Recent research has shown subthreshold operation to reduce active energy in low-power MCUs , , . However, some applications impose additional constraints. For battery-powered sensor nodes deployed in remote locations, the MCU may lie dormant for long periods, only waking up when a sensor detects activity. For such cases, the MCU needs very low sleep power to maximize battery lifetime, deterministic real-time response to capture rare sensor events, and energy-efficient operation with enough compute and memory to run useful workloads. This work shows a 65nm Arm Cortex-M33 SoC for constrained battery-powered sensor nodes achieving 10 nW sleep power (4 KB retention) with fine-grained DVFS and performance regulation from 0.8 MHz (0.40 V) to 38 MHz (0.75V) to address a range of real-time requirements across the operating range of 0-to-85° C and 1.0-to-1.5V battery voltage. Digital circuit optimizations reduce active power to 47μ W (20pJ/cycle) on a high-activity keyword spotting (KWS) workload.