Library finding for high-level synthesis of analog systems

The behavior of an analog system may be represented as a signal flow graph (SFG), where nodes represent operations (transfer functions and non-linear functions) on signals and edges represent signal flow. Library binding or technology mapping transforms the given behavior into a netlist of cells from the target library, optimized for performance. Analog library binding is different from its digital counterpart, and presents new problems that we address in this work. (1) The transformation from analog behavior to structure is not unique. Our algorithm performs architecture generation and explores various possible circuit architectures for the best area and noise performance. (2) The occurrence of feedback in analog systems introduces cycles in the SFG. Hence the structural approach to analog library binding involves matching and covering of directed graphs with cycles. We employ a multi-level pattern library representation, together with hierarchical matching and covering approach to solve this problem. (3) In order to have acceptable signal-to-noise ratio at the system´s output terminals, low-noise circuit structures are selected by optimizing both the gain distribution and the noise figure (NF) of the system. Due to its frequency-dependent nature, noise figure is determined over multiple frequency points during fitness evaluation