Figure 1a provides a schematic diagram of the Pd nanowire H2 sensor employed in this study. The proposed Pd nanowire H2 sensor undergoes a change in electrical resistance when exposed to H2 at room temperature. This phenomenon arises as hydrogen (H) atoms are adsorbed and diffused within the Pd material, disrupting the flow of electrons (e−) (Fig. 1b-i, 1c-i). As a result, electron scattering is induced by the H atoms within Pd, which is proportional to H2 concentration, resulting in a change in electrical resistance (R). It is noteworthy that reducing the size of the Pd material improves surface adsorption and scattering, which significantly enhances sensing performance in terms of reaction time and resistance change (sensitivity). However, as previously mentioned, Pd nanostructures can be easily affected by common environmental contaminants (black solid particles in Fig. 1c-ii)27,28,29. Therefore, when exposed to ambient air conditions for an extended amount of time, the device can experience a significant performance degradation by the adsorbed and/or absorbed contaminants (Fig. 1b-ii, c-ii). To eliminate the adsorbed contaminants and recover their sensing performance, we propose a thermal refreshing method (Fig. 1c-iii). The applied thermal energy is capable of generating the necessary energy at the surface, leading to the controlled rupture of bonds with the adsorbed and/or absorbed contaminants. This process results in an efficient and full recovery of the Pd nanowire surface and sensing performance (Fig. 1b-iv, c-iv).
To perform an in-depth study regarding the Pd contamination and validate the proposed recovery concept, we first fabricated a Pd nanowire H2 sensor utilizing an emerging nanofabrication technique developed in our previous work (Supplementary Information Figure S1)24. Since the utilized process is based on a conventional semiconductor fabrication, a perfectly-aligned Pd nanowire array was fabricated in a highly reliable manner (Supplementary Information Figure S2). A photograph of a diced single-sensor device is shown in Fig. 1d. From a magnified optical microscope (OM), we confirmed that the Pd nanowires are formed on the specifically designed region between the two separated gold (Au) electrodes (Figs. 1d and 1e). Figure 1f also confirms the ohmic I-V curve, measured in ambient air. Then, from a scanning electron microscope (SEM) (upper panel of Fig. 1g) and transmission electron microscope (TEM) (lower panel of Fig. 1g), we also verified that the Pd (width = 160 nm, length = 100 µm, thickness = 25 nm) nanowires are perfectly separated and positioned, as designed (Supplementary Information Figure S3). The details of the fabrication processes are shown in the Method section.
Using the fabricated device, we first evaluated its chemo-electrical response. The fabricated sensor was located in a customized chamber, where the concentrations of H2 gas (H2 ≤ 4%) could be precisely controlled through a designed nozzle; the electrical resistance of the sensor was measured in real-time using a conventional digital multi-meter (USB-4065, National Instruments, USA), Figure S4. The measured chemo-electrical response of the as-fabricated Pd nanowire sensor is represented on the left panel of Fig. 2a. The fabricated sensor shows stable base resistance when exposed to nitrogen (N2) gas; however, a sudden change in resistance is detected as the H2 is injected into the chamber. The change in resistance in response to H2 also increases as the concentration increases from 1–4%. It is important to note that the changed resistance is stably maintained during the entire duration of H2 injection. However, after extended time, the performance of the chemo-electrical sensor is degraded. After storing the identical device in typical ambient air conditions for 7 days, chemo-electrical gas measurements were conducted (right panel in Fig. 2a). While the device showed increasing resistance in response to H2 injection, significant degradation in response, stability, and response time were observed. To quantitatively confirm the degraded performance of the sensor, we extracted the response time (\({\tau }_{res}\)) and the gas response (\({\varDelta R/R}_{0}\)) of the sensor with respect to H2 concentration (Fig. 2b). The initially fabricated device exhibited fast \({\tau }_{res}\) of less than 25 seconds for each concentration (1, 2, 3, and 4%), but after 7 days of storage, the device exhibited a prolonged \({\tau }_{res}\) of at least 40 seconds (left graph of Fig. 2b). The \({\varDelta R/R}_{0}\) of the sensor also decreased (right graph of Fig. 2b). Initially, the fabricated sensor showed 9.97, 15.51, 16.66, and 16.67% changes in resistance to 1, 2, 3, and 4% of H2, respectively, yet each response is degraded to 7.30, 10.57, 11.78, and 12.38%, respectively. Note that \({\tau }_{res}\) was calculated as the time for the sensor to reach 90% of the difference between the minimum and maximum values of each response, and \({\varDelta R/R}_{0}\) was calculated as the difference between the minimum and maximum values per minimum value.
Few studies have previously reported that the degraded performance of the Pd-based H2 sensor might be due to the carbon molecules in ambient air30,31. However, as to which specific contaminants are involved, how they function, and the recovery methods, to the best of our knowledge, have not been clearly identified. Thus, we have conducted in-depth studies to identify the origin of the degraded performance with diverse material analysis. For each material analysis, we fabricated a thin film Pd (50 nm-thick) on a silicon (Si) wafer and observed its change in grain structure, morphology, and chemical state using SEM, atomic force microscope (AFM), X-ray diffraction analysis (XRD), and X-ray photoelectron spectroscope (XPS) with respect to storage time. The sample preparation is shown in the Method section. Visual changes via SEM, AFM, or XRD results are negligible between the initial and contaminated states (Supplementary Information Figure S5); however, XPS analysis revealed significant differences (Fig. 2c, d). It should be noted that the ‘Contaminated state’ defines the Pd specimen after 7 days of storage. While the peaks that identify as pure Pd, carbon (C), and carbon-oxygen single bond (C-O) were found to be the same in both the initial and contaminated states, C = O was observed only in the contaminated state. This suggests that carbon dioxide (CO2) from the atmosphere is adsorbed onto the Pd surface, generating specific bonds with oxygen. The full spectrum and detailed XPS results of both specimens are shown in the Supplementary Information Figure S6.
To validate that the C = O adsorption on the Pd surface occurs naturally and inevitably in ambient air, density functional theory (DFT) was conducted. The key molecules selected for analysis include not only CO2, as revealed by the XPS results in Fig. 2d, but also H2, H2O, and CO, which are known to be the main concerns in H2 detection31,32. Figure 2e illustrates DFT model systems where these molecules are individually adsorb on the Pd (111) surface. The adsorption energy was determined by calculating the difference between the energy of the molecule attached to the Pd surface with the sum of the energies of the Pd’s clean surface and the isolated molecule. In Fig. 2f, when comparing the adsorption energies of H2, CO, and CO2, it can be observed that CO strongly adsorbs onto the surface of Pd with an adsorption energy of − 2.02 eV. These results are well corresponded with the XPS results. Due to the strong adsorption energy between the Pd surface and CO gas, CO readily adsorbs to the Pd surface naturally, resulting in the intrinsic C-O peak in the XPS result (Fig. 2c). However, despite the natural adsorption of CO to the Pd surface, Pd is still able to react with H2 gas due to the extremely low atmospheric partial pressure of CO. The minimal CO adsorption on the Pd surfaces in practical application is nearly negligible, and it does not significantly interfere with H2 sensing33. The second strongest adsorption can be seen from the H2. When H2 gas is adsorbed, it undergoes H-H dissociation and exhibits a strong adsorption energy of − 0.87 eV, thereby indicating Pd as a suitable H2 gas sensor. Lastly, the negative adsorption energy (–0.168 eV) can also be calculated from CO2, indicating a natural adsorption of CO2 on the Pd surface. Additionally, as opposed to CO, CO2’s atmospheric partial pressure is about 4000 times larger than that of CO (assuming standard atmospheric conditions at sea level)34. While the adsorption energy may be small, the abundant presence of CO2 poses a significant threat as a contaminant for long-term use, preventing H2 gas adsorption and degrading the H2 sensing performances.
To eliminate the C = O from the Pd surface and recover device performance, we propose a thermal treatment method. Previous studies have reported that the removal of CO2 on a metal surface while negligibly changing the physical and chemical properties of the material, such as grain-structure/-size variation and oxidation, occurs at a temperature range of 400 ~ 500 Kelvin35. To verify the proposed method, a DFT simulation was conducted. To simulate the thermal treatment process, ab-initio molecular dynamics (AIMD) calculations were conducted at various temperatures. In Fig. 3a, we compared the trajectory of carbon (C) atoms in CO and CO2 with respect to time at temperatures of 300 K and 500 K. At 300 K, CO and CO2 exhibit slight movement near the surface but do not move away from Pd. However, at 500 K, CO2 detaches from the surface, freely moving away from the Pd surface while CO is still strongly bounded to Pd. According to Fig. 3b, we can track the detachment from the Pd surface by monitoring the z-axis coordinates of C atoms. At 500 K, the carbon atom within the CO2 molecule exhibits significant fluctuations and detaches from the surface. This phenomenon can be explained by the weak adsorption strength of CO2, making it possible to remove CO2 impurities through high-temperature treatment. DFT simulations with other temperatures such as 400 K and 600 K were also conducted, showing similar tendencies with 300 K and 500 K results (Supplementary Figure S7). We also conducted the electron localization function (ELF) to analyze the bonding characteristics between CO and Pd. The C of the CO molecules shares electrons with the Pd surface, forming a very strong covalent bond (Fig. 3c). However, as previously mentioned, the C of CO2 does not form a strong bond with Pd at 300 K, resulting in CO2 evolution at a higher temperature (T = 500 K).
To experimentally verify the proposed thermal-refreshing technique, we thermally treated the specimen on a hot plate set to 200 ℃ in air condition for 10 min. The change in surface chemical bonding of the same sample from Fig. 2 was observed using XPS. From the data, a noticeable change in C bonding was present (Fig. 3d). For clear juxtaposition, we extracted the C 1s, C-O, and C = O peaks from the raw data of the contaminated and thermal treated specimen, and compared the change in peak-intensity (Fig. 3e). While the C-O peaks are maintained even after heat treatment, due to their strong adsorption energy, the intensity of C = O is significantly decreased, further corroborating our findings.
Additionally, H2 gas sensing performance was analyzed for initial, contaminated (1 day), and recovered states of the device. First, to find the optimal thermal refreshing parameters, we varied the annealing temperatures from 100 ℃ to 300 ℃, following the results of our DFT simulations, with a fixed time of 10 min. Then, the gas sensing performances to 1% H2 gas, including the \({{\tau }}_{res}\) (response-time) and \({\varDelta R/R}_{0}\) (gas-response), of the Pd-nanowire devices (n = 3) have been measured (Fig. 3f). All of the devices showed degraded sensing performance after 1 day of ambient storage, but the extent of recovery varied with thermal refreshing temperature. On average, devices annealed at 100 ℃ recovered its response time to be 20% slower (\({{\tau }}_{res}\)=120%) with 93% response recovery (\({\varDelta R/R}_{0}\)), indicating insufficient thermal treatment. However, devices treated at higher temperatures showed improved recovery characteristics with \({{\tau }}_{res}\) and \({\varDelta R/R}_{0}\) to 98% (2% faster) and 97%, respectively, at 200 ℃, and to 68% (32% faster) and 98%, respectively, at 300 ℃. It is interesting to note that the response time of the devices after annealing at 200 ℃ and 300°C showed faster response times when compared to their initial states, indicating an ‘over-recovery’. This phenomenon can be attributed to the widely-known releasing of the lattice defects which occur at high temperatures. As a result, for the focus of this research, we have chosen 200 ℃ as the optimal thermal refreshing temperature. Next, we evaluated the annealing time dependency of the device (Fig. 3g). The devices were treated to various annealing times of 5, 10, and 15 min at an annealing temperature of 200 ℃ and exposed to 1% H2 gas. Similar to previous measurements, the device annealed at 5 min showed longer \({{\tau }}_{res}\) and lower \({\varDelta R/R}_{0}\), while the 10-min condition resulted in complete recovery of \({{\tau }}_{res}\) (100%) and 93% recovery of \({\varDelta R/R}_{0}\). Annealing time of 15 min showed ‘over-recovery’ due to the similar phenomenon as mentioned above. Therefore, for near-100% performance recovery, we chose the optimal thermal refreshing condition as 200℃ for 10 min. Raw data of Fig. 3f are shown in Supplementary Figure S8.
Based on the optimized thermal treatment condition (200 ℃ and 10 min), we also tested the device’s recovery characteristics to a wide range of H2 concentrations. Three states of the device were measured and compared under 1 to 4% of H2 concentrations to validate our proposed concept: as-fabricated (‘Initial’), after ‘1-day’ in ambient air storage, and post-thermal treatment on a 200 ℃ hot plate for 10 min (‘Annealed’). The extracted \({\tau }_{res}\) and \({\varDelta R/R}_{0}\) are shown in Fig. 3h. Considering the initial \({\tau }_{res}\) as 100%, the response times of some devices increased by as much as two-folds (\({\tau }_{res}\)= 200%) at higher H2 concentrations. However, with optimal thermal refreshing conditions, the sensor showed near-perfect recovery for all H2 concentrations. In terms of gas response, the sensors showed about 20% degraded \({\varDelta R/R}_{0}\), but near-perfectly recovered to their initial states ( > 90%) following thermal refreshing. Raw data are shown in Supplementary Figure S9.
After the thermal refreshing process, we also confirmed that the granular structure of Pd is not affected through surface morphology using AFM (Fig. 3i). The roughness of Pd after thermal treatment (Ra) is about 0.284 nm, showing negligible change in roughness from its initial measurements, Ra=0.275 nm. As a result, we can conclude that the proposed recovery technique can easily and successfully recover the chemo-mechanical device’s performance against H2 gas by selectively removing the contaminants, such as C = O bonds, on the Pd surface without any side effects.
To practically adapt the proposed concept, we developed an all-electrical wireless H2 gas sensor system. In this concept, the thermal treatment on the Pd nanowire can be conducted by joule heating from a controllable power source unit. In detail, through a circuit using a Wheatstone bridge and operational amplifier (OPAMP), the change in the sensor’s resistance is converted into a change in voltage. Data is transmitted and received by an analog-digital-converter (ADC) in Bluetooth low energy (BLE) integrated system on chip (SoC) for wireless applications. The heating circuit for joule heating was designed to allow a specific voltage to flow through the sensor using a 5 V regulator containing an ON/OFF switch and a SoC's GPIO. Additionally, to monitor the ambient conditions of the sensor, a commercial temperature and humidity sensor was also integrated onto our developed wireless module (Fig. 4a, b). The details regarding the sensor hardware are in the Method section and Supplementary Figure S10. Using the manufactured hardware, we were able to measure H2 concentrations of 1 to 4% (Fig. 4c) in a wireless manner. Then, to ensure that the joule heating method can mimic the proposed thermal refreshing technique, we compared the simulated and measured data. Simulation results, using the finite element method (FEM), indicated that 4.4 V and 406 mA of power will be needed for the Pd nanowires to reach 200 ℃ (Fig. 4d and Supplementary Information Table S1). Subsequently using a thermal imaging (infrared) camera, we experimentally confirmed that the Pd nanowires can be effectively heated to the desired temperature with coinciding values from our simulation results (Fig. 4e). FEM simulation details are shown in the Method section and Supplementary Figure S11. To further validate the functionality of the developed wireless sensor module even to low gas concentrations, we measured the change in ADC of the sensor for H2 gas concentrations ranging from 0.1–1.0%. The as-fabricated H2 gas sensor module successfully shows stable electrical output with respect to varying H2 gas. However, consistent with previous findings, a significant degradation in the chemo-electrical characteristics of the sensor was observed after a day of storage in ambient air (Fig. 4f). By applying joule heating within the wireless module for 10 min at 200°C, we successfully recovered the sensor’s performance to its initial state (Fig. 4g). To ensure impartial comparison, we simultaneously measured the relative humidity (RH) of the sensor, thereby excluding its effects on device performance (upper panel in Fig. 4f and 4g). We quantitatively extracted that the \({\tau }_{res}\) and \({\varDelta R/R}_{0}\) degraded by an average of up to 57% (\({\tau }_{res}\)=157%) and 83%, respectively. Nevertheless, the device demonstrated near-perfect recovery to an average of 100% after thermal refreshing through joule heating (Fig. 4h).
Finally, we verified the sensor’s repeated performance. Figure 4i presents a 3D graph of the chemo-electrical response of the module to various H2 gas concentrations (0.1 ~ 1.0%). Repeated gas measurements were conducted after each ‘1-day’ storage and thermal refreshing cycle. Even when the sensor displayed degraded performance after each ‘1-day’ storage, the device showed near-perfect recovery to its initial performance even after repeated contamination and thermal refreshing cycles (inset in Fig. 4i). To quantify the differences in these contaminated and recovery state graphs, we used the concept of dynamic time warping (DTW). DTW is an algorithm for determining the similarity between two-time series data. When comparing the DTW values of the three different states, we found that the recovered and initial values are identical, while the contaminated state showed more than a threefold increase (Supplementary Figure S12). Lastly, extended contamination test was conducted to verify its long-term capability. Shown in Fig. 4j, the sensor’s response and response time are both heavily degraded after 61 days of contamination. However, with the proposed recovery method, the device returned 100% to its initial performance even after long-term contamination (raw data in Supplementary Figure S13). While ‘over-recovery’ was observed in terms of normalized gas response, it is important to note that the proposed concept holds true even with long-term contamination. In conclusion, the wireless sensor module exhibiting self-healing capabilities and its ability to sustain repetitive operations as well as long-term utilization, the practicality and its applicability of our proposed system and thermal refreshing technique is verified.