Author:

Abstract:

As the miniaturization of semiconductor technology continues, electronic systems on chips offer a more extensive and more complex functionality with better performance, higher frequencies and less power consumption. Whereas digital designers can take full advantage of the availability of design automation tools to build huge systems, the lack of support by computer programs for different abstraction levels makes analog design a time-consuming handcraft which limits the possibilities to implement large systems. Various approaches for finding optimal values for the parameters of analog cells, like op amps, have been investigated since the mid-1980s, and they have made their entrance in commercial applications. However, a larger impact on the performance is expected if tools are developed which operate on a higher abstraction level and consider multiple architectural choices to realize a particular functionality. In this work, the opportunities, conditions, problems, solutions and systematic methodologies for this new generation of analog CAD tools have been examined. First, the characteristics of the analog design process are systematically analyzed and several approaches for automated analog synthesis are summarized. Comparison of their properties with the requirements for high-level synthesis of analog and mixed-signal systems results in a new design paradigm: the high-level design flow based on generic behavior. This design approach involves a modeling strategy using generic behavioral models and a synthesis strategy leading to the exploration of a heterogeneous design space containing different architectures. Generic behavioral models allow to easily represent a wide range of diverse architectures with several non-idealities and, at the same time, to exploit specific aspects of a class of systems leading to efficient evaluation of performances. Hence, they stand midway between simulations of general-purpose languages and dedicated approaches using specific behavioral models. This work contributes to the development of generic behavioral models by defining and implementing two novel models. The first one adopts a time-domain approach and is suited for classes like Delta-Sigma modulators and sampled-data systems. Experiments with a custom SystemC module have been performed. For the second model, a new frequency-domain framework has been developed (the