A Meta-Framework for Design Space Exploration

Complex software systems have a large number of choices in terms of selection of software components and hardware architectures for implementation. These design choices create a large space of possible design solutions called the design space. The design process requires exploring through this design space to find valid design solutions before the actual implementation. Design space exploration (DSE) is the process of searching through the design space to find feasible and optimal design solutions. The main challenge in DSE is to deal with an exponential number of design alternatives, which is further complicated by the various conflicting requirements. Thus, there is a clear need for tool support to automate DSE. Over the years domain-experts have frequently relied on different search techniques (mathematical programming, constraint techniques, heuristics) to automate DSE. Different approaches can be used to integrate the search techniques into the existing development environment. One of the approaches is development of black-box frameworks for DSE. Literature survey reveals that existing DSE frameworks are domain-specific where the representation and exploration algorithm are tightly coupled with domain-dependent assumptions. Besides this, most frameworks use a single search technique to automate exploration, which may not work well as the problem evolves over time. Thus, most frameworks lack reusability and flexibility. This paper presents a survey of existing approaches for automated DSE. The goal is to highlight the requirements and research challenges in the development of a meta-framework for design space exploration. This framework can be meta-programmed by domain-experts to work for a class of DSE problems in any domain, thus providing reusability. The meta-framework also supports a set of search techniques providing the flexibility to choose a technique based on the characteristics of the problem instance.