Compiler Infrastructure for Specializing Domain-Specific Memory Templates

Abstract: Specialized hardware accelerators are becoming important for more and more applications. Thanks to specialization, they can achieve high performance and energy efficiency but their design is complex and time consuming. This problem is exacerbated when large amounts of data must be processed, like in modern big data and machine learning applications. The designer has not only to optimize the accelerator logic but also produce efficient memory architectures. To simplify this process, we propose a multi-level compilation flow that specializes a domain-specific memory template to match data, application, and technology requirements.