Application of Machine Learning in De-embedding of Signal Integrity Parameters for High Speed Serial Link

This Signal Integrity measurements are imperative to High-Speed Serial Link characterization. However, as new protocol definitions introduce higher data rates and enhanced tighter specifications, the characterization activity becomes challenging with existing equipment. For high-speed serial links, for data rates above 1 Gbps, board traces/channel affects the Signal Integrity (SI) measurements (like peak-to-peak voltage and jitter) that lead to the test failing the target specification. Therefore, de-embedding is performed to isolate the performance of the DUT (Transmitter) and eliminate the degradation due to the channel's effect. In essence, the same SI evaluation is being done twice (once on non-de-embedded signal and again on deembedded signal) that results in time overhead and increased cost of equipment and software required for the de-embedding process. In this paper, we explore the use of machine learning-based methods in de-embedding a broad class of signal integrity (SI) measurements - jitter (and all its components) and amplitude of the signal. In this paper, we talk about different ML models' and their efficiency in predicting the de-embedded SI measurements. We discuss in detail the conventual deembedding approach and the proposed methodology. We show how the proposed method can save cost in terms of both time and money with an average error of less than 7% and timesaving of almost 25%.