Thinned and Welded Silver Nanowires for Intelligent Pressure and Humidity Sensing Enabled by Machine Learning

Abstract

The rapid development of flexible and wearable technologies urgently requires high-performance and compatible flexible electrodes. Silver nanowires (AgNWs) are considered a promising conductive material, but their inherent structural drawbacks significantly hinder their widespread application. Here, an innovative and facile strategy is employed to simultaneously enhance the electrical and optical properties of AgNWs by precisely tuning their microstructure. Thinner AgNWs are achieved through the selective etching, alongside enhanced heating and mechanical properties, stemming from the welded structure and preserved conductive network integrity. A pressure sensor constructed with modified AgNWs demonstrates improved sensitivity compared to one using pristine AgNWs. By leveraging machine learning, the sensor can identify pressing behaviors with different fingers, achieving a high accuracy of 94.5%. The letters of the alphabet are also accurately recognized through analysis of unique resistance patterns in their Morse code. With the incorporation of a moisture-sensitive graphene oxide layer, the device is capable of recognizing human respiration behaviors and detecting voices based on their distinctive respiration and pronunciation patterns. The strategy employed to tailor the functionality of AgNWs, combined with further integration of machine learning, presents a promising avenue for advancing flexible and wearable electronics.

Keywords: conductive network; machine learning; morse code; selective etching; sensing.

Figure 1

a) SEM images of pristine and HCl‐treated AgNWs. Scale bar: 500 nm. b) XRD patterns of pristine and HCl‐treated AgNWs. c) Average diameters of pristine and HCl‐treated AgNWs. d) Schematic of AgNWs after acid treatments. e) EDS element mapping of i, ii) pristine and iii, iv) HCl‐treated AgNWs. Scale bar: 500 nm. f) Transmittance of films based on pristine and HCl‐treated AgNWs (excluding the substrate). The inset photograph shows AgNWs electrodes treated with and without HCl. Scale bar: 2 cm. g) The corresponding sheet resistance and FOM of AgNWs films.

Figure 2

a,b) Tilted SEM images of pristine and treated AgNWs. Scale bar: 500 nm. c) Schematic of contacted AgNWs with low and high R_c. d,e) TEM images of pristine and treated AgNWs. Scale bar: 10 nm. f) AFM images of i) pristine AgNWs and ii) AgNWs‐40 h. The roughness spectra correspond to the line scan results.

Figure 3

a) Resistance evolution of AgNWs with varying HCl treatments under an electric field. b) Simulated electric potential distribution of i) separated and ii) contacted AgNWs. c) Heating performance of acid‐treated AgNWs (32 h). d) Simulated temperature distribution of i) pristine AgNWs and ii) acid‐treated AgNWs before failure. e) Diagram detailing the bending test parameters. Relative resistance evolution of AgNWs subjected to 10 000 bending cycles at bending radii of f) 10 mm, g) 5 mm, and h) 2.5 mm.

Figure 4

a) A photograph of AgNWs sensors. Scale bar: 1 cm. b) Relative resistance changes induced by different weights applied to the sensor fabricated from AgNWs treated with HCl for 32 h. c) The relative resistance changes of the pressure sensor with five different fingers. d) Radar plot representing the total accuracy of five fingers with various machine learning models. e) Confusion matrix after 80 training iterations. f) Alphabet classification showing the accuracy of distinguishing each letter after 10 sets of training iterations. The insets show the resistance changes representing the Morse codes of “UNSW” and “Morse”.

Figure 5

FTIR analysis of GO film under a) moisture exposure and b) dry N₂ blowing over time. c) Relative resistance changes of the sensor under different humidities controlled by wet N₂. Sensor response under d) normal breath, e) deep breath, and f) breath after exercise. g) Confusion matrix with 10 iterations of each breath mode. h) Resistance variation of the sensor triggered by different voices. i) Confusion matrix of voice classification.