System Services for Reconfigurable Hardware Acceleration in Mobile Devices

FPGAs have been deployed to provide custom hardware acceleration for applications such as networking, vision, and machine learning in embedded devices. A mobile device serves increasingly diverse computing applications under constrained resources, hence, making FPGA-based acceleration more attractive to achieve high performance, and yet, energy efficiency. However, due to lack of systematic interface between mobile application software framework and FPGA design flow, software developers avoid using FPGA accelerators in developing Apps for mobile applications. In this paper, we introduce a software framework that integrates the Android operating system with programmable hardware accelerator design flow. We present an acceleration service in Android platform that manages the mapping of compute-intensive kernels to the accelerators on FPGA. The proposed framework enables multiple applications to access multiple FPGA-based accelerators, simultaneously. We present an Android OS porting on the Xilinx Zynq SoC, equipped with dual-core ARM Cortex-A9 processors and programmable logic. The experimental results show performance gain in case studies of canny edge detection algorithm, digit recognition algorithm, and neural network, which are used by mobile Apps.