Platform-based System Level Optimization for Mixed Systems
Alberto L. Sangiovanni-Vincentelli and Xuening Sun
Traditional electronic designs have separated the design of analog and digital components. This design gap helps to reduce the complexity of design; however, it generally results in sub-optimal systems since the decisions of system partition and resource allocation are made in an ad hoc manner based on the expertise of the system designer with almost no knowledge of the feasible physical implementations.
Although there has been an increased interest in mixed-signal design in the recent decade, the system level optimization of analog/digital partitioning and resource allocation have remained relatively unexplored. The focus of this research is to develop a platform-based design methodology for system level optimization. Platform-based design is a meet-in-the-middle methodology where system level design decisions are supported by performance models obtained via bottom-up extraction of the physical implementation . The goal is to develop a framework such that the entire system, consisting of analog and digital blocks, can be designed and optimized. We hope such a methodology can help in the design of any signal processing system. The most direct and relevant application of this research is on ultra-low power radio design. Modern ultra-low power radios for wireless sensor nodes have extremely tight constraints in terms of power, area, and timing. These constraints sometimes cannot be met using the traditional design flow. In short, analog/digital co-design of the entire radio system, from the RF front-end to digital baseband, is required to obtain the optimal design that meets the constraints. This research is work in progress, currently we are working on defining the common semantic domain to describe the analog and digital components.
Figure 1: Simplified illustration of platform-based mixed system optimization applied to radios. As shown, we do not attempt to replace traditional design flows, but rather, we are building a higher abstraction layer such that the entire system can be considered for design space exploration. At this higher abstraction level, system partition and resource allocation decisions are products of the optimization engine. Because we are using PBD, the physical implementation performance parameters are retained even at the higher abstractions. This guarantees feasibility of the output of the optimization engine.
- F. De Bernardinis and A. Sangiovanni Vincentelli, "A Methodology for System-level Analog Design Space Exploration," Proceedings of Design, Automation and Test in Europe Conference and Exhibition, Vol. 1, February 2004, pp. 676-677.