An efficient FPGA-based memory architecture for compute-intensive applications on embedded devices

FPGAs are increasingly being utilized to accelerate real-time compute and data intensive applications on embedded platforms. FPGAs achieve high speed-performance by exploiting a variety of parallelisms in computations. However, on-chip memories of current FPGAs are typically dual-port, which hinders multiple simultaneous read/write (R/W) operations required for parallel processing. Although several multi-ported memories are proposed in the literature to address this issue, there is a tradeoff associated with the existing architectures; that is, increasing the number of ports, reduces the total available memory on chip for the block RAMs to store essential data for real-time processing. This tradeoff is not desirable, especially for real-time compute/data intensive applications on embedded platforms, due to the significant amount of time spent on accessing the external memory. In this research work, we introduce a novel and efficient multi-ported memory architecture to bridge the gap between this tradeoff. Experiments are performed to evaluate the feasibility and efficiency of our multi-ported memory architecture. Our unique memory architecture is generic and parameterized. Our memory can be configured to provide a sufficient number of ports for simultaneous R/W operations, while utilizing the total available on-chip memory to store the essential data.