Event-Driven Simulation Methodology for Analog/Mixed-Signal Systems

Abstract

Recent system-on-chip's (SoCs) are composed of tightly coupled analog and digital components. The resulting mixed-signal systems call for efficient system-level behavioral simulators for fast and systematic verifications. As the system-level verifications rely heavily on digital verification tools, it is desirable to build the mixed-signal simulator based on a digital simulator. However, the existing solutions in digital simulators suffer from a trade-off between simulation speed and accuracy. This work breaks down the trade-off and realizes a fast and accurate analog/mixed-signal behavior simulation in a digital simulator SystemVerilog. The main difference of the proposed methodology from existing ones is its way of representing continuous-time signals. Specifically, a clock signal expresses accurate timing information by carrying an additional real-value time offset, and an analog signal represents its continuous-time waveform in a functional form by employing a set of coefficients. With these signal representations, the proposed method accurately simulates mixed-signal behaviors independently of a simulator's time-step and achieves a purely event-driven simulation without involving any numerical iteration. The speed and accuracy of the proposed methodology are examined for various types of analog/mixed-signal systems. First, timing-sensitive circuits (a phase-locked loops and a clock and data recovery loop) and linear analog circuits (a channel and linear equalizers) are simulated in a high-speed I/O interface example. Second, a switched-linear-behavior simulation is demonstrated through switching power supplies, such as a boost converter and a switched-capacitor converter. Additionally, the proposed method is applied to weakly nonlinear behaviors modeled with a Volterra series for an RF power amplifier and a high-speed I/O linear equalizer. Furthermore, the nonlinear behavior simulation is extended to three different types of injection-locked oscillators exhibiting time-varying nonlinear behaviors. The experimental results show that the proposed simulation methodology achieved tens to hundreds of speed-ups while maintaining the same accuracy as commercial analog simulators.