Multilevel Modeling Methodology For Reconfigurable Computing Systems Based On Silicon Photonics

Reconfigurable computing systems, e.g. FPGA, represent an increasingly attractive architectural solution for high-end supercomputing due to high aggregated computational resources, high energy-efficiency and flexibility. However, to further increase the computing bandwidth of such systems while decreasing the energy consumption, emerging technologies such as silicon photonics, are urgently needed. In this context, we have proposed the OLUT architecture, which could serve as the core building block of prospective silicon photonic reconfigurable computing systems. To implement this concept, a modeling methodology is required to evaluate the impact of technological constraints on system metrics. In this paper, we propose a multilevel design methodology to optimize the performance of reconfigurable computing systems based on silicon photonics by exploring the design space from multiple perspectives of system dimensions, device parameters and technologies. This method allows us to efficiently implement a functional and energy-efficient reconfigurable computing architecture that could reach similar to 50fJ/bit logic operation at a BER of <10(-18), as a potential answer to the future demands of reconfigurable computing.