Resumen:
Timing and crosstalk are determined by the complicated joint effects of layout, electrical, and logic parameters, which make conventional estimation methods either too slow or very inaccurate. Thanks to their strong knowledge extraction and reuse capability, machine learning (ML) techniques have been adopted to improve the predictability of timing and crosstalk effects at different design stages. Many of these works develop net-based models, whose outcomes provide detailed information to guide timing optimization and crosstalk avoidance techniques. In this chapter, we first present a comprehensive review of net-based ML-aided approaches for timing and crosstalk prediction. Then, four representative case studies are introduced in detail with the focus on problem formulation, prediction flow, feature engineering, and machine learning engines. Finally, a few conclusion remarks are given.
