Feed¬back Equalizer (DFE) are responsible for realizing the equalization requirements in the receiver side. This work discusses the imperfections of channels used in wire link transceivers ranging from Extra Short Reach (XSR), found in die-to-die wire links, to Long Reach (LR) ones found in backplanes. Then it shows the impact of such impurities on the quality of the high-speed transmitted data as a degradation in the quality of received data in form of a high Bit Error Rate (BER) or a closed eye diagram. The thesis also provides an overview of serial links with the mostly commonly used signaling techniques, Non-Return-to-Zero (NRZ) and 4-levels Pulse Amplitude Modulation (PAM4). Next, it presents the increase of the versatile equalization possibilities with the shift from NRZ-based to PAM4-based serial links. Then it emphasizes the channel equalization challenges for the latter case as it is dominating the most recent serial links targeting multi-gigabit rates. The thesis also provides a complete analysis of the different major blocks used in channel equalization in serial links and the corresponding design metrics. Then it shows the immense need of having an adaptation algorithm to control the equalizers deployed in the two ends of a serial link: transmitter (TX) and receiver (RX) to provide adaptive compensation for channel imperfections. A MATLAB model for one of the state-of-the-art serial links is built. This model is used to validate the effectiveness of using such adaptation algorithms like the most frequently used Least Mean Squares (LMS) algo¬rithm with the usage of a practical long reach channel model. The thesis proposes the utilization of one evolution algorithm, particle swarm optimization (PSO), instead of LMS. A comparison between both is presented showing the different convergence rates between both.