Transmitter and Receiver Equalizers Optimization for PCI Express Gen6.0 based on PAM4

An Overview of RF and Microwave Research in Latin America: Scanning Latin American Research on Microwaves

“…ITESO -The Jesuit University of Guadalajara, focuses its research on CAD techniques for RF and microwave modeling, design, and optimization of circuits and systems, most of it in close collaboration with local industry [97], [98]. More specifically, ITESO develops research on design optimization methods [99]- [101], surrogate-based approaches [102], [103], neuronal analog fault identification [104], and space mapping techniques [105]- [108], with emphasis on applications to signal-integrity and high-speed interconnects [109], [110], post-silicon validation of high-speed computer platforms, [111]- [113], power delivery networks and power integrity [114], [115].…”

A Multi-Stage CTLE Design and Optimization for PCI Express Gen6.0 Link Equalization

“…A conventional CTLE cannot compensate for the small amount of low-frequency channel loss since its primary objective is to compensate for high-frequency channel losses [11]. Since the slope of the low-frequency loss is quite flat (<3dB/dec), an extra circuit is required.…”

“…We need to ensure the optimal system margin response also meets an eye linearity, elinearity, larger than 0.85, and a vertical eye closure (VEC) below 6 dB. An initial optimization problem can be defined through a constrained formulation, * arg min ( ) u = x xx (11) subject to elinearity(x) > 0.85 and VEC(x) < 6dB, where u(x) is the total area of the PAM4 eye diagram, Additionally, we need to ensure the optimal system response is within a suitable area in the coefficients search space of the EQ map. Here we follow our work in [6] and [11] to redefine the corresponding objective function.…”

“…An initial optimization problem can be defined through a constrained formulation, * arg min ( ) u = x xx (11) subject to elinearity(x) > 0.85 and VEC(x) < 6dB, where u(x) is the total area of the PAM4 eye diagram, Additionally, we need to ensure the optimal system response is within a suitable area in the coefficients search space of the EQ map. Here we follow our work in [6] and [11] to redefine the corresponding objective function. The four responses around u'(x * ) must be at least 80% of the value of u'(x * ), as shown in Fig.…”