May_EDFA_Digital

edfas.org 35 ELECTRONIC DEV ICE FA I LURE ANALYSIS | VOLUME 24 NO . 2 is proficient with machine learning. Finally, Aysu’s team developed power-balancing circuits and quantified that they can remedy the side-channel leakage at the expense of higher area and power consumption. MACHINE LEARNING ENABLED ANALYSIS OF HIGH-SPEED SIGNAL BUSSES A research team led by Prof. Xu Chen at Illinois devel- oped a Stochastic Finite-Difference Time-Domain (FDTD) method to simulate printed circuit board and package structures, despite uncertainty about the geometry and material properties—those uncertainties arising from manufacturing variations. The FDTD method is a well- established approach to simulating the time evolution of electromagnetic fields. Those fields are determined uniquely by the systemgeometry, materials, and sources. Statistical variation in the system parameters causes uncertainty in the field and current values; those values affect performance metrics such as bit error rate and eye opening, the latter being a measure of the signal to noise ratio at the input to the data sampler. However, in the presence of uncertainty, conventional FDTD can provide only the average field values. Therefore, a stochastic FDTD simulation method must be used to obtain the perfor- mance metric statistics. The stochastic FDTD method developed by Chen’s group is formulated using the intrusive Stochastic Galerkin Method, which can provide the electric and magnetic field statistics from a single simulation. Traditionally, such simulations were performed using Monte Carlo analysis, which involves many time-con- suming simulations. Graduate student Der-Han Huang received the Best Student Paper Award at the 2020 IEEE NEMO (Numerical Electromagnetic and Multi- physics Modeling and Optimization) conference for his early work on this topic. POWER, PERFORMANCE, AREA Recently, CAEML supported two research projects that aim to improve PPA prediction in an early phase of the IC design process. The power, performance, and area (PPA) of digital blocks can vary over almost a 3:1 range, depend- ing on the parameters input to the place and route (P&R) Fig. 1 Pareto fronts of area and speed over different process corners for an OpenRISC 1200 core. [2]