Memory allocation and optimization in system-level architectural synthesis

In this paper, we present a novel approach to optimally allocate memory resources in a system-level synthesis flow, which converts a dataflow style system description (synchronous data flow) into the register-transfer level description in the specified implementation style (ASIC, FPGA or CGRA). The first problem is encountered by the synthesis flow is that since it covers different implementation styles, a generic model is required to support resource allocation and optimization. The second problem is the memory allocation method to optimally allocate memory resources in the RTL model. The contribution of this paper has two parts, which are 1) a generic memory model for different memory architectures in ASIC, FPGA and CGRA, and 2) a memory allocation and optimization method for optimally allocating storage elements in the intermediate representation with actual implementations (e.g. on-chip SRAM for ASIC, memory controller and off-chip SDRAM for FPGA). The memory allocation method is an implementation style dependent procedure and has three steps: architecture independent optimization, resource allocation and architecture depended optimization. The experimental result shows that the proposed method is efficient and effective. The automatically generated implementation uses only approximately 4% more resources compared to manual implementation. The fast and automatic memory allocation method enables fast design space exploration that requires little effort form the system designer.