Broadband Thermoelectric Energy Harvesting for Wearable Biosensors Using Plasmonic Field-Enhancement and Machine-Learning-Guided Device Optimization

TL;DR

This paper presents a hybrid thermoplasmonic and thermoelectric energy harvester with machine-learning optimization, significantly enhancing power output for wearable biosensors by capturing a broad infrared spectrum and increasing temperature differences.

Contribution

It introduces a novel broadband metasurface with plasmonic nanoantennas combined with machine-learning-guided design to improve energy harvesting efficiency in wearables.

Findings

Achieved a four- to six-fold increase in power density over existing devices.

Predicted device performance with high accuracy using a surrogate model.

Identified optimal device geometries through Pareto-front analysis.

Abstract

Wearable biosensors increasingly require continuous and battery-free power sources, but conventional skin-mounted thermoelectric generators are limited by the small temperature differences available in real environments. This work introduces a hybrid thermoplasmonic and thermoelectric energy harvester that combines multiband plasmonic absorption with machine-learning-guided optimization to improve on-body energy conversion. A broadband metasurface made of cross-bowtie nanoantennas is designed to absorb infrared radiation across the 2 to 12 micron range, capturing human body emission, ambient infrared radiation, and near-infrared sunlight. Electromagnetic simulations show strong field enhancement in nanoscale antenna gaps, producing localized thermoplasmonic heating directly above flexible Bi2Te3 thermoelectric junctions. Coupled optical, thermal, and electrical modeling indicates that this localized heating increases the effective temperature difference from the typical 3 to 4 degrees C of standard wearable thermoelectric generators to approximately 13 degrees C. This results in a power density of about 0.15 mW per cm^2 under indoor-relevant infrared flux, representing a four- to six-fold improvement over existing flexible devices. A machine-learning surrogate model trained on multiphysics data predicts temperature rise and electrical output with high accuracy (R2 greater than 0.92) and identifies optimal device geometries through Pareto-front analysis. The proposed hybrid thermoplasmonic, thermoelectric, and machine-learning framework provides a scalable route toward more efficient, compact, and flexible energy harvesters for autonomous and long-term wearable physiological monitoring.