Space applications nowadays require onboard processors combining high performance with strict reliability and energy-efficiency constraints. The AI/DSP workloads for image processing, communications, and autonomous navigation must operate under limited power budgets and in harsh radiation environments, while also being affected by PVT variations. To tackle this challenge, the FPGAs offer both hardware acceleration and low-level configurability, and thus, enable customization of performance-per-Watt and system reliability. In this talk, we present radiation testing and mitigation techniques applied to our FPGAs in the lab, we show how to analyze the PVT variability of commercial devices, and we propose frameworks to fine-tune the operating margins via DVFS and PVT-aware placement. Furthermore, we explore mapping strategies and DVFS on AMD/Xilinx Versal to efficiently execute GEMM kernels and support emerging AI applications.