A hardware framework for on-chip FPGA acceleration

In this work, we present a new framework to dynamically load hardware accelerators on reconfigurable platforms (FPGAs). Provided a library of application-specific processors, we load on-the-fly the specific processor in the FPGA, and we transfer the execution from the CPU to the FPGA-based accelerator. Results show that significant speed-up can be obtained by the proposed acceleration framework on system-on-chips where reconfigurable fabric is placed next to the CPUs. The speed-up is due to both the intrinsic acceleration in the application-specific processors, and to the increased parallelism.