Designing Battery-Free IoT Devices: Energy Harvesting for Embedded Engineers

Circuit board with tiny solar cell and antenna designed for battery-free IoT energy harvesting

A single CR2032 coin cell costs $0.35. Deploying it inside an asset-tracking tag costs $0.35. Sending a technician to replace it three years later costs $15–$50 per visit. Multiply that across 10,000 devices in a warehouse, hospital, or logistics network, and battery replacement becomes the dominant lifecycle cost of your IoT deployment, dwarfing the hardware BOM by an order of magnitude.

You already know this. That’s probably why someone on your team has asked whether you can kill the battery entirely. The answer is yes, but only if you treat the problem correctly. Energy harvesting isn’t a drop-in replacement for a battery. It’s a fundamentally different design discipline where your energy budget dictates every architectural decision, from PMIC selection down to how your firmware handles a cold boot.

This guide walks you through the complete design flow for a battery-free IoT sensor node powered by indoor solar and RF energy harvesting. By the end, you’ll have a validated reference design you can take from bench to production BOM. We’ll follow a specific sequence: energy audit → source selection → PMIC and storage sizing → ultra-low-power MCU/radio selection → firmware duty-cycle optimization → DFM review.

The Energy Budget: The One Step Most Engineers Skip

Most engineers start an energy harvesting design by browsing PMICs. That’s backwards. You need to start with a hard accounting of how much energy your device consumes per duty cycle, and whether any realistic harvester can cover that bill.

Here’s a concrete example. Consider an indoor temperature/humidity sensor transmitting a BLE advertisement every 60 seconds:

+---------------------+----------+-----------+------------+
| State               | Current  | Duration  | Energy (µJ)|
+---------------------+----------+-----------+------------+
| Deep Sleep          | 1.2 µA   | 59.85 s   |   237      |
| Wake + Sense        | 3.5 mA   | 50 ms     |   577      |
| TX (BLE adv.)       | 8.0 mA   | 100 ms    | 2,640      |
+---------------------+----------+-----------+------------+
| TOTAL per cycle     |          | 60 s      | 3,454 µJ   |
| Avg. power required |          |           |  ~57.6 µW  |
+---------------------+----------+-----------+------------+

That ~58 µW average is your design target. If your harvester can’t reliably deliver that with headroom (say, 80–100 µW to account for MPPT losses and storage leakage) your device will brown out. No amount of clever firmware will save you.

A critical practical tip: measure real current profiles with a Nordic PPK2 or µCurrent Gold before trusting datasheet typicals. Datasheets report current under ideal conditions. Your actual wake-and-sense sequence, including oscillator startup and sensor initialization, will often draw 20–40% more than the sum of individual typical specs suggests.

Indoor Solar Harvesting: What a 4 cm² Panel Actually Delivers

If you’ve worked with outdoor solar, recalibrate your expectations immediately. A bright office sits at 300–500 lux. Direct sunlight is 50,000+ lux. That’s a 100× difference, and it translates directly to available power.

At 300 lux, indoor solar panels typically deliver 10–20 µW/cm². That’s the real number from peer-reviewed data (Mathews et al., Joule, 2019), not a vendor’s best-case figure.

Panel technology matters enormously indoors. Amorphous silicon (a-Si) outperforms mono- or polycrystalline silicon in indoor conditions because its spectral response better matches the emission profiles of LED and fluorescent lighting. The Panasonic AM-1816 (a-Si) and Epishine LEH3 (organic PV) are both designed specifically for this niche.

Sizing rule of thumb: For our 60 µW average target, plan for roughly 4–6 cm² of a-Si panel at 300 lux, assuming 70–80% MPPT efficiency. That gives you approximately 28–96 µW of harvested power depending on actual light conditions. Tight, but workable with a storage buffer.

DFM callout: Solar panel cost scales non-linearly. Standard rectangular panels are stocked; custom shapes (round, L-shaped to fit an enclosure) carry $2K–$10K in NRE tooling. Make your enclosure accommodate a standard panel, not the other way around.

RF Energy Harvesting: A Supplement, Not a Silver Bullet

RF energy harvesting sounds magical: power from thin air. The reality is more constrained. Ambient RF from WiFi access points, BLE devices, or cellular base stations typically provides 1–10 µW at realistic distances (3–5 meters). That’s useful, but it won’t run your device alone.

The basic architecture is a rectenna: an antenna impedance-matched to a rectifier circuit that converts RF energy to DC. The critical spec is sensitivity threshold, the minimum input power at which the rectifier produces usable DC. A practical floor is around -20 dBm (10 µW), which limits your effective range from any given RF source.

Dedicated RF power beacons operating at 915 MHz or 2.4 GHz can improve this dramatically, delivering 100+ µW at short range. But if you’re deploying beacons, you’re adding infrastructure cost.

The real value of RF harvesting for battery-free IoT is as a complementary source alongside solar. When lights are off (nights, weekends, unoccupied spaces) ambient RF may still trickle in from always-on access points. The e-peas AEM40940 is a dual-source PMIC designed exactly for this: it manages both solar and RF inputs simultaneously with independent MPPT on each channel.

DFM callout: Rectenna tuning is highly sensitive to enclosure material and geometry. ABS and polycarbonate have different dielectric properties and will detune your antenna by 1–2 dB. Test RF performance in the final housing early in your design cycle, not after tooling.

The PMIC and Storage: Where Your Design Succeeds or Fails

The power management IC is the heart of any energy harvesting embedded system. It performs four functions: maximum power point tracking (MPPT), cold-start management, voltage regulation, and energy storage management.

Harvester → [PMIC w/ MPPT] → [Energy Storage] → [LDO/Buck] → [MCU + Radio]
               ↑                    ↑
          Cold-start V         Supercap vs.
          MPPT ratio           thin-film battery

Cold-start voltage is the gating specification. This is the minimum voltage and power at which the PMIC can boot from a completely depleted state, with no stored energy and no backup. If your solar panel’s open-circuit voltage at low lux dips below this threshold, your device is a brick.

Concrete comparison: the TI BQ25570 cold-starts at 330 mV with 15 µW minimum input. The e-peas AEM10941 cold-starts at 380 mV but needs only 3 µW. That 5× difference in minimum power can determine whether your device boots in a dimly lit corridor or not.

For energy storage, you’re choosing between supercapacitors and thin-film batteries:

  • Supercapacitors offer essentially unlimited charge cycles, low ESR, and simple charging, but suffer from leakage current (especially at elevated temperatures) and lower energy density.
  • Thin-film batteries (e.g., Ilika Stereax) pack more energy per volume but have limited recharge cycles (thousands, not millions) and higher unit cost.

Size your storage to buffer enough energy for N transmit cycles during periods with no harvesting (nights, covered panels). The formula:

C = (N × E_cycle) / ((V_max² - V_min²) / 2)

For 100 cycles of 3,454 µJ each, with a supercap charged to 3.3 V and a brownout at 1.8 V: C = (100 × 3,454 µJ) / ((3.3² - 1.8²) / 2) ≈ 91 mF. A 100 µF supercap gets you there. For deeper dives, see our guide on capacitor sizing for embedded systems.

DFM callout: Supercap leakage at 85°C can be 5–10× the 25°C datasheet value. Specify leakage at your maximum operating temperature. Also confirm your chosen supercap is reflow-safe. Many aren’t, which forces a separate hand-solder step in production.

MCU and Radio Selection for Intermittent Power

Your batteryless sensor will lose power. Not might. Will. This changes what you need from an MCU.

Key selection criteria: shutdown current below 500 nA, wake time under 5 µs, and, ideally, non-volatile memory for state retention across power outages. FRAM is preferred here because it lets you checkpoint state without the wear and energy cost of flash writes.

Strong candidates for energy harvesting IoT designs:

  • TI MSP430FR series: 350 nA standby, integrated FRAM, excellent ultra-low-power ecosystem.
  • Ambiq Apollo4 Lite: Sub-µA sleep, subthreshold power processing.
  • Nordic nRF52833: 1.7 µA sleep with RTC, integrated BLE 5.3 radio. A good fit when BLE is your transport.

See our ultra-low-power MCU selection guide for a detailed comparison.

Firmware architecture must be event-driven, not a polling main loop. Design for checkpoint/restore: save critical state to FRAM on every cycle so a power loss mid-operation doesn’t corrupt your sensor readings or BLE stack state. This “intermittent computing” pattern (well-studied in academic research by Lucia et al.) is now practically necessary in production zero-power IoT devices.

Reference Design: Dual-Source Indoor Sensor Node

Here’s the complete system for an indoor temperature/humidity tag with BLE advertising, pulling together everything above:

 [a-Si Panel 4cm²]──┐
                     ├──[AEM10941 PMIC]──[100µF Supercap]──[1.8V LDO]──[nRF52833]──[SHT40]
 [2.4GHz Rectenna]───┘         │                                            │
                          [MPPT + Cold                                [BLE Adv.]
                           Start Mgmt]

Key specs:

  • Harvested power: ~40–80 µW (solar, 300 lux) + ~2–5 µW (ambient RF)
  • Sleep current: ~1.8 µA (nRF52833 + SHT40 idle)
  • TX interval: 60 seconds (adjustable based on available energy)
  • Storage autonomy: ~100 cycles in darkness ≈ ~1.7 hours with 100 µF supercap
  • Cold-start: functional down to 50 lux with AEM10941’s 3 µW threshold

Validation checklist before calling this done:

  1. Profile current across the full duty cycle with PPK2; verify the energy budget matches predictions within 20%.
  2. Test cold-start from a fully depleted supercap under minimum expected light.
  3. Run a 72-hour dark-room soak test. Confirm the device recovers and resumes operation when light returns.
  4. Verify BLE advertising optimization is tuned for minimum TX energy.

DFM Tradeoffs That Will Dominate Your Production BOM

The bench prototype works. The hard part is making it manufacturable at volume. Use our DFM checklist for IoT hardware as a companion to this section.

Solar panel: Conformal coating or a protective window over the panel will reduce light transmission 5–15%. Account for this in your energy budget. Standard rectangular panels from Panasonic or Epishine drop significantly in cost at 10K+ volumes.

Antenna/Rectenna: Plan for 1–2 dB enclosure loss. If your RF margin was already thin, this kills your supplementary harvest. Prototype in the production housing.

Supercap: Specify leakage at your maximum operating temperature in the production spec. A supercap that looks great at 25°C may hemorrhage charge at 60°C in a sunlit warehouse.

Assembly: Reflow-safe supercaps exist (e.g., Murata DMF series) but limit your chemistry choices. Thin-film batteries typically cannot survive reflow and require a separate assembly step. Factor this into your line cost.

Production test: You must verify cold-start on every unit. Define a go/no-go threshold: illuminate the panel at a standardized lux level and confirm the device boots within a specified time.

+-----------------------+----------+-------------+
| Component             | Proto $  | @10K Vol $  |
+-----------------------+----------+-------------+
| a-Si solar panel      | $3.50    | $1.10       |
| PMIC (AEM10941)       | $2.80    | $1.60       |
| Supercap 100µF        | $0.90    | $0.35       |
| MCU (nRF52833)        | $2.50    | $1.80       |
| Rectenna components   | $1.20    | $0.45       |
+-----------------------+----------+-------------+
| Harvest subsystem tot | ~$10.90  | ~$5.30      |
+-----------------------+----------+-------------+

At $5.30 for the harvest subsystem at volume, you’re well under the lifetime cost of a single battery replacement visit. That’s the business case.

Turning This Into Your First Build

The design flow is sequential, and each step gates the next: energy budget → source selection → PMIC + storage → ultra-low-power compute → firmware architecture → DFM. Skip the energy audit and you’ll discover in testing that your device can’t boot in the actual deployment environment.

Start by measuring your current application’s real power profile with a PPK2. Build the energy budget table. Then check whether indoor solar at your deployment’s worst-case lux can cover it. If the math works, you have a buildable battery-free IoT product. If it doesn’t, you know exactly which knob to turn: duty cycle, TX power, or panel area.

The full Prototype to Production series covers adjacent topics, from MCU selection to BLE optimization to production DFM, that connect directly to the design choices outlined here.

Hubble Network connects battery-free IoT devices directly to satellite, eliminating ground infrastructure as a constraint on where you deploy. See how it works →