Mender-FPGA: An Open Source Framework for FPGA Remote Update for ML Applications

The last decade has seen a massive adoption of FPGA devices in various fields like electronics, robotics, machine vision, etc. Over-the-Air (OTA) updates are becoming an important feature for internet-connected intelligent embedded devices. Such remote updates offer flexibility in updating an existing system remotely without physical access to it. The intelligence-based embedded devices' software modules (like AI/ML models) may need to be updated frequently. The traditional approach to update Field Programmable Gate Array (FPGA) uses JTAG programming, which is limited to wired connectivity with 5 to 10 devices. However, when these FPGA devices are deployed at a remote location, the need for automatic wireless or wired updates increases over manual updates. In this paper, we discuss the design and implementation of an open-source framework for over-the-air and remote wired update for SoC FPGA devices using the open-source Mender platform. We specifically address two features: (1) full system update, (2) delta-update to update the FPGA functionality similar to full software updates and software patches respectively, and include a rollback mechanism for device recovery. We implemented and tested all the above features on Zed board to demonstrate the proposed approach and framework for machine learning applications on such devices. Our result shows the successful implementation of the remote update method for FPGAs and is the first to develop an open-source framework targeting Zed board.