The conservation core accelerator presents a promising avenue for improving the computational efficiency of specific applications. However, its design space is ultra-high-dimensional, significantly increasing the exploratory effort required to identify the optimal design across performance, power, and area metrics. Furthermore, the discrete nature of the microarchitecture space renders conventional search methods ineffective. To tackle these challenges, we propose the Discrete Continuous Transformation to speedup Design Space Exploration, namely DCTDSE. It can operate in either offline or online mode. In offline mode, it transforms the original discrete design space into a continuous space, builds predictive models, performs parallel gradient-based optimization, and maps the results back to the discrete domain. In the online mode, DCTDSE refines the models by iteratively resampling previously found solutions, thereby enhancing exploration quality while maintaining moderate runtime overhead. Experimental results indicate that DCTDSE achieves a 3.9× to 40× speedup over benchmark methods in offline mode. In online mode, it provides a 2.5× speedup, with a 21% reduction in exploration quality relative to the most accurate comparison method.

Coarse-Grained Reconfigurable Arrays (CGRAs) are a promising architecture that warrants thorough design space exploration (DSE). However, traditional DSE methods for CGRAs often get trapped in local optima due to singularities, i.e., invalid design points caused by CGRA mapping failures. In this paper, we propose a singularity-aware framework based on the integration of reinforcement learning (RL) and imitation learning (IL) for DSE of CGRAs. Our approach learns from both valid and invalid points, substantially reducing the probability of sampling singularities and accelerating the escape from inefficient regions, ultimately achieving high-quality Pareto points. Experimental results demonstrate that our framework improves the hypervolume (HV) of the Pareto front by 23.56% compared to state-of-the-art methods, with a comparable time overhead.

Surrogate model-based design space exploration (DSE) is the mainstream method to search for optimal microarchitecture designs. However, it is hard to build accurate models for accelerator-rich systems within limited samples due to its high dimensional characteristic. Moreover, it is easy to fall into local optimal or difficult to converge. To solve these two problems, we propose a DSE flow based on active learning, namely ARS-Flow. It is featured with Pareto-region-oriented stochastic resampling method (PRSRS) and multiobjective genetic algorithm with self-adaptive hyperparameter control (SAMOGA). Taking the gem5-SALAM system for illustration, the proposed method can build more accurate models and find better microarchitecture designs with acceptable runtime costs.