### The DSE approaches has been increasing in the recent

The modern Electronic Systems design often confronts sophisticated requirements with a wide range of choices from operating modes, components, interfaces, algorithms, etc. To deal with the problem, Codesign technique tries to bring the design process to a higher level of abstraction with the goal of improving the design productivity by high-level system modeling and simulation 4. By using this approach, we can optimize for better modeling effort and execution speed. Therefore, DSE has become an essential part of Codesign technology 5. In the next sections, DSE will be the focus of the discussion. The DSE phase will be conducted at the beginning of the development process, at which a large variety of different design alternatives will be considered. The decisions made after DSE have significant influence on the later phases, thus heavily affecting the success or failure of the final design. However, this process is very time consuming and challenging since the design space that needs to be explored is typically enormous. Therefore, the need for efficient and effective DSE approaches has been increasing in the recent years 3.During DSE phase, the target is multiple objectives optimization, where the ambition is to improve performance, power/energy consumption, cost, etc. simultaneously. These objectives are called parameters of DSE, and are different across systems. Generally, there will be a solution that best fits with each parameter, which is evaluated by a so-called fitness function. Since the parameters are often in conflict, i.e. the enhancement of one quality number often comes at a deterioration of another, there cannot be a single optimal solution that simultaneously optimizes all objectives 3. Therefore, optimal decisions need to be taken in the presence of trade-offs between different design criteria. Hence, we can typically compare different solutions in the case of multiple objectives by using the Pareto front relation which indicates the dominance of some solutions over others. The example of a Pareto front is shown in Fig. 4. Here, f1 and f2 are fitness functions corresponding to each solution with a preference of low values, the set of solutions from A to F belongs to Pareto front of solutions because they are more optimized with respect to f1 and f2. The solution H is dominated by solutions B, C, and D because their values are lower for both f1 and f2 compared with that of H. Similarly, solution H is a better solution in comparison with M, N, and O. Finally, some of the solutions are not comparable to H since they have better score for one objective but worse for the other. Therefore, it is up to the designers to decide which of the set of solutions is the most adequate.The results obtained from the DSE step is the Pareto-optimal front of solutions where designers now have only a few of best choices left for further investigation. An efficient DSE requires both evaluation and navigation approaches 6. In fact, because design spaces grow exponentially with increasing number of parameters, an exhaustive search (evaluating every possible design point) is unfeasible. With this method, the number of design points can easily outgrow any practical limit on execution time which is limited by the evaluation algorithm or simulator. Therefore, many advanced methods are proposed for the search problem, which can be classified into four different classes 3. The first class is heuristics and pseudo-random optimization which attempt to reduce the design space under scrutiny and focus the exploration on regions of interest 7, 8. The goal of algorithms in this class is to drastically reduce the number of solutions to evaluate (e.g., using the sum instead of the product of the number of tunable parameters 7). For example, they try to identify sub-spaces (called clusters) in the design search space that can be efficiently explored in an exhaustive fashion 8. Subsequently, the global Pareto front can be built from the Pareto-optimal solutions of each partition. The exploration results of the algorithms in this class remain sub-optimal due to their simplicity.