How do you calculate the energy-optimal frequency for a sub-threshold MCU in a battery-less IoT sensor?

Short answer

Assess the candidate's capability to design sophisticated DVFS algorithms and power management strategies for severely energy-constrained IoT devices operating on harvested energy.

Race-to-sleep ignores static leakage

The naive approach to power management in embedded systems is the "race-to-sleep" strategy: executing workloads at the maximum possible frequency to return to a deep sleep state as quickly as possible. For severely energy-constrained, battery-less IoT devices operating in sub-threshold or near-threshold regimes, this is a fatal flaw. While dynamic power scales quadratically with voltage, executing tasks at minimum voltage increases execution time. In modern sub-threshold CMOS, static leakage current dominates the energy budget, meaning running too slowly consumes more total energy due to prolonged leakage.

Balancing dynamic power and static leakage

An authoritative Dynamic Voltage and Frequency Scaling (DVFS) governor must calculate the exact "energy-optimal" frequency for compute-bound tasks. This is the precise operating point where the reduction in dynamic power perfectly balances the increase in static leakage energy caused by extended execution times.

For memory-bound tasks, where the CPU stalls waiting for non-volatile memory, lowering the CPU frequency and voltage yields massive energy savings with negligible performance degradation. Utilizing hardware performance counters to estimate instruction-level parallelism in real-time is essential for categorizing these workloads accurately.

The cost of transition overheads

State transitions are not free. Adjusting the voltage requires the on-chip DC-DC converter to charge or discharge internal capacitors, consuming energy and introducing latency during which the processor stalls. Transitioning to a lower power state for a highly transient task frequently consumes more energy in transition overhead than it saves during execution. The DVFS algorithm must incorporate transition penalties into its cost function, maintaining the current state if the calculated savings do not exceed the overhead.

Adaptive thresholding and opportunistic execution

An energy-harvesting system must co-design task scheduling with DVFS. When the supercapacitor State of Charge (SoC) is high and energy yield is abundant, the governor should aggressively scale up frequency to process queues and minimize radio active time. When SoC is critical, the system must force a sub-threshold regime, sacrificing performance to avoid a brownout reset. Strictly periodic tasks (like sampling) must be isolated from opportunistic tasks (like DSP and transmission), deferring the latter until the energy harvester operates at peak efficiency.

flowchart TD
    A["Task Arrival"] --> B["Estimate Task Duration & Type"]
    B --> C["Monitor Supercapacitor SoC"]
    C --> D{"SoC Level?"}
    D -- "Critical" --> E["Select Minimum V/F (Sub-threshold)"]
    D -- "Nominal" --> F["Calculate Energy-Optimal V/F"]
    D -- "Abundant" --> G["Select Maximum V/F (Race-to-Sleep)"]
    F --> H{"Transition Overhead < Savings?"}
    H -- "Yes" --> I["Initiate DVFS Transition"]
    H -- "No" --> J["Maintain Current State"]
    I --> K["Execute Task"]
    J --> K
    K --> L["Update Energy Models"]
    E --> K
    G --> K

Optimising DVFS for energy-harvesting IoT nodes requires moving beyond simple race-to-sleep strategies to implement energy-optimal frequency selection that balances dynamic power savings against static leakage penalties and state transition overheads, continuously adapting to both workload profiles and unpredictable energy generation.

© 2026 Preptima. Originally published at preptima.com.