Abstract-A Σ∆ GmC integrator refinement flow is presented. The classically simplified GmC integrator small-signal model was upgraded to be extremely accurate by considering the complete transistor small-signal model. A circuit-level knowledge-based tool was used to execute the designer defined sizing procedure and to extract small signal parameters. By associating the symbolic transfer function to small-signal parameters, the flow, entirely implemented with C++, is able to compute poles and zeros to permit precise behavioral simulations. A 2 nd order Σ∆ modulator was chosen to visualize performance degradations while the specifications were not achievable.
I. INTRODUCTIONThe essential target of simulation is evaluation of performances. When measuring non-linear devices such as Σ∆ modulators, a time-domain analysis is required. Performances are extracted through spectral analysis, requiring a sufficient number of simulation samples to be accurate enough. Running those simulations at transistor-level is time consuming. Designers have to deal with levels of abstraction to handle the trade-off between simulation speed and accuracy of results. That's why a great challenge is to refine behavioral models with a large amount of non-idealities.Σ∆ modulators are well known for their performances. They are used in high resolution audio and wireless applications. Continuous-time modulators consume low power and operates at high sampling rate. Many publications have presented non-idealities modeling either in the discrete-time case [7]. They model poles/zeros limitations, clock jitter, noise, saturation, harmonic distortion, other non-linearities. Characterization of these non-idealities is still often based on circuit-level simulation. Expressing non-idealities as function of circuit small signal parameters is always valuable since it permits automatic model refinement and architecture exploration [8].Focusing on poles/zeros characterization, symbolic analysis [2] [9] has proved its worth to refine behavioral transfer function. This way, technology independent functions can describe a circuit. By choosing a technology, small-signal parameters of the circuit allow to finally determine the values of poles and zeros. Such extraction is generally simulationbased [9]. There are many variants, for example, [7] introduced curve fitting also based on simulation.In this paper, we investigated continuous-time Σ∆ GmC integrators. First, we present a precise small-signal model of a GmC integrator. Then, a circuit-level knowledge-based