Energy Harvesting

Converts light energy, typically from the sun or indoor lighting, into electricity using semiconductor materials like silicon or thin-film cells. It is the most mature and widely deployed harvesting technology.

• Key Mechanism: The photovoltaic effect, where photons with sufficient energy create electron-hole pairs in a semiconductor, generating a voltage.
• Power Density: Highly variable. Outdoor sunlight can deliver 10-100 mW/cm², while indoor lighting yields only 10-100 µW/cm².
• Primary Use Cases: Powering outdoor sensors, calculators, and, at larger scales, grid-tied solar panels. For edge AI, it's suitable for devices with access to consistent light.

Generates electrical charge from applied mechanical stress or vibration using materials like PZT (lead zirconate titanate) or PVDF. The charge is produced when the material's crystalline structure is deformed.

• Key Mechanism: The direct piezoelectric effect, where mechanical strain causes a displacement of ions, creating a dipole moment and surface charge.
• Power Density: Typically in the range of 10-100 µW/cm³, highly dependent on vibration frequency and amplitude.
• Primary Use Cases: Harvesting energy from industrial machinery vibrations, foot traffic, or structural strain. Ideal for wake-on-inference sensors in manufacturing or infrastructure monitoring.

Converts temperature differences (thermal gradients) directly into electrical voltage using the Seebeck effect. A thermoelectric generator (TEG) is made from pairs of p-type and n-type semiconductors.

• Key Mechanism: The Seebeck effect, where charge carriers diffuse from the hot side to the cold side of a material, creating a voltage proportional to the temperature difference (ΔT).
• Power Density: Generally low, around 10-100 µW/cm² for modest ΔT (e.g., 5-10°C). Scales with the square of ΔT.
• Primary Use Cases: Scavenging waste heat from engines, industrial processes, or the human body. Enables energy-autonomous sensors for predictive maintenance in heated environments.

Captures ambient electromagnetic energy from radio waves (e.g., TV broadcasts, cellular signals, Wi-Fi) and rectifies it into DC power. Efficiency depends heavily on distance from the transmitter and frequency.

• Key Mechanism: An antenna captures RF waves, and a rectifying circuit (a rectenna) converts the alternating current to direct current.
• Power Density: Extremely low, often in the nanowatt to microwatt range at practical distances from sources.
• Primary Use Cases: Powering ultra-low-power devices like RFID tags, wireless sensor nodes in smart buildings, or intermittent computing systems that can operate on tiny, sporadic energy bursts.

Generates electricity from varying magnetic fields or motion using the principle of electromagnetic induction (Faraday's Law). A changing magnetic flux through a coil induces a voltage.

• Key Mechanism: Faraday's Law of Induction. Can be implemented via linear motion (shaking a magnet through a coil) or rotational motion (miniature generators).
• Power Density: Varies widely; human-motion devices (e.g., shake flashlights) produce milliwatts intermittently.
• Primary Use Cases: Kinetic energy harvesters in wearable devices, self-powered switches, or energy from rotating machinery. Supports duty-cycled operation for activity-tracking edge AI.

Critical ancillary systems that condition, store, and manage harvested energy, which is often irregular and low-power. This subsystem is essential for reliable device operation.

• Key Components:
  • Power Management IC (PMIC): Rectifies, boosts, and regulates the erratic harvested voltage to a stable level usable by electronics.
  • Energy Storage: A rechargeable battery or, more commonly for harvester-only systems, a supercapacitor that can handle frequent charge/discharge cycles.
  • Intermittent Computing Runtime: Software/hardware layer that enables computation to survive power failures by checkpointing state to non-volatile memory.
• Primary Role: Bridges the gap between the unpredictable energy source and the stable power rail required for on-device model compression and inference, enabling true energy autonomy.

Vibration, thermal, or acoustic energy harvesters power tinyML models that analyze equipment signatures directly on industrial machinery. These models perform anomaly detection and predict failures, transmitting alerts only when necessary via low-power radio. This eliminates wiring and battery replacement in hard-to-access locations, enabling condition-based maintenance in factories, wind turbines, and pipelines.

Embedded in bridges, buildings, or aircraft, these systems harvest energy from ambient vibration or thermal gradients. They run compressed neural networks to process data from strain gauges and accelerometers, detecting cracks, corrosion, or stress anomalies. The system operates for decades without maintenance, providing continuous safety monitoring and enabling preventive interventions based on real-time structural data.

Solar or RF energy harvesters power networks of sensors that perform on-device analytics for agriculture, conservation, and climate science. Use cases include:

• Crop monitoring: Analyzing spectral data for disease detection.
• Wildlife tracking: Running lightweight audio models for species identification.
• Air/water quality: Detecting pollutant levels and chemical signatures. Data is processed locally, with only summary insights transmitted, drastically reducing communication energy.

Harvesting energy from body heat, motion, or light enables medical devices that never need charging. These devices run specialized health inference models for:

• Continuous glucose monitoring and predictive alerts.
• Cardiac arrhythmia detection from ECG signals.
• Neurological disorder monitoring (e.g., seizure prediction). The ultra-low milliwatt budget requires extreme model compression (binary neural networks, pruning) to enable real-time, life-critical analytics on harvested power.

Self-powered sensors placed throughout a building harvest energy from indoor light or small temperature differentials. They run reinforcement learning agents or forecasting models to optimize energy use. Examples:

• Occupancy detection via infrared or acoustic sensing to control lighting/HVAC.
• Personalized thermal comfort models that adjust micro-environments.
• Predictive load balancing for district heating/cooling systems. This creates a fully autonomous, self-configuring network for maximizing building efficiency.

RFID-like tags enhanced with energy harvesting and a micro-controller can run simple computer vision or sensor fusion models. Powered by ambient RF or light, they enable:

• Visual inventory counting and product identification on shelves.
• Condition monitoring (e.g., detecting if a perishable good has been exposed to adverse temperatures).
• Smart logistics: Determining package orientation or detecting drops during shipping. This brings ambient intelligence to every item in a supply chain without manual scanning.

A strict power consumption constraint, typically in the range of single-digit to tens of milliwatts, imposed on an edge device or subsystem.

• Purpose: To ensure operation within the energy limits of a small battery or the continuous output of an energy harvester.
• Design Implication: Dictates every architectural choice, from the selection of the microcontroller and sensors to the complexity of the ML model that can be run.
• Example: A solar cell on a smart sensor might provide a 5 mW average budget, forcing the use of extreme model compression and aggressive duty cycling.

An event-driven system architecture that minimizes energy use by keeping the main, power-hungry processor asleep until absolutely needed.

• Mechanism: A tiny, always-on coprocessor (e.g., a microcontroller or microNPU) runs a minimal, ultra-efficient detection model (e.g., for keyword spotting or motion detection).
• Trigger: Only when this first-stage inference detects a relevant event does it activate the main AI accelerator for more complex processing.
• Energy Impact: Dramatically reduces average system power by avoiding continuous operation of the largest power consumer.

A fundamental power management strategy where a system alternates between short active periods and long sleep periods.

• Principle: Reduces average power consumption by limiting the time components are fully powered. Power_avg = (Active_Power * Active_Time + Sleep_Power * Sleep_Time) / Period
• Application in Harvesting: Critical for matching the device's power consumption profile to the harvester's often-trickle charge rate. The system wakes, performs a sensing/inference task, transmits data if necessary, and returns to deep sleep.
• Optimization: The ratio of active to sleep time is tuned based on available energy and task latency requirements.

A dedicated hardware block or integrated circuit responsible for the generation, regulation, sequencing, and control of power within a system.

• Critical Functions for Harvesting:
  • Maximum Power Point Tracking (MPPT): Dynamically adjusts the electrical load to extract the maximum possible power from a variable energy harvester (e.g., solar, piezoelectric).
  • Voltage Conversion: Steps up/down the harvested voltage to usable levels for system components.
  • Energy Storage Management: Safely charges/discharges a buffer element like a capacitor or thin-film battery.
  • Power Gating: Controls which subsystems receive power based on the operational state.
• System Role: The PMU is the essential interface between the chaotic energy source and the stable, reliable power rails required by digital compute and sensing elements.