FPGA & 5G: Simplify deployment of FPGA and associated firmware/software @scale

Say, you have 5G-DU that leverage FPGA for O-RAN, FEC acceleration; 5G-UPF that leverage FPGA for UPF acceleration; or applications that leverage FPGA for ML/DL & Video encode/decode acceleration . Hence, you want to make your Edge platform expose FPGA to applications. FPGA being programmable, you like to program the FPGA with bitstream based on the application/network-function being deployed. If new application or network function is getting deployed on the node, you want to reprogram the FPGA with new bitstream. FPGA based acceleration exposure in K8s environment involves basic infrastructure setup phase and then the acceleration setup phase that is based on the type of applications being deployed. Complexity goes up when the upgrade of infrastructure (image upgrade) is required. Number of steps required in each phase could be in tens. Complexity goes up with the number of FPGAs in each node, multiple nodes in each edge and multiple edge locations. Hence, automation becomes important. In this webinar, we will talk about one of the most popular FPGA cards, N3000 from Intel. N3000 is used for multiple accelerations. DU vendors use FGPA to accelerate L1 processing. Analytics application vendors use it for accelerating training & inferencing. This talk takes one through the complexity in enabling N3000 and updating application bitstreams that involves verifying the image signatures. Finally, it talks about few mechanisms to simplify this process – K8s operators, Central automation system and hybrid approaches.