Even with the intense activity in the development of such mixed-signal hardware-description languages as Verilog-AMS and VHDL-AMS, behavioral modeling is neither a new nor a difficult concept. The behavioral description of an analog or mixed-signal system is closely coupled with top-down design methodology. That modeling technology, through analog behavioral language, provides a base for design specification and documentation. A direct benefit is improved design efficiency, which better delivers intrinsic performance characteristics.
Fundamentally, a behavioral model is a set of equations between inputs/outputs and internal variables. It also allows multiple modeling approaches to be used from a basic set of equations that is to be passed to a simulation engine. In any case, the equations are application-dependent and can be technology-independent as well, depending on the abstraction level, facilitating mixed-technology, domain simulation.
Behavioral models are defined as customized equations and software. Their primary drawbacks are inefficiencies in the modeling process and the need to develop specialized simulation engines. With today's designer challenges, one must address the requirements of system-level simulation to catch high-level incompatibilities early in the design cycle in order to validate system specifications, generating as a by-product better documentation of the complete design process.
More specifically, one must be able to describe different levels of abstractions or different levels of detail, providing an effective method to manage complexity and optimize simulation speed and efficiency. To accommodate different application domains, the language-based methodology can define the appropriate models or modeling techniques. That lets designers speed simulation by reducing the model detail level, without sacrificing model accuracy.
Mixed-signal HDLs (MS-HDLs) are a specific class of languages with a similar purpose, though they may be at varying development levels in the IEEE standardization process. Ideally, such languages enable multiple benefits from system-level design: reduction of design iterations, model reuse and system-level verification and documentation. In their ability to provide better control of detail, abstraction level and model accuracy, HDLs represent an incremental evolution from current circuit-simulation technology.
Behavioral modeling enables true top-down analog and mixed-signal design in the development of a specification and documentation. Behavioral modeling facilitates architecture-trade-off analysis and verification, and it supports multiple levels of abstraction. The model writer will need to understand language fundamentals to develop effective models for multidomain simulation (e.g., non-electrical), to capture a mathematical representation and to address hierarchical requirements.
HDL-A provides a mechanism for model characterization, verification, numerical robustness, stability, continuity and portability. The process must also reveal second-order effects and non-ideal behavior. Thus, language-based design within a complete design process will extend-but not replace-the traditional, bottom-up design cycle.
Within the applications domain of MS-HDLs, advanced topics for modeling include s- and z-domain, statistical, reliability and noise models. Model building will need to grow from such simple building blocks as op amps, oscillators, PLLs and data converters to sigma-delta modulators and such elaborate sensors as microelectromechanical structures. More applications will use mixed technology, such as electrothermal and mechtronic descriptions. Beyond the enormous manpower investment required to develop such model libraries, the next obvious development will concentrate on portability and manufacturability.
Even among the most common building blocks used for a system-level approach, the use of macromodels-such as an equivalent representation of an operational amplifier presenting such characteristics as gain, bandwidth and slew rate-must be easily correlated with device-level description designs. A behavioral representation will not include, as the device-level description, all the non-ideal effects, but it can still model nonlinear behavior with adequate accuracy. Typical constraints, such as regions of validity of the model, can be embedded within the behavioral description and monitor out-of-range use. In general, a complete model description will be based on parameters that will include terminal and state variables such as algebraic, differential or difference equations; Boolean or state equations; linear and non-linear behaviors; and finally, time-varying or invariant descriptions.
Model accuracy-especially the confusion between accuracy and model detail-is an important concern. The level of abstraction defines the required detail. Thus, it is essential to highlight some basic rules of model development, such as limited inclusion of complex second-order effects, and to focus on orthogonal properties in order to simplify the model description. The model will be smoothed or linearized when appropriate.
A behavioral model is not a direct translation of its implementation; rather, it is the representation of the behavior or effect to be monitored between different building blocks, assuming that a particular aspect may not affect the design and can be abstracted.
The hierarchical nature of any design illustrates the power of behavioral and language-based methodology. The characterization, extraction and optimization methods of the behavioral model to define a complete verification methodology will directly influence the modeling strategy used.
A more general requirement is the mathematical nature of different behavioral models, taking into account numerical noise and accuracy. In addition, it is necessary to consider the continuity issue with first to nth derivatives. Convergence assists in correspondence with Newton or other algorithms to resolve circuit-structure problems. Obviously, as more models are created in the emerging MS-HDL standards, additional tools will be required to assess the robustness of the model or to be used in conjunction with symbolic analysis packages such as Matlab or Maple.