Specification and design-space exploration for high-level synthesis of analog and mixed-signal systems

High-level synthesis of mixed-signal systems is the task of automatically producing analog/digital implementations from system specifications described with a Hardware Description Language (HDL). Main arguments for high-level synthesis are besides the correctness of design by construction, its ability of achieving short design cycles with less effort. Analog and mixed-signal synthesis constitute critical research problems because of the increasing demand for analog/digital applications. Emphasis of the thesis is on specification languages and techniques for automated analog and mixed-signal synthesis. The thesis begins by introducing synthesis-oriented computational (formal) models for mixed-signal and analog systems. Formal models were used for developing sound specification languages and synthesis techniques. The thesis continues with discussing issues on specification languages for synthesis of analog and mixed-signal systems. High-level synthesis of analog systems is the second topic of the thesis. High-level specifications do not explicitly indicate the kind, connection and parameters of the analog circuits of the final implementation. Hence, our analog synthesis methodology includes following tasks: (1)  Architecture generation for mapping a specification to hardware architectures, (2) Generation of equational performance models that relate system parameters to circuit parameters and (3) Optimization of system parameters using the performance models. The thesis presents an efficient technique for linear performance model generation. The method exploits system hierarchy and regularity to produce polynomial-size performance models. Two architecture generation methods are presented. The first algorithm follows the traditional pattern-matching approach. The second method is a novel approach based on tabu-search optimization. It does not request a pattern library and thus, it does not enforce implementations only in terms of a static set of mapping patterns. The thesis also suggests a hierarchical parameter optimization technique for optimizing system area and AC behavior of an analog system. The third part of the thesis addresses high-level mixed-signal synthesis. Mixed-signal synthesis considers the switching noise (DSN) transferred from the digital to the analog domain as the major design concern. We assumed an analog-domain centered approach for mixed-signal synthesis in that the analog domain is completely designed (including placement and signal routing) before digital synthesis is conducted. We present an original method for AC and transient analysis of DSN propagation. The technique includes modeling of substrate coupling, contacts to the power supplies and power buses. We also suggest an original technique for operation scheduling and resource binding given that latency and DSN constraints have to be contemplated. We conducted multiple experimental studies to validate the correctness of the proposed methods and study their effectiveness. Experiments targeted various applications i.e. telecommunications or signal processing. Results motivate the ability of the suggested techniques to automatically generate a wide range of solutions with a minimal amount of design knowledge requested as an input.