Towards performance modeling of 3D memory integrated FPGA architectures

Abstract

Recent advances in three dimensional integrated circuits have enabled large amounts of memory to be stacked in layers and accessed by a logic unit using high bandwidth vertical interconnects. Several 3D architectures have been proposed with different organizations of memory and logic layers. In particular, 3D stacks of memory dies can be interfaced with a reconfigurable logic layer such as FPGA to enable highly optimized implementation of memory-intensive applications.We refer to these as 3D Memory Integrated FPGAs. Mapping algorithms to such architectures is a challenging task due to the complex interaction between memory and logic and the relation between energy consumption and memory access. Performance modeling of these architectures can enable the design space to be systematically explored while mapping a specific algorithm. In this paper, we analyze the current landscape of 3D Memory Integrated FPGAs and identify the key parameters that have a significant impact on bandwidth and energy. We specify an “abstract architecture” that captures the features of such architectures and provide a parameterization of the design space with the eventual goal of developing a performance model for optimizing algorithm implementation.