Abstract—The design of today’s embedded systems involves a complex Design Space Exploration (DSE) process. Typically, multiple and conflicting criteria (objectives) should be optimized simultaneously such as performance, power, cost, etc. Usually, Multi-Objective Evolutionary Algorithms (MOEAs) are used to explore a large design space with a finite number of design point evaluations, providing the designer a set of tradable solutions with respect to the design criteria. Analyzing how such evolutionary algorithms searched the design space, understanding the characteristics of the optimum design points, the effect of design parameters on each objective and the relationships between different criteria is of invaluable importance to the designer. To this end, this paper proposes a novel interactive visualization tool, VMODEX (Visualization of Multi-Objective Design spacE eXploration), to realize the search dynamics of a MOEA and to visualize where the optimum design points are located in the design space and what objective values they have. In our tool, we provide several interactive capabilities, which enable designers to look at the exploration data from different perspectives and provide better analysis of the search results. Keywords—Design space exploration, embedded systems, multi-objective evolutionary algorithms, visualization I. INTRODUCTION The complexity of modern embedded systems has led to the emergence of system-level design. A key issue of system-level design is the notion of high-level modeling and simulation in which the models allow for capturing the behavior of system components and their interactions at a high level of abstraction. As these high-level models minimize the modeling effort and are optimized for execution speed, they can be applied at the very early design stages to perform, for example, architectural Design Space Exploration (DSE). Such early design space exploration is of eminent importance as early design choices heavily influence the success or failure of the final product. System-level simulation frameworks that are deployed for DSE of embedded systems that are based on heterogeneous Multi-Processor System-on-Chip (MPSoC) architectures, usually use independent application and architecture models. The application model describes the functional behavior of the system expressed as processes (computations) and channels (communications). The architecture model represents the hardware components in the system, such as processors, reconfigurable modules, memories, etc. Then, different mappings of processes and communication channels to the various architectural components are evaluated by simulation to find the optimum mapping solutions. Each mapping decision taken in this step corresponds to a single point in the design space. Generally, for designing complex embedded systems, multiple conflicting criteria need to be considered simultaneously such as performance, power, cost, etc. Therefore, there exist no single optimum solution, which simultaneously optimizes all objectives. Instead, a set of optimal solutions, denoted as the Pareto optimal set or non- dominated set, has to be found. This is the set of those solutions for which one objective cannot be improved further without causing a simultaneous degradation in at least one other objective. These optimal solutions provide the designer trade-offs between the design objectives. In order to find a Pareto optimal set, the designer should ideally evaluate and compare every single point in the design space. However, such an exhaustive search quickly becomes infeasible, as the design space grows exponentially with the size of the application(s) and the number of possible architecture components. In general, to trim down an exponential design space into a finite set of points, which are more interesting (or superior) with respect to design criteria, design space pruning can be used. In [1], e.g., the mapping decision problem is formulated as a multi-objective optimization problem in which three criteria are considered: the processing time, energy consumption and cost of the architecture. To solve this problem, a Multi-Objective Evolutionary Algorithm (MOEA) has been used to achieve a set of optimal alternative mapping decisions under the aforementioned criteria. MOEAs evaluate a population of design points (solutions) over several iterations, called generations. With the help of genetic operators, a MOEA progresses iteratively towards the best possible solutions. As the searched design space still is vast, interpreting all evaluation data and understanding how the MOEA searches through or prunes the design space is cumbersome. Such analysis is, however, essential to the designer as it provides insight into the “landscape” of the design space (e.g., indicating which design parameters are more important than others). To address these problems, we develop a novel interactive visualization tool, VMODEX, to understand how an evolutionary algorithm, such as presented in [1], searches the design space, where the optimum design points are An Interactive Visualization Tool for the Analysis of Multi-Objective Embedded Systems Design Space Exploration Toktam Taghavi, Andy D. Pimentel Computer Systems Architecture Group, Informatics Institute University of Amsterdam, Amsterdam, the Netherlands {T.TaghaviRazaviZadeh, A.D.Pimentel}@uva.nl