A comprehensive review of hydrogen sensor for thermal runaway monitoring: fundamentals, recent advancements, and challenges

Abstract

Thermal runaway (TR) in lithium-ion batteries (LIBs) remains an intrinsic safety issue, posing significant risks of fire and explosion. Among various technologies employed to assess LIB status- including temperature, pressure, voltage, and gas measurements-gas sensors exhibit superior response speed and stronger sensing abilities. Notably, H₂ has been identified as the first gas released during the TR process when compared to other gases such as CO₂, CO and CH₄. Furthermore, H₂ serves as an indicator for the formation of trace Li dendrites, which are inducements of LIBs safety issues. Consequently, development of high performance H₂ sensors is essential for providing timely early safety warning. Compared with other types of H₂ sensors, chemiresistive H₂ sensors have garnered significant attention owing to their good sensitivity, low cost, and easy of miniaturization and integration into LIB cells. This review presents a comprehensive overview of chemiresistive H₂ sensors through classifying them into different categories based on sensing material systems. Within each category, the inherent fundamental sensing mechanisms and current strategies aimed at enhancing sensor performance have been systematically discussed. It is believed that chemiresistive H₂ sensors would play an important role in TR monitoring. Moreover, a more accuracy prediction could be implemented when H₂ sensors are integrated with other existing warning methods.

Introduction

Global environmental challenges and the rapid growth of portable electronics and electric vehicles have heightened the demand for cleaner, more efficient energy conversion technologies and high-density lithium-ion batteries (LIBs) with sustainable electrochemical performance. LIBs typically consist of cathode, anode, and electrolyte. To enhance the energy density of LIBs, researchers have made significant efforts to develop novel cathode and anode materials^1,2,3. However, the LIBs with higher energy density may reduce thermal stability and cause safety problems, e.g., thermal runaway (TR). Owing to the irreversible increase of heat caused by TR, accidents relating to battery fire or even explosion still occur frequently^4,5,6,7. During the past few years, lots of electric products were recalled ascribing to the unsafe batteries⁸. According to the existed data, more than half of the fires are caused by electric vehicle battery failure⁹. Therefore, development of advanced technologies for providing early warning of TR are essential to minimize the safety-related problems in practical LIB applications.

The TR is generally defined as the condition in which the rate of heat generation exceeds the rate of thermal dissipation, leading to a significantly increase in LIB temperature. A commonly accepted index for TR is when the rate of battery temperature rise surpasses 1 °C/s^10,11. During the practical applications, LIBs may be subjected to various abuse conditions that can lead to TR, including mechanical abuse (e.g., collision and crush), electrical abuse (e.g., external short circuit, overcharge, and overdischarge), and thermal abuse (e.g., overheating)^12,13. Mechanical abuse may cause the deformation of LIBs, which can result in contact of cathode and anode and generate internal short circuit. Electrical abuse is usually caused by abnormal operation conditions, such as overcharge and overdischarge, leading to dendrite growth and internal short circuit. Thermal abuse will boost side reactions in the LIBs to generation of numerous heat and gases, such as decomposition of electrolyte. The TR process typically undergoes three stages as illustrated in Fig. 1a. At initial stage, the LIBs transition from a normal operational state to an abnormal one, such as overcharging, overheating, internal short circuits, etc., resulting in an increase of LIB temperatures. During this stage, the self-heating rate remains relatively low (0.2 °C/min), and sometimes can be dissipated within the LIB packs. While an increase in temperature may trigger sustained exothermic reactions that further elevate the temperature, the LIBs will go to stage 2. Once the stage 2 commences, the LIBs undergo a series of complex reactions, including the decomposition of solid electrolyte interphase (SEI) and cathode materials, as well as the melting of polymer separators. These reactions can accelerate the heat accumulation and gas release processes. At this stage, it is hardly to quench the TR process via any external cooling mechanism. The aforementioned exothermic reactions will lead to a dramatic increase in temperature, pressure, gas production, etc., and then the LIBs proceed to stage 3. As stage 3 starts, irreversible damage may occur within the LIBs, including risks of combustion and explosion. Furthermore, adjacent LIBs could be destroyed, potentially leading to a disaster.

Fig. 1: The stages of TR process and the changes of relevant parameters during TR process.

a Three stages for the TR process. Stage 1: The onset of overheating. The batteries change from a normal to an abnormal state, and the internal temperature starts to increase. Stage 2: Heat accumulation and gas release process. The internal temperature quickly rises, and the battery undergoes exothermal reactions. Stage 3: Combustion and explosion. The flammable electrolyte combusts, leading to fires and even explosions⁴⁹⁶. Copyright 2018, American Association for the Advancement of Science. b Readings of the seven sensors in thermal runaway test. The dashed line labeled with “visible venting” indicates the time when venting was visible outside of the battery housing. All sensors detect the thermal runaway effects of the cell within a time window of about 20 s³². Copyright 2018, Multidisciplinary Digital Publishing Institute

Therefore, to ensure the safety of human life and personality, high-precision technologies are essential for providing an early warning of the TR in individual LIB cell. Such advancements could afford people a couple of few minutes to stay away from potential hazards. Currently, several safety warning systems for LIBs are available, primarily focusing on the detection of electrical, thermal, mechanical signals, and gas emissions. Although the end voltage of LIBs is a widely recognized electrical parameter for evaluating the TR, the process of voltage drop exhibits variability when encountered different TR trigger factors^14,15. Moreover, an obvious change in voltage often occurs when the TR process evolves into an irretrievable state^16,17. Thus, relying solely on battery voltage detection for early warning of TR is unreliable. In addition, temperature is a typical thermal indicator for judging whether TR has occurred within the LIBs. Currently, commonly used temperature sensors primarily monitor the surface temperature and usually fail to supply adequate and in-time information due to the significant disparity between the internal and external temperatures of LIBs^18,19. Although there are several types of temperature sensors (e.g., thermocouples^20,21,22, and thermal resistance^23,24, optical fibers¹⁸) capable of detecting the internal temperatures, they often come with high costs and technical complexities^25,26,27. Since TR-induced elevated temperatures and gas release can cause pouch cell swelling and deformation, detection of the mechanical signal such as pressure is more appropriate than temperature for serving as an early safety indicator^28,29. However, the pressure sensors fail to trigger timely alerts under pressure abuse conditions²⁸. Notably, during the stage 1, a large amount of gas is generated from electrochemical reactions inside the cell that allows gas sensors to be an alternative technology for early warning alert. Additionally, during the TR process, two distinct venting events can be observed, with the second event signifying battery failure. Consequently, the time gap between the two venting events provides a window for early warning³⁰. Wenger et al.³¹ carried out a series of overcharging experiments by varying the charge rate of current (e.g., 5A(1C), 30A(6C), and 60A (12C)) and found that the gas sensor detected the failure earlier than sensors based on temperature and voltage measurements. Koch et al.³² selected a set of sensors to assess their capability for fast and reliable early warning by detecting various signals, including voltage, temperature, pressure, smoke, creep distance, force, and gas (as shown in Fig. 1b). Among these sensors, the SnO₂-based gas sensor, which is sensitive to CO and CH₂, provided the fastest detection speed and strongest sensing response³². Huang et al.³⁰ built a simulation model demonstrating that the gas sensor responds faster than the external temperature and pressure sensors. Consequently, gas sensors are capable of accurately monitoring released gases for practical early safety warnings.

The TR process of LIBs involves the decomposition of SEI and cathode materials, melting of polymer separators, and formation of Li dentaries. This process releases a lot of gases such as H₂, O₂, CO₂, CO, C_xH_x (CH₂, C₂H₄, etc.), HF and some organic vapor^33,34,35. Golubkov et al.³⁶ investigated the components of venting gases in LIBs with different cathode materials, revealing that CO₂, CO and H₂ are three main gas components. Additionally, a small amount of HF can only be detected when the LIBs contain specific fluorine-containing materials³⁷. To date, many researchers have analyzed the composition and released time of venting gases to identify typical gas indicator as early warning signs of TR through precise instruments such as differential electrochemical mass spectrometry³⁸, gas chromatograph (GC)³⁹, on-line electrochemical mass spectrometry (OMES)⁴⁰ and Raman^41,42. Yang et al.³⁹ selected eight types of commercial LiFePO₄ LIBs to analyze the gas components using GC under overcharge abuse conditions. The results reveal that H₂ and CO₂ account for the highest proportions among five main gases, as shown in Fig. 2a. However, the accuracy of CO₂ detection in practical applications may be impacted by environmental factors. Thereby, Yang et al.³⁹ chose H₂, CO, and CH₂ as gas signals of overcharging to determine which one is the best for early warning, as shown in Fig. 2b, c. Obviously, H₂ was detected first, 204 s earlier than CO and 619s earlier than CH₂, and significantly earlier (579 s) than TR³⁹. Zhang et al.⁴⁰ conducted operando OEMS characterizations under practical operation conditions, including overcharging, high temperature, and cycles of ageing/storage (Fig. 2d, e). Similarly, the release of H₂ is detected significantly earlier than that of CO₂ and CO under both normal charge and overcharge conditions at 25 °C (Fig. 2d). At high-temperature, the evolution of H₂ is much more distinguishable compared with CO and CO₂ under normal charge cycles and H₂ is the first detected gas under overcharge state (Fig. 2e). Figure 2f-h provide a micro-scale perspective on why detection of H₂ is essential for safety early warning because the release of H₂ gas can serve as an indicator for identifying the formation of trace Li dendrites, which are inducements of LIBs safety issues⁴³. For the sake of safety warning, United States Department of Energy (US DOE) set a goal for the advancement of H₂ sensor in the applications of TR monitoring, hydrogen energy transportation and storage, including achieving the response/recovery time (t_res/t_rec) within 1 s for H₂ ≥ 1%, a broad detection concentration range of 0.1–10%, and a lifespan exceeding 10 years⁴⁴. Therefore, development of H₂ sensors with fast response speed, high responsibility and stability is essential in real applications.

Fig. 2: The index gases released during TR process.

a Histogram of proportion of main gas components according to the results of gas chromatography, which is obtained based on GC results of ref. ³⁹; b, c Overcharging test of battery 1 with real-time monitoring of three gases in the chamber³⁹. Copyright 2023, Elsevier. The galvanostatic time–voltage profiles of stepwise increasing cut-off voltages (4.4–4.8 V) and real-time gassing behavior under different storage temperatures d RT, e high temperature, respectively⁴⁰. Copyright 2023, Elsevier. f H₂ gas was captured at 683 and 472 s for the two kinds of assembled LIBs through automatic GC detecting and relative microscopy images of graphite anode surface during the charging process; g, h Overcharge experiment of a LiFePO₄ battery pack with online detection of six gases (t₁ represents the initial time of overcharge; t₂ represents the time when the gas sensor detects the characteristic gas; t₃ represents the time when white smoke is observed; t₄ represents the time when fire explosion occurs)⁴³. Copyright 2020, Elsevier

To date, various H₂ sensors have been briefly reported in the realm of H₂ sensing, including chemiresistive, electrochemical, optical, conductometric and catalytic-combustion H₂ sensors. As summarized in Table 1, electrochemical and optical sensors, while capable of sensitive and selective detection of H₂, suffer from high fabrication cost and performance degradation due to their sensitivity to ambient condition fluctuations^45,46. Catalytic sensors exhibit poor selectivity and pose safety risks in LIB applications due to their flameless operation⁴⁷. Thermal conductive sensors lack the accuracy and sensitivity required for early warning detection⁴⁸. Compared with other types of gas sensors, chemiresistive gas sensor offer more superiorities such as high sensitivity, acceptable stability and low cost of fabrication. Moreover, due to their ease of integration and compatibility with Micro-Electro-Mechanical-Systems (MEMS) techniques^37,49, miniaturized chemiresistive gas sensors have already demonstrated the potential to monitor released gases during TR process^32,50.

Table 1 The mechanism and characteristics of different types of hydrogen sensors

Therefore, this review primarily focuses on presenting a comprehensive overview of chemiresistive H₂-sensing mechanisms, research advances, and remaining challenges in H₂ sensors. first, the authors elucidate the essentiality of developing chemiresistive H₂ sensor with high performance (e.g., high sensitivity, sub-second t_res/t_rec, high stability, humidity tolerance, good selectivity, etc.) for the sake of safety issues of LIBs. Although the importance of hydrogen monitoring in new energy systems, storage, and transportation is well-established, its critical role in battery TR warning requires more detailed discussion. This review provides a distinctive focus on systematically analyzing why H₂ serves as one of the most essential gaseous markers for early TR detection in LIBs. Different to prior review articles which are merely focusing on few kinds of materials (e.g., Pd based H₂ sensors⁵¹, 2 dimensional (2D) material H₂ sensor⁵², or nanogap determined H₂ sensors⁵³), this review provides a more comprehensive discussion of chemiresistive H₂ sensors based on various sensing materials, which are divided into the following categories metal-, metal oxide-, carbon-, transition metal dichalcogenide-, and Ⅲ–Ⅴ wide bandgap semiconductor-based H₂ sensors. Moreover, this review not only presents the advancements in enhancing H₂ sensor performance with respect to material and device design, but also emphasizes significant breakthroughs aligned with the standards set by US DOE for LIB safety. We believe this review will provide a forward-looking perspective on development H₂ sensors to meet US DOE’s critical metrics.

Metal-based chemiresistive H₂ sensor

Among various metals, Pd is the most widely utilized metal for H₂ sensing and storage as it can reversibly absorb a volume of H₂ around 1000 times than its own^{54,55,56,57,58}, and has been reported to have a higher binding energy with H₂ compared with Pt, Au, Ru, etc^59,60. Thus, in this review, we mainly focus on Pd-based and Pd functionalized H₂ sensors.

At room-temperature (RT), Pd reacts with H₂ molecules to form PdH_x hybrids, resulting in a significant increase in the overall resistance of the sensor, which transitions into a high-resistance state. Consequently, numerous Pd-based H₂ sensors have been developed, including those utilizing Pd nanomaterials, Pd-based alloys, and Pd-based composites. In this section, we review the fundamental sensing mechanism of Pd based sensors as well as recent advancements in metal-based H₂ sensors.

Fundamental mechanism of Pd–H interactions

Since Graham et al.⁶¹ first reported the hydrogen-storage property of Pd in 1866, the PdH_x hybrid has been systematically studied. During the adsorption process, H₂ molecules dissociate into hydrogen (H) atoms upon exposure to Pd, and these H atoms occupy the interstitial octahedral sites of a face-centered cubic (fcc) lattice as shown in Fig. 3a and b. Therefore, the transportation of free electrons within Pd crystal is impeded (namely electron scattering), and volume expansion of the Pd lattice occurs. The alterations in electrical properties and lattice expansion of the Pd crystal are significantly influenced by the composition of H atoms (x). When the x is in a low region (0 < x < 0.02), the PdH_x is stable at α phase with local expansion in Pd lattice, and the PdH_x is stable at β phase with huge lattice expansion when the x is in a high region (x > 0.6)⁶². While the two phases coexist when 0.02 < x < 0.6. Researchers have demonstrated that the lattice expansion of Pd is positively related to the concentration of H atoms via powder X-ray diffraction (XRD) characterization^{63,64,65,66,67}. Figure 3c presents the pressure-composition-temperature (PCT) curve of Pd-H system. With the increase of H₂ pressure, PdH_x undergoes a first-order phase transformation. During the coexistence of the α- and β-phases, a temperature-dependent plateau pressure is observed and as well as the hysteresis, which is generated by the PdH_x hydride formation and decomposition at constant temperature. Both the width of plateau and the extent of hysteresis shrink with increasing temperature, and ultimately vanishing as the temperature approaches the critical threshold (T_c). Figure 3d demonstrates a schematic graph of the energy landscape encountered by an H₂ molecule when interacting with a Pd surface. Obviously, the rate-limiting step in H₂ absorption involves the diffusion of H atoms from the Pd surface into its bulk.

Fig. 3: The fundamental mechanism of Pd-based H₂ sensor.

Schematic illustrations for (100) surface of fcc Pd (a) and the phases formed during H₂ adsorption (b); c PCT curves of bulk Pd-H on desorption process; d Energy landscape encountered by a H₂ molecule upon interaction with a Pd surface. In the first step, the H₂ molecule dissociates on the Pd surface into hydrogen atoms (H). In the next step, the H atoms diffuse into the subsurface region and occupy subsurface interstitial lattice sites. Subsequently, H diffuses interstitially further into the bulk

As previously mentioned, the adsorption process of H₂ induces electronic change and lattice expansion. Therefore, electron scattering is the main sensing mechanism for Pd-based H₂ sensor, leading to an increase in sensor resistance^68,69. A secondary mechanism associated with Pd-based sensor pertains to the volume expansion of Pd, which can be achieved through the design of discontinuous Pd nanostructures or films (nanogap-controlled Pd-based H₂ sensors). For the second type of sensor, the resistance decreases upon H₂ adsorption due to the formation of new electrical contact points within the Pd nanostructures/films^70,71,72,73. However, the pressure plateau and the hysteresis of H₂ adsorption/desorption also cause a series of problems in H₂ sensing. First, according to the PCT curve, the coexistence of two phases occurs within a narrow pressure range, leading to low sensitivity. Second, owing to the hysteresis loop between H₂ adsorption and desorption, it becomes challenging to obtain a precise signal related to H₂ pressure at any specific point in time. Therefore, to enhance the performance of Pd-based H₂ sensor (e.g., sensitivity, t_res/t_rec, and limit of detection (LOD)), researchers have proposed various approaches such as the design of Pd nanostructures, Pd nanogap, and Pd-M alloy nanomaterials. Compared to their bulk counterparts, Pd-based nanomaterials offer greater degrees of freedom such as size, shape, alloy composition, and nanogap distance, which can dramatically affect the H₂ sensing performance. Furthermore, linear detection of H₂ in a wide concentration range and long-term duration of Pd-based H₂ sensors are achieved via device structure design.

Pd nanostructure-based H₂ sensor

Recently, Pd nanomaterials have been utilized as H₂ sensing material due to their unique selectivity toward H₂, high specific surface ratio and quantum size effect. Yamauchi et al.⁷⁴ reported that decreasing palladium nanoparticle (NP) size leads to increased hydrogen solubility in α-phase, whereas β-phase shows a distinct behavior. Thus, the pressure plateau narrows and the T_c significantly decreases with reduced size of Pd NPs. In addition, decreasing the size of Pd nanomaterials may also reduce the t_res by shortening the diffusion distances for H atoms^75,76,77. Langhammer et al.⁷⁸ were the first to measure the kinetics of H₂ adsorption and desorption in Pd NPs. They demonstrated that the kinetics exhibit a strong size dependence since both the adsorption/desorption times of H₂ decline with the reduction in Pd NP size, which is consistent with Monte Carlo simulation results of diffusion-controlled adsorption kinetics⁷⁸. Thus, engineering efforts aimed at downsizing Pd nanomaterials would weaken the impact of pressure plateau and hysteresis^79,80,81,82. Cho et al.⁸³ reported a H₂ sensor based on ultrasmall grained Pd NPs, which exhibited no hysteresis. Figure 4a displays the H₂ device structure derived from the fabricated Pd nanopattern arrays with the smallest grain size and interface dimensions of 5 nm and 2 nm, respectively. Attributing to the small grain size/interface, this unique Pd nanopattern array demonstrated the capability to detect a wide range of H₂ concentrations (2.5–30,000 ppm) with minimal hysteresis effects (Fig. 4b, c). It exhibited both lattice expansion and electron scattering mechanisms across different ranges of H₂ concentration. The lattice expansion mechanism predominates at low concentration ranges (2.5–100 ppm), resulting in a reduction in resistance upon exposure to H₂. In contrast, the sensor transitions to an electron scattering-dominated regime under high H₂ concentration ranges (0.5–3%), exhibiting an increase in sensor resistance. Kumar et al.⁸⁴ revealed that the polycrystalline Pd nanowires (NWs) exhibited faster t_res/t_rec and higher sensitivity throughout 0.1–1% H₂ compared to quasi-single-crystalline Pd NWs (Fig. 4d-h). The improved sensing properties is caused by the reduced grain size as well as exposed high-index active facets^84,85. Li et al.⁸⁶ reported that Pd octahedrons with {111} facets have a fast response for H₂ adsorption compared with Pd cubes with {100} facets through XRD and in-situ solid-state ²H NMR characterization. Except for reducing Pd NP size, increasing of exposed surface area is also an effective strategy for improvement sensor sensitivity. For instance, employing low-cost porous paper as a substrate greatly enhanced the sensitivity of polycrystalline Pd NWs-based H₂ sensor ascribing to the increased contact area of H₂ and Pd NWs, as shown in Fig. 4i. Different from previous reports that focus on reducing the size of Pd nanomaterials, Li et al.⁸⁷ fabricated highly sensitive H₂ sensors using Pd hollow shells (with ~10 nm thickness) by using Si nanoparticles (283 to 633) as template (Fig. 4j-l). The sensor achieved the LOD at 75 ppm when utilizing the largest Pd hollow shell size (633/10). Such sensitivity can be attributed to a larger volume expansion, which can effectively tune the sensor’s resistance even at low H₂ concentrations. Consequently, decrease of the Pd particle size, increase of the exposure area of Pd, and construction of grain boundaries can considerably enhance the sensitivity and lower the t_res/t_rec of sensor toward H₂, as well as extend the detection range, achieving several ppm level LOD. However, the sensor exhibits t_res/t_rec spanning several hundred seconds, which hinders real-time application in providing early warning for TR in LIBs. Therefore, it is imperative to explore innovative strategies to enhance key sensing performance.

Fig. 4: H₂ sensors based on Pd nanograins.

a–c Schematic graph of H₂ sensors based on a high-resolution Pd nanopattern with ultrasmall grain boundaries and relevant real-time dual-switching H₂ sensing behavior with variation of the H₂ concentration range⁸³. Copyright 2018, American Chemical Society. d–f Schematic graph of paper-based H₂ sensor using polycrystalline Pd NWs and relevant dynamic H₂ sensing performance of two sensors; g, h t_res/t_rec vs H₂ concentration (0.1–1 vol %) of the sensors; I the influence of substrate to the sensing response of polycrystalline Pd NWs-based H₂ sensor⁸⁴. Copyright 2023, American Chemical Society. J Schematic fabrication of Pd hollow shells; k, l Real-time response of Pd hollow shells to different concentrations of H₂ in N₂ at 25 °C and its plotted response curve (the core diameter is 633 nm and the shell thickness is ~10 nm)⁸⁷. Copyright 2019, Elsevier

Pd nanogap-controlled H₂ sensor

Among Pd-based H₂ sensors, the sensing mechanism of Pd nanogap-controlled H₂ sensors is dominated by lattice expansion, resulting in an on-off behavior. These sensors usually exhibit high sensitivity, and fast t_res/t_rec. For this type of sensor, detecting conductive signals in the initial state (off-state) poses a challenge due to the presence of the formed nanogap. However, upon exposure to H₂ (on-state), there is a significant decrease in resistance attributed to lattice expansion that effectively bridges the nanogaps. In 2001 year, a Pd nanogap-based H₂ sensor was firstly proposed by electro-deposition of Pd mesowire arrays⁷¹. Although this Pd nanogap-controlled H₂ sensor demonstrated a rapid t_res (<75 ms), it exhibited a disadvantage in detecting trace H₂ since the LOD was 2.25% H₂⁷¹. While, by increasing the density of Pd NPs, the Pd mesowire-based sensor can operate as an always-on-state sensor, resulting in a reduced LOD of 0.5% H₂. Thus, the width of nanogap plays an essential role in determining the performance of H₂ sensing. To date, there are two primary strategies for forming nanogaps: (ⅰ) a lithography-free method that uses an elastomeric substrate (ES) to fabricate Pd nanogap-based H₂ sensors; (ⅱ) a lithography method employing rigid substrates.

For the first strategy, the formation mechanism of Pd nanogaps is ascribed to the differences in properties between ES and Pd films. Specifically, ascribing to the distinct Young’s moduli of ES and Pd films, a Pd film-crack-based H₂ sensor was firstly fabricated through mechanically stretching the ES-polydimethylsiloxane (PDMS)-to form nanogaps at the edge of the broken Pd films^88,89. In addition, owing to the different volume expansion upon H₂ between Pd film and ES (which is insensitive to H₂), cracks are generated in the Pd film when removing the H₂ and the width of Pd nanogap can be modulated by varying the initial concentration of H₂^90,91. In addition, the disparity in thermal expansion coefficients between Pd film and ES provides an additional force for nanogap formation when subjected to liquid nitrogen freezing treatment^{92,93,94,95,96}. However, since the Pd films are simply deposited onto the surface of ESs, those sensors are suffering from poor mechanical stability and limited stretchability, thereby hindering their application in wearable electronic devices. To address this issue, Won et al.⁹⁷ proposed a sensitive and stretchable H₂ sensor by embedding Pd NP networks in PU fiber through a facile two-step chemical solution process as shown in Fig. 5a, b. The PU fiber was firstly immersed in a Pd precursor solution, subsequently, it was air-dried before being immersed in a reducing agent. Figure 5c demonstrates that the obtained H₂ sensor exhibits a wide sensing ranging from 5 to 100000 ppm, with the capability to detect under strains of up to 110%. Son et al.⁹⁸ used the similar method to develop a H₂ sensor based on PdO NPs-embedded carbon nanotube (CNT) yarns, which demonstrated well-maintained sensing performance under two deformation states, namely, bending and loading. Therefore, embedding sensing material networks into ESs presents a promising approach for enhancing stability, despite the random distribution of nanogaps with varying widths. To precisely control the width of nanogap, a rigid substrate (e.g., steel foil) was used to fabricate Pd/Cr nanogap H₂ sensor via one-step bending deformation by using a cylinder⁹⁹. The width of the nanogap can be adjusted from 8 to 80 nm by varying the diameter of the cylinder (1–6 mm) (Fig. 5d). Figure 5d illustrates a schematic diagram of the fabricated sensor, which consists of 4 layers, namely, a steel substrate, Kapton tape, a 3 nm Cr film, and a 2 nm Pd film, respectively. The sensor based a single nanogap (15 nm) works as an on-off behavior and exhibits an optimized sensing performance, including broad detection range (0.0001–4% H₂), good repeatability and fast t_res/t_rec (3/4.5 s) as displayed in Fig. 5e-g. Moreover, lithography method is also an another effective strategy to precisely control the width of the Pd nanogap with high uniformity and repeatability^100,101. However, achieving sub-100 nm gaps using conventional UV lithographic techniques poses significant challenges^101,102,103. To address this problem, a self-aligned nanogap is formed using oxidized Cr film as a shadow by using two steps of photolithography and the width of nanogap (45–300 nm) depends on the thickness of Cr film (Fig. 5h)¹⁰¹. Figure 5h also displays the cross-section diagram of a sensor, clearly showing that the Au and Pd films are partially suspended due to the third photolithography for selective wet etching of the underlying SiO₂ layer. This suspended film structure provides enough space for the expansion of the Pd film when exposure to high concentration of H₂ (4%). Figure 5i-k presents the dynamic sensing response of the suspended Pd film, demonstrating that the sensor is capable of detecting H₂ concentrations ranging from 1% to 4%. To improve the sensitivity to low H₂ concentrations, Pak et al.¹⁰⁰ proposed a strategy for fabricating a 15 nm nanogap sensor through transferring a deposited PdAu film onto a polystyrene substrate using a Si stamp with periodic gratings measuring 500 nm in width and 100 nm in height (Fig. 5l, m). During the detachment process, the width of the PdAu nanogap decreases in a controllable manner (from 40 nm to overlap) due to variations in the shrinkage degree of the PS substrate at different detachment temperatures. The 15 nm PdAu nanogap sensor demonstrates a fast t_res (11.54 s) across the full detection range of H₂ concentrations (0.005–10%) and the sensing response is much higher than that of sensor based on pure Pd nanogap (Fig. 5n and o). Therefore, precisely control the nanogap width of nanogap architectures has shown promise in enhancing sensitivity (ΔR/R₀ > 1), expanding detection range (several ppm to 4% H₂), and accelerating response/recovery dynamics (few seconds). However, there is remain a significant gap in meeting the targets set by US DOE, which specify a t_res/t_rec within 1 s for H₂ concentrations ≥ 1%. While this disadvantage could be addressed by integrating with algorithms as discussed in the following section “Pd-based H₂ sensor integration of algorithms”.

Fig. 5: Different strategies for the design of Pd nanogap controlled H₂ sensors.

a Schematic graph of the fiber multimodal sensor and b structure of the conducting network regarding Pd NPs and PdH_x NPs on the surface of fiber; c response of the 0–100% strained sensor for 10% to 5 ppm H₂ gas⁹⁷. Copyright 2020, American Chemical Society. d Schematic diagram of fabricated Pd/Cr nanogap sensor and e its on-off operation mechanism; f Real-time resistance sensing response of optimized Pd/Cr gap sensor for various H₂ concentrations (10–40,000 ppm) and g resultant linear correlation⁹⁹. Copyright 2020, Elsevier. h Optical image of the Pd nanogap H₂ sensor and its schematic cross-sectional structure; i–k Schematic illustrations of the sensor structures and sensing operations in response to two different H₂ concentration ranges¹⁰¹. Copyright 2022, American Chemical Society. l Schematic fabrication procedure of PdAu nanogap sensor; m SEM images of four samples with different nanogap widths when detached at different temperatures; n dynamic sensing response of a 15 nm Pd_0.6Au_0.4 sensor to 0.005–10% H₂ and o the repeatability test to 0.5% H₂¹⁰⁰. Copyright 2018, John Wiley and Sons

Pd alloy-based gas sensor

As discussed above, there are two primary sensing mechanisms for Pd-based H₂ sensor. However, current pure Pd-based sensors face challenges due to their high cost and inherent hysteresis associated with hydride formation and decomposition, which significantly diminishes sensor accuracy¹⁰⁴. The fabrication of Pd alloys with other metals (Ms) presents an effective approach to reduce costs and eliminate hysteresis, as confirmed by both theoretical calculations and experimental tests^105,106. The presence of the second M, which is non- or weak hydride former, increases the energy barrier for hydride formation during hydrogen absorption^51,105. When the atomic radius of the alloyants exceeds that of Pd, such as Au, Y, and Mo, the lattice structure of Pd expands, thereby reducing the strain-induced energy barrier caused by hydrogen absorption¹⁰⁴. Mamatkulov et al.¹⁰⁵ used density functional theory (DFT) to study the influence of other Ms and presented that the hysteresis of Pd toward H₂ can be dramatically suppressed by introducing Au and Ta. According to their calculations, the fraction (f) of the second M, H-M interaction (ε_HM), and effective M-M interaction (ε*) play crucial roles in suppression of hysteresis in Pd, as shown in Fig. 6a-c. In pure Pd (Fig. 6a), H₂ is absorbed at RT up to an H/Pd ratio of ~0.66, which exhibits considerable hysteresis with a high T_c (600 K). While the hysteresis loop in the alloys shifts to the left, and the T_c of Pd_0.85Ta_0.15 decreases to ~320 K (Fig. 6b). Notably, the reduction in T_c exhibits a positive correlation with ε_HM when compared with other additives¹⁰⁵. Figure 6c illustrates that an increase in f can significantly suppress hysteresis as well. Therefore, considerable efforts have been made to develop hysteresis-free H₂ sensors based on Pd alloys, particularly in comparison to sensors based on pure Pd, especially in the realm of optical H₂ sensors^107,108,109.

Fig. 6: The advantages of H₂ sensors based on Pd-M alloy.

Calculated H₂ adsorption isotherms for a pure Pd and b Pd₈₅Ta₁₅ alloy at four different temperatures; c calculated H₂ absorption isotherms for the Pd-Ta alloys at 30 °C¹⁰⁵. Copyright 2021, Elsevier. d Schematic diagram of prepared PdSn alloy NTs; Normalized response versus H₂ concentrations for PdSn alloy NTS, PdSn NFs, Pd NFs, and Sn NFs at H₂ concentrations of e 0.5–200 ppm and f 500–30,000 ppm¹¹¹. Copyright 2022, American Chemical Society. g repeatability test for 1% H₂ of PdNi alloy film annealed at 250 °C¹¹³. Copyright 2020, Elsevier. h Repeatability test for 1% H₂ of PdNi alloy film prepared under 5 Pa and I the variations in _ΔR values of three PdNi alloy samples deposited under different pressures¹¹⁴. Copyright 2024, Elsevier. j Long-term repeatability curves of the nonalloyed Pd and Pd_0.62Au_0.38 sensors at 5% H₂ atmosphere and k schematic of the mechanism of resistance change when H₂ penetration occurs in Pd and PdAu¹¹⁵. Copyright 2024, American Chemical Society

In addition, the introduction of a second M can significantly improve the sensing performances of the sensor, including sensitivity, stability and t_res/t_rec. Zhao et al.¹¹⁰ demonstrated that PdAu alloy-based sensors achieved a 3.9-fold faster t_res and a 1.1-fold faster t_rec compared to pure Pd-based devices. Additionally, the PdAg alloys exhibited a higher sensitivity for H₂ detection than pristine Pd. Kim et al.¹¹¹, employed Sn as a second M to fabricate low-cost PdSn nanotube (NT)-based H₂ sensors. Although Sn itself is not particularly sensitive for H₂ sensing, its presence significantly improves both the sensitivity and tolerance of the sensor up to 4% H₂ (Fig. 6d-f). Because PdSn alloys exhibit lower oxygen adsorption energy compared to pristine Pd, they can facilitate a greater amount of chemisorbed oxygen species, thereby enhancing the sensitivity of the sensor. Furthermore, when exposed to high concentrations of H₂ (0.05–3%), the presence of Sn effectively inhibits phase transitions and the growth of Pd nanograins, significantly improving the endurance and stability of sensor to under elevated H₂ concentrations. Similarly, the incorporation of 8% Ni into Pd film also can significantly inhibit the α-β transformation, thereby enhancing the sensor’s tolerance of the up to 4%¹¹². Additionally, a zero-drift suppression in PdNi alloys has been reported through the reduction of defects in PdNi alloy films via annealing treatments (Fig. 6g)¹¹³. Liu et al.¹¹⁴ obtained PdNi alloy films with different morphologies by adjusting the deposition pressure and founded that the degree of rearrangements in the alloy structures is influenced by their morphologies. As a result, the sensors (PdNi-5 Pa-0) featuring tiny cracks, which were obtained under a pressure of 5 Pa, exhibited superior stability compared to those obtained under different deposition pressures (Fig. 6h, i). This enhanced stability can be attributed to the minimal structural rearrangement observed after exposure to H₂. Kim et al.¹¹⁵ reported a H₂ sensor based on PdAu alloys, which demonstrated excellent durability with an optimized composition of Pd_0.62Au_0.38. The drift of the baseline and changes in sensing response are neglective (only 0.02% per cycle) during over 35 h of repeated operation in 5% H₂, as illustrated in Fig. 6j. Whereas the resistance of Pd drifted significantly, averaging 1.6% of the initial value per cycle, and failed completely after 15 cycles (Fig. 6j). Notably, in contrast to sensors based on pure Pd, the Pd_0.62Au_0.38-based sensor illustrated a decrease in resistance when exposed to H₂. The relative sensing mechanism is illustrated in Fig. 6k. For sensors based on pure Pd, the formation of PdH_x in Pd lattice results an increase in resistance. Whereas, when H atoms penetrate the octahedral sites of Pd₄Au₂, it induces the displacement of Au atoms while simultaneously compressing the Pd matrix. The displaced Au atoms are then compressed together with those Au atoms sites that remain unoccupied by H atoms, leading to the formation of a compressed layer. This induced pressure consequently results in a decrease in resistivity within the compressed metal structure. Thereby, construction of Pd alloys cannot only reduce the cost but also can improve sensing properties, including suppressing hysteresis enhancing sensor sensitivity and stability. Particularly, the long-term stability over months has been verified experimentally, during which both the signal response and baseline resistance remained virtually unchanged. Such remarkable stability underscores the reliable deployment of Pd alloys in the demanding environment of battery safety monitoring.

Novel Pd-based H₂ sensor design

Pd nanograins, Pd nanogaps, and Pd alloys have been reported to enhance the sensor performance, such as sensitivity, stability and reduce of hysteresis. However, due to the inevitable phase transition upon H₂ concentrations above 2%, it induces undesirable electrical and mechanical alterations for the sensors based on Pd. In particular, nonlinear gas response (ΔR/R₀) associated with phase transitions has posed a significant challenge for detection of high H₂ concentrations. Although fabrication of Pd alloys is an efficient approach to suppress phase transition and shows the capability to detect H₂ in high concentrations, it is difficult to explicitly control the material composition with the alloying method for nanofabrication with high reproducibility^116,117. Unlike material alloying, employing a mechanical approach to suppress lattice expansion offers greater flexibility in the design of nanomaterials. Kim et al.⁷⁹ reported that the phase transition is suppressed by introducing of a buffer layer at the interface of the Pd film and substrate. In 2022, Yoon’s group designed a phase transition-inhibited Pd nanowire H₂ sensor that can linearly detect H₂ up to 4% with high sensitivity by inserting a layer of Al₂O₃ between Pd and substrate¹¹⁸. Figure 7a shows the schematical graph of the Pd NWs-based sensor. The device comprises an aligned array of Pd NWs serving as the sensing material, equipped with four electrodes. Figure 7b presents the mechanism of suppression of phase transition. As mentioned, the formation of PdH_x leads to the lattice expansion of Pd. However, as the bottom of the Pd NWs is anchored to the substrate, a compressive stress is formed at the interface between Pd NWs and substrate to constrain the volume expansion. Considering this factor, the phase transition would be inhibited, particularly in regions adjacent to the substrate where high levels of compressive stress are generated. It was calculated that at least 0.25 GPa stress is needed to prevent phase transition at 4% H₂ (Fig. 7c). Based on the assumption that PdH_x expands with a linear expansion coefficient of 3.5%, the author found that the stress distribution of the Pd NWs depends on their geometrical dimensions such as the thickness (t_NW) and width (w_NW) of Pd NWs by using a finite element method. Figure 7d demonstrates the stress distribution of a Pd NW results when the proportion of the area with a stress greater than 0.25 GPa to the total area (A_{>0.25 GPa}/A_total). Thus, to accommodate the stress between Pd NWs and the substrate during absorption of H₂, a lot of efforts have been done to modulate the thickness and width of Pd NWs to implement linear detection of H₂ concentrations up to 4% (Fig. 7e and f). Highly linear and distinguishable gas response (linearity = 98.9%) for up to 4% H₂ was achieved by optimizing the geometrical dimensions of Pd NWs (t_NW = 15 nm, w_NW = 160 nm), as displayed in Fig. 7g. In 2023, to further broaden the linear detection up to 10% H₂, Yoon’s group employed thermal activation mechanism to achieve fast response rate and inhibit the phase transition of PdH_x through device structural design¹¹⁹. Because providing thermal energy to Pd is an effective approach for enhancing the response rate by accelerating the hydrogen adsorption process and thermodynamically inhibiting the phase transition. The schematic graph of the device is displayed in Fig. 7h. Different to conventional device structures, in which a heater, an insulating layer, and sensing materials are vertically arranged, the Pt heaters are laterally placed at the bottom side of the suspended Pd NWs with C-channel-shape feature (Fig. 7i, j). This unique structure can maximize the exposed reaction sites of the Pd NWs and raising the temperature by lateral heaters. By measuring the continual gas response of the device depending on the temperature, the operating temperature for a linear gas response without phase transition is 65 °C (Fig. 7k). The designed sensor demonstrates the capability to detect H₂ within 0.6 s, which is the fastest detection rate in metal-based chemiresistive sensors at the time of publication, with high linearity (98.8%) across a concentration range of 0.1–10% at 65 °C when the thickness of Pd is 20 nm (Fig. 7l and m). Therefore, ingenious device structure design can also suppress phase transition and expand the linear detection range. This strategy demonstrated a potential approach for the early detection of H₂ in TR process. Moreover, if incorporating strategies discussed in prior sections, further improvements in sensitivity could be achieved to realize the practical deployment in LIB safety systems.

Fig. 7: The strategies for improving the sensing linearity of Pd based H₂ sensors.

a Schematic illustration of perfectly aligned Pd NWs based H₂ sensor with inhibited phase transition. b Schematic of the internal lattice structure of Pd NWs with two phases; c Gibbs free energy change of PdH_x phase transition (ΔG_α→β, kJ/mol) according to stress (σ) at four different concentrations of H₂. d Ratio of a cross-sectional area with a stress greater than 0.25 GPa to the total area with respect to w_NW and t_NW; Influence of e t_NW and f w_NW of the Pd NWs to the H₂ sensing performance; G Gas response based on optimized Pd NWs sensor (w_NW = 160 nm, t_NW = 15 nm) as increasing and decreasing H₂ concentration¹¹⁸. Copyright 2022, American Chemical Society. h Schematic graph and I, j SEM images of the nanoelectromechanical H₂ sensor; k Hysteresis characteristics depend on the sensor temperature; l Response time and m linear gas response according to the H₂ concentrations at 65 °C¹¹⁹. Copyright 2023, American Chemical Society

Pd-based H₂ sensor integration of algorithms

Rapid H₂ detection is the prerequisite of LIB TR monitoring. Therefore, it is essential to develop a fast H₂ sensing system with a wide detection range. Although Pd alloys can efficiently broaden the linear detection range and enhance the stability, while the t_res/t_rec of most the reported chemiresistive sensors still unable to meet the requirement of US DOE^{111,112,113,114,118}. To address this issue, integration of algorithm in the H₂ sensing system to predict the H₂ concentration based on first few data samples could achieve ultrafast alarming (<1 s). Leveraging the advantages of Pd alloys and the principle that heating can mitigate hysteresis, Huang et al.¹²⁰ developed a PdNi alloy based H₂ sensor with the capability of fast detection of 0.1–4% H₂ at 40 °C (<1 s) as shown in Fig. 8a-d. Although the t_res of the sensor meets the requirement of US DOE under specific conditions, the t_res still exceed 1 s toward H₂ concentration lower than 1% (Fig. 8c, d). To tackle this problem, the authors develop a neural network concentration prediction method utilizing an autoencoder architecture for faster concentration output (Fig. 8e). The underlying mechanism is grounded in the adsorption kinetics of H₂ on Pd. After acquiring the real sensor response dataset, zero drift compensation and normalization were applied to minimize drift impact data scale differences. The mean (μ) and standard deviation (ω) of the first 50 values of the baseline was calculated and used to set the threshold (μ + 3ω) for determining the start of a response. The start value r[t] is determined when r[t]−r[t-1] > μ + 3ω. Once the initial r[t] is obtained, a short segment of early response signal (40 data points, 150 Hz, ~0.3 s) is used as the input to the algorithm. Therefore, by extracting ~0.3 s data segment to the H₂ concentration model which is consisted of an encoder and a fully connected neural network (FCNN) can successfully predict H₂ concentration. It can be clearly observed that most of the predicted concentrations are almost consistent with the true concentrations, the relative error has a mean error of 0.17% and standard error of 1.36%. A similar PdNi based H₂ sensor, which is using polyimide film as substrate, with the ability to detect as low as 5 ppm is reported by the same research group (Fig. 8f-g)¹²¹. Furthermore, by integration with the prediction model algorithm, the alarm triggering time was greatly shortened to 0. 4 s compared to the system without the algorithm (Fig. 8h, i). Therefore, integration algorithm to the H₂ detection system offers a promising solution to decreasing the alarm time within sub-second under the full detection range.

Fig. 8: Advantages of integrating algorithms into H₂ sensors.

a Schematic diagram of the coplanar-structured H₂ sensor based on a PdNi film; b Hysteresis characteristics depend on the sensing material and temperature; c t_res and response value of the sensor towards 0.4–4% H₂; d sensor response towards various random H₂ concentrations; e working principle and flowchart of the prediction algorithm and its predicted results¹²⁰. Copyright 2025, Elsevier. f, g Dynamic sensing response of PdNi film-based sensor upon 5–40,000 ppm H₂; h prediction algorithm of H₂ sensing system; i Comparison of the response time of the detection system with and without the introduced prediction algorithm for different concentrations of H₂¹²¹. Copyright 2025, American Chemical Society

Table 2 presented sensing properties of metal-based H₂ sensors. In summary, Pd is a typical metallic chemiresistor for H₂ detection through phase transition mechanism and lattice expansion mechanism. Although Pd-based sensors are capable of detecting H₂ at RT with high selectivity, they usually displayed poor sensing properties, such as low responses and long t_res/t_rec. To date, various approaches have been reported to enhance their sensing performance including reducing size of Pd NPs, employing Pd alloys, construction of Pd nanogaps, design novel device structures, and integration with prediction algorithms. Some of these strategies are capable to enhance the sensor’s stability, broaden the linear detection range, and improve the t_res/t_rec for real-time TR monitoring. However, the expense associated with noble-metal-based sensors is significantly higher than that of other sensing materials, which may pose a challenge for their commercialization.

Table 2 Summary of sensing properties of metal-based H₂ sensors

MOS-based chemiresistive H₂ sensor

The MOS-based resistive gas sensors have garnered great attention due to their high chemical and thermal stability, rapid t_res/t_rec, and low cost, positioning them as a primary research direction for H₂ sensing. However, the sensitivity of MOS-based sensors is fundamentally reliant on the redox reactions that occur between adsorbed oxygen species and target gases. This intrinsic mechanism cannot completely eliminate cross-sensitivity, which ultimately constrains gas selectivity. In addition, many MOS-based gas sensors require high operation temperatures (>200 °C) to ensure an adequate number of charge carriers enter the conduction band for active participation in these reactions. This requirement not only increases power consumption and impacts the long-term stability of the sensors but also poses a safety concerns when detecting combustible gases. Additionally, the sensitivity of sensors based on a single MOS usually shows low sensitivity. Consequently, various investigations have been performed to enhancing sensitivity, improving selectivity and reducing the operational temperature of MOS-based H₂ sensors. In this section, we present the fundamental sensing mechanisms and recent advancements in the development of MOS-based H₂ sensors, including single MOSs, heterojunction MOS nanostructures, and MOSs modified with noble metal NPs.

Fundamental sensing mechanisms for MOS-based H₂ sensors

In general, the basic mechanism underlying MOS-based sensors involves a change in current or resistance induced by the interaction between H₂ and absorbed oxygen species. This process typically encompasses two key steps: (i) adsorption of O₂ to form reactive oxygen species, and (ii) subsequent reactions between these oxygen species and H₂ molecules, as shown in Fig. 9a^122,123. The absorbed O₂ molecules serve as electron acceptors, capturing electrons from the MOSs and dissociating into different oxygen species (O₂⁻, O⁻, O²⁻), which are temperature dependent^124,125,126. MOSs can be categorized as either n-type or p-type MOSs based on their dominant carriers (electrons for n-type and holes for p-type semiconductors). When O₂ adsorb on the surface of n-MOSs, there is a reduction in electron density at the surface of n-MOSs, resulting in the formation of an electron depletion layer (EDL). The presence of the EDL dramatically increases the resistance of n-MOSs due to the decrease in electron density and the formation of potential barriers at grain boundaries. Upon exposure to reducing gases (e.g., H₂), H₂ reacts with oxygen species to produce H₂O and release electrons back to n-MOSs. This reaction leads to a decrease in the width of the EDL and consequently results in a decrease in resistance. Whereas, for p-MOSs, a hole accumulation layer (HAL) is formed after absorption of oxygen species through electron extraction. As the predominant carrier of p-MOSs is hole, the resistance of p-MOS-based sensors decrease with the formation of HAL. The reaction between H₂ and adsorbed oxygen species effectively neutralizes these holes by releasing electrons, which leads to a shrinkage of HAL and increase in resistance of p-MOS-based sensors. Therefore, for both n-type and p-type MOS-based sensors, it is essential to enhance the specific surface area and elevate reactivity towards analytes to improve their sensing capabilities.

Fig. 9: Fundamental sensing mechanisms of H₂ sensor based on MOSs.

Schematic representation of H₂ sensing mechanisms: a Reducing gas effect mechanism of n- and p-type MOSs; b Metallization effect mechanism of ZnO nanograins

In addition to the reaction between oxygen species with H₂ molecules, certain MOSs, such as ZnO and CuO can directly react with H₂ molecules, leading to the metallization of MOSs^{127,128,129,130}. As shown in Fig. 9b, the metallization mechanism of ZnO-based H₂ sensors involves the formation of a metalized region on the ZnO surface upon exposure to H₂. This process facilitates electron flow from the metallic Zn surface back to ZnO, significantly lowering the sensor’s resistance and thereby improving its sensing performance. When these materials are re-exposed to air, the metalized region oxidizes and reverts back to ZnO, allowing the resistance to return to its original baseline level.

H₂ sensors based on single MOSs

Because of the advantages of low cost, stable physical and chemical properties, and the abundance of available materials, MOSs have been extensively employed in H₂ sensing. The fundamental sensing mechanism of MOSs indicates that their sensing response is greatly influenced by the reactivity of materials towards H₂ and the quantity of adsorbed oxygen species. Therefore, numerous researchers are concentrating on obtaining MOSs with high reactive facets and increasing adsorption sites of MOSs^{131,132,133,134,135}.

In terms of exposed facets, since the surface energy of various facets differs arising from distinct atomic arrangements, researchers have studied the correlation between sensing performance and exposed crystal facets of MOSs from both theoretical and experimental perspectives^{131,136,137,138,139}. Zhou et al.¹⁴⁰ successfully tuned the exposed facets of rutile TiO₂ by varying the ethanol content in the hydrothermal solvent. Figure 10a demonstrates the XRD patterns for rutile TiO₂ prepared on fluorine-doped tin oxide (FTO) substrates. E_x is used to label the TiO₂ samples, where x denotes to the specific volume of ethanol used, such as 2, 4, 6, 8, and 10 mL. Obviously, the ratio of (002) facet increases along with the ethanol content. The sensing response values, t_res/t_rec of different TiO₂ samples are demonstrated in Fig. 10b, c. DFT simulation results reveal that H₂ preferentially adsorb and dissociate on the (002) and (101) facets of TiO_2, attributing to a low energy barrier. This phenomenon leads to improved sensitivity and rapid t_res. While the negligible reaction barrier for atomic H to recombine into H₂ molecules on the (110) surface of TiO₂ facilitates fast t_rec. Thereby, it is essential to rationally design MOSs with appropriate facet ratios to balance the sensing performance of sensors^140,141,142.

Fig. 10: Different approaches for regulating the sensing performance of pristine MOSs.

a XRD patterns of TiO₂ films synthesized with different ethanol contents. * indicates the diffraction peaks from FTO substrate; b intrinsic resistance of four samples without H₂ and the response of four samples to different H₂ concentrations; c Response and recovery time of the hydrogen sensors to various H₂ concentrations¹⁴⁰. Copyright 2018, American Chemical Society. d Schematic of the formation process of SnO₂ nanospheres and e, f the sensing performance of four sensors toward H₂¹⁵³. Copyright 2022, Elsevier. g Synthesis process for 2D holey/unholey ZnO NS; h Linearity sensing response of holey/unholey ZnO based sensors to different H₂ concentrations and i The response/recovery time of optimized holey ZnO-based sensor toward 100 ppm H₂¹²⁹. Copyright 2021, Elsevier

So as to achieve a greater number of adsorption sites, researchers typically employ the following two strategies: (ⅰ) modulation of defect concentrations^143,144,145 and (ⅱ) increase of specific surface area^146,147. To date, annealing, plasma etching, and chemical reactant treatments have emerged as three predominant approaches for adjusting defect concentrations^{148,149,150,151,152}. Figure 10d presents a general annealing treatment to obtain SnO₂ with different concentrations of oxygen vacancies (O_V) under H₂ atmosphere by tuning the temperatures¹⁵³. The percentage of O_V for SnO₂, SnO₂-D3, SnO₂-D4 and SnO₂-D5 are 21.11%, 22.32%, 29.47%, and 23.65%, respectively. Among these samples, SnO₂-D4 illustrates the highest sensing response toward H₂, attributed to its elevated concentration of O_V (Fig. 10e, f). This enhancement in sensitivity can be explained by the role of O_V as potential active centers for gas adsorption^154,155. Additionally, the presence of O_V increases the electron density in the conduction band of SnO₂-D4, thereby facilitating the formation of reactive oxygen species^153,156. Wang et al.¹⁵⁷ also revealed that O_V can promote the H₂ sensing performance of ZnO. Moreover, they found that both O_V and metal vacancy (M_v) may influence the H₂ sensing behavior, as the position of Fermi level (E_f) is affected by the types of existing defects. To enhance the specific surface area, in addition to reducing the dimensional size of sensing materials such as 0D quantum dots^158,159, 1D NWs/NTs^160,161,162, and 2D nanosheets (NSs)^163,164, designing porous structures presents as a promising strategy^165,166,167. Due to their inherently high specific surface area, porous structures facilitate the migration of gases in and out more effectively than non-porous structures. Kumar et al.¹²⁹ synthesized 2D holey ZnO NSs for H₂ detection at RT by programmatically tuning the pore/hole size through annealing treatment, as shown in Fig. 10g. The density of pore in ZnO NSs consistently reduces as the annealing temperature increases from 400 °C to 800 °C. Owing to the high specific surface area and increased channels for gas diffusion and transport, the 2D holey ZnO NSs obtained at 400 °C exhibit the highest response of ~115 (20 times more than ZnO@800 sample) towards 100 ppm H₂ at RT with fast t_res/t_rec (9/6 s), as displayed in Fig. 10h and i. Currently, many researchers use metal-organic-frame (MOF), such as ZIF-8, MIL-88, MIL-68, to develop porous MOSs for highly sensitive gas sensing^{168,169,170,171,172,173}. He et al.¹⁷³ synthesized hexahedral hollow porous In₂O₃, which is derived from MIL-68(In), for dual gas sensing of NO₂ (at RT) and H₂ (at 160 °C).

The above-discussed H₂ sensors are primarily using n-MOSs as sensing materials because n-MOSs is more sensitive to p-MOSs. The relation between resistance and the band banding in MOSs can be expressed as ${R}_{p}=\exp ({eVs}/2{kT})$ and ${R}_{n}=\exp ({eVs}/{kT})$ for p-MOSs and n-MOSs, respectively^174,175. Therefore, the response of p-MOSs is equal to the square of the response of n-MOSs ($S{R}_{p} \sim \sqrt{S{R}_{n}}$) with a similar morphology toward the same target gases¹⁷⁵. Although p-MOSs based sensor is theoretically poor than n-MOSs, the p-MOSs also arouse significant attention due to their excellent catalytic effects, and less affected by humidity and high temperature. To date, some p-MOSs such as CuO^176,177, NiO¹⁷⁸ and Co₃O₄^179,180 have been used for H₂ sensing. Volanti et al.¹⁷⁷ designed various morphologies of CuO for H₂ sensing with the ability to detect 10 ppm H_2, such as urchin-shaped NPs, fibers and nanorods. Zhao et al.¹⁷⁶ designed a nano-patterned CuO NWs with voids for ppb-level H₂ sensing by optimizing the channel width (33 nm) as shown in Fig. 11a-c. The CuO NWs consist of small CuO grains (<10 nm) with voids were obtained by “pre-H₂ annealing” fresh Cu NWs to enhance their crystalline and form semicircular Cu NWs, and then followed by ex situ oxidizing process. Attributing to pre-H₂ annealing treatment, no hysteresis or baseline shift was observed (Fig. 11d). In addition, the CuO NW with voids exhibited the ability to detect 5 ppb H₂ with improved t_res/t_rec (<10 s) at 150 °C (Fig. 11e). Although the t_res and t_rec are slightly prolonged under 50% relative humidity (RH), the sensitivity to H₂ at 50% RH showed almost equal to that under dry air (Fig. 11f). The humidity tolerance of CuO is also corroborated by another research work that reported nearly constant response values under 150 °C while RH varied from 25–55%¹⁸¹. In a word, the ultralow LOD, negligible baseline drift and humidity resistance properties of CuO NWs position them as a promising candidate for trace H₂ detection under demanding conditions during the TR monitoring. While it still falls short of the US DOE standard to a certain extent when discussing the t_res/t_rec of the device. Except for solely designing of sensing materials, using pulsed heating mode (PHM) is an effective strategy for enhancing device sensitivity, selectively identifying gas species, and lowering power consumption^182,183. Different to conventional operation mode, PHM has been rigorously explored as a power-saving strategy, minimizing the active operation time and average power consumption by applying pulse heating voltages^184,185. Yan et al.¹⁷⁸ significantly enhanced the sensitivity of p-NiO sensor toward H₂ by employing PHM strategy compared with conventional DC mode (Fig. 11g and h). Moreover, based on this working mode, it enhances the extrapolation of data derived from isothermal measurements of MOS sensors. This mode can significantly enhance the device ability for discrimination multi-gases when combining with algorithms (Fig. 11i-l)¹⁸³.

Fig. 11: The H₂ sensors based on p-type MOSs.

a Schematic diagram and SEM images of Cu NWs b before and c after Ar/H₂ and dry air annealing; d, e Dynamic sensing resistance and response of CuO NW sensors toward H₂ from 5 ppb–5000 ppm; f Dynamic sensing response under different RH conditions¹⁷⁶. Copyright 2024, John Wiley and Sons. g Schematic design concept and h surface temperature of the micro-hotplate; i Hydrogen sensing performance across different pulsed operation modes. j, k Transient current response to 200–1000 ppm H₂ and l linear gas response according to the H₂ concentrations under DC and pulsed operation¹⁷⁸. Copyright 2024, Elsevier

To date, the sensing properties of single MOSs-based H₂ sensors, such as sensitivity and t_res/t_rec, could be improved through facet engineering, porosity design, as well as vacancy regulation. However, many reported single MOSs exhibit a low response ratio of H₂ to its strongest interferent, affecting the accuracy of detection under TR conditions with complex gas mixture⁴⁴. This cross-sensitivity of single MOSs hinders their further advancements. Consequently, it is essential to investigate and design new functional MOSs materials to enhance the sensitivity, selectivity, and t_res/t_rec of H₂ sensors. The subsequent sections will provide a detailed discussion on heteroatom doping, construction of heterojunction, fabrication of ternary or quaternary MOSs, and decoration MOSs with metals.

Hetero MOS-based H₂ sensor

To achieve enhanced H₂ sensing performance in MOSs, several strategies have been performed, including heteroatom doping¹⁸⁶, the construction of heterojunction, and the fabrication of ternary or quaternary MOSs.

Doping is an effective method for improving the sensing performance of MOS through modulating their electronic structure and influencing their physicochemical properties such as bandgap, conductivity, and defect concentration^{187,188,189,190,191}. For instance, after doping of Mn into Fe₂O₃ lattice, it weakens the bonding between neighboring oxygen ions, leading to the release of oxygen ions and the formation of O_V¹⁸⁶. Kim et al.¹⁸⁸ reported that the percentage of O_V increased from 26.42% to 33.58% by doping 3% Cu²⁺ into the SnO₂ lattice, and the sensitivity of the sensor increased from 0.241%/ppm to 0.286%/ppm. Catalytic La³⁺, which facilitates fast dissociative adsorption of H₂ through polaron effect^192,193, was selected to dope SnO₂ nanofibers (NFs) with varying atomic ratios (At%) relative to Sn (0%, 0.5%, 1.0%, and 3%) via electrospinning and calcination. Figure 12a shows the SEM images of the fabricated four La-SnO₂ samples¹⁹⁴. The response value of 1% La-SnO₂ NFs for H₂ detection is 9.9 towards 100 ppm H₂ at 300 °C, which present an enhancement of 2.5 times and 1.6 times compared to pure SnO₂ and 3% La-SnO₂ NFs, respectively (Fig. 12b). When the At. % of La reaches a high level (≥3%), La³⁺ ions are expelled from the SnO₂ NFs, leading to the formation of a p-n junction between n-type SnO₂ and p-type La₂O₃. This p-n heterojunction significantly contributes to the enhanced sensing mechanism. In contrast, when the At. % of La is low, La³⁺ ions are incorporated into the SnO₂ crystal lattice by replacing the Sn sites, resulting in partial p-type doping regions within the SnO₂. The area occupied by these p-type doping regions undergoes expansion upon exposure to air and shrinkage when exposed to H₂. Owing to the abundant pores present in the La- SnO₂ NF film, H₂ can easily penetrate into/depart from the entire sensing film. This contributes to rapid t_res/t_rec (~1 s), which is approaching the detection speed set by US DOE as shown in Fig. 12c. In addition, Li et al.¹⁸⁷ revealed that the dopants can also enhance the reactivity of surface lattice oxygen (O_L) in MOSs. The H₂ sensing response of Sn_0.8Ge_0.2O₂ (SGO) significantly surpasses that of pure SnO₂, with the LOD reaching as low as 50 ppb, as illustrated in Fig. 12d and e. According to the DFT calculations, the introduction of Ge induces lattice distortion, which in turn results in an increased concentration of O_V (Fig. 12f). Specifically, compared to pure SnO₂, the length of the Sn-O bond in SGO along both the a-axis and b-axis is elongated from 2.08 Å to 2.10 Å, while the length along the c-axis is changed to 2.05 Å. When viewed from the (110) direction, the bond angle of Sn-O-Sn changes from 180 ° to 162 °, thereby altering the electronic structure of SGO (Fig. 12g). Moreover, the oxygen escape energy in SGO is dramatically lower than that in SnO₂ (2.45 eV vs 3.26 eV). This finding implies that the O_L in SGO transitions more readily into chemisorbed oxygen species compared to its counterpart in SnO₂, as depicted in Fig. 12h. This difference is primarily derived from the variations in the positions of O 2p orbitals between these two samples, and a higher O 2p-band center (relative to E_f) thermodynamically facilitates the transformation from O_V into chemisorbed oxygen species (Fig. 12i-k). Since the SGO exhibits a comparable or even higher H₂ response value in Ar, this further provides experimental evidence for the involvement of O_L in H₂ sensing^45,187,195. Thus, the H₂ sensing mechanism of SGO-based sensors involves a four-step process (Fig. 12i): (ⅰ) H₂ molecules react with adsorbed oxygen species; (ⅱ) surface O_L transforms into chemisorbed oxygen species and generates O_V; (ⅲ) newly generated chemisorbed oxygen species continue to react with H₂ molecules; and (ⅳ) upon exchange to air, the O₂ molecules adsorb on the surface of SOG and undergo conversion back into surface O_L and chemically adsorbed oxygen. Furthermore, some nonmetal elements such as C, N, S and Si are often utilized as doping heteroatoms to enhance gas sensing response^{196,197,198,199}. Li et al.²⁰⁰ used waste honeycomb (HC) as a bio-template to fabricate N-doped TiO₂ with a hierarchical porous nanostructure for rapid and selective H₂ detection. In comparison to pure TiO₂ synthesized without HCs, the obtained N-TiO₂, which was calcinated under 600 °C, exhibited a higher concentration of O_V (Fig. 12m). Beneficial from the introduction of active sites, the sensor based on N-TiO₂-600 demonstrated a high response, rapid t_res/t_rec (8/3.8 s) at 250 °C for 1000 ppm H₂, as illustrated in Fig. 12n, o.

Fig. 12: The enhanced sensitivity of MOSs based H₂ sensors via heteroatom.

a SEM images of the SnO₂ and La-doped SnO₂ NFs; b Temperature-dependent response of SnO₂ and La-doped SnO₂ NFs to 100 ppm H₂ and C dynamic sensing response of SnO₂ and La-doped SnO₂ NFs to different H₂ concentrations¹⁹⁴. Copyright 2019, Elsevier. d Response of sensors based on SnO₂ and Ga-doped SnO₂ (SGO) to 500 ppm H₂ at different temperatures and e real-time sensing response of SGO to various H₂ concentrations; f Schematic crystal structure and g electron local function and the bader charge of SGO; h Escape energy of O atoms in SGO and SnO₂; The electronic density of states of I SGO and j SnO₂; k the relationship between the position of O p-band and the conversion from surface lattice oxygen to chemisorbed oxygen; l Schematic diagram of the mechanism of the gas-sensing reaction with lattice oxygen participation¹⁸⁷. Copyright 2024, Springer Nature. m O 1s of TiO₂-600 and TiO₂ without HC; n Response of four sensor at different temperatures toward 1000 ppm H₂; o Dynamic response-recovery curves of four sensors at 250 °C toward 1000 ppm²⁰⁰. Copyright 2025, Elsevier

The construction of heterojunctions by incorporating two or three types of MOSs is one of the most important strategies for enhancing sensor performance, such as selectivity and sensitivity. Heterojunctions can modulate the charge distribution and electronic structure of the host material, thereby further optimizing sensor performance via both tuning the resistance value and adjusting the concentration of adsorbed oxygen species. Upon establishing heterojunction interfaces, the E_f of two components equilibrate at the interface, resulting in electron transfer from the component with a higher E_f to that with a lower E_f. This process leads to the formation of an EDL. The increased interfacial potential barrier energy and synergistic surface reactions greatly contribute to improvements of sensing performance. In general, the heterojunctions can be classified into p-n, n-n, and p-p junctions, as well as complex heterojunctions (e.g., n-p-n, p-n-p), depending on the types of MOSs employed. Among these classifications, the construction of p-n heterojunctions has emerged as a widely used strategy for selective detection of H₂, such as n-SnO₂/p-Co₃O₄^201,202, n-In₂O₃/p-Co₃O₄²⁰³, n-WO₃/p-PdO^204,205, n-WO₃/p-CoO²⁰⁶, n-TiO₂/p-NiO²⁰⁷, and n-SnO₂/p-Cr₂O₃²⁰⁸. During the industrial H₂ production and TR process, CO not only serves as the interfering gas but also possesses similar chemical properties with H₂. Consequently, many researchers have aimed at enhancing the selectivity for H₂ in the presence of CO interference^201,208,209. In the realm of gas sensing, the adsorption energy (E_ads) is an essential index for evaluating the selectivity of the sensor. Thereby, the NiO (100) site supported TiO₂ (101) exhibits a good selectivity to H₂ since its E_ads to H₂ (−0.377 eV) is much more negative than CO (−0.022 eV) (Fig. 13a-e)²⁰⁷. The change in the heterojunction energy barrier after interaction with H₂ ($\Delta {E}_{total({H}_{2})}$) is significantly lower than that of $\Delta {E}_{total({O}_{2})}$, while ΔE_total(CO) remains nearly equivalent to $\Delta {E}_{total({O}_{2})}$ (Fig. 13b). Consequently, the TiO₂-NiO heterojunction exhibits an n-type sensing response to H₂ and exhibits no response to CO (Fig. 13c-e). Similarly, since the E_ads of H₂ (−2.38 eV) is more negative than that of CO (−1.9 eV) on the surface of n-SnO₂/p-Co₃O₄ composites, resulting to an outstanding anti-interference with an impressive H₂ to CO response ratio (S${H}_{2}$/S_CO) of 11.67 (Fig. 13f-h)²⁰¹. Additionally, both p-p and n-n heterojunctions also possess the capability to modulate the barrier energy at their heterointerface, several H₂ sensors with enhanced sensing response have been reported by using the MOSs with the same semiconductor type^210,211,212. Motaung et al.²¹² reported a highly sensitive H₂ sensor based on CeO₂-SnO₂ n-n heterojunctions, as illustrated in Fig. 13i-k. The response value of the n-n heterojunction is significantly higher than that of pure CeO₂ and pure SnO₂ owing to the presence of a large number of defects induced by heterojunctions. Moreover, instead of confining the search domain to construction of heterojunctions, the fabrication of ternary MOSs (TMOSs) represents a promising strategy for enhancing sensor performance^213,214. Further, numerous reports have demonstrated that TMOSs-based sensors exhibit superior sensing performance compared to binary MOSs. Owing to the presence of two differently sized cations in TMOSs with multiple oxidation states, these TMOSs are endowed with a higher density of interstitial defects and O_V, and a stronger surface adsorption of atmospheric oxygen. This ultimately leads to enhanced gas sensing performance. Figure 13l displays the SEM and AFM images of AgInO₂ film, which is deposited at RT for the first time²¹³. The AgInO₂-based sensor exhibited high sensitivity and excellent selectivity toward H₂, achieving an impressive ratio of H₂ to NH₃ response ratio (S${H}_{2}$/S${{NH}}_{3}$) of ~7.5, as displayed in Fig. 13m, n.

Fig. 13: The influence of MOS heterojunction design on H₂ sensor selectivity.

a Differential charge density of NiO/TiO₂ and TiO₂/NiO heterojunctions after absorption of O₂-H₂ and O₂-CO; b Gas selectivity mechanism of the NiO-TiO₂ heterojunction in the presence of Air, H₂, and CO; c cycle response curves of NiO/TiO₂ to 400 ppm of CO, H₂, and the mixture of H₂ and CO²⁰⁷. Copyright 2023, American Chemical Society. f TEM images of SnO₂-Co₃O₄ nanocomposite; g, h Gas selectivity of four samples to different gases at 325 °C²⁰¹. Copyright 2025, Elsevier. I TEM images of SnO₂, CeO₂, and SnO₂-CeO₂; j Response curves of the CeO₂, SnO₂, CeO₂-SnO₂ exposed to different concentrations of H₂ at 300 °C; b dynamic response curves of CeO₂-SnO₂ sensor to various H₂ concentrations at different temperatures²¹². Copyright 2018, Elsevier. l SEM and AFM images of AgInO₂ thin film; m Selectivity of AgInO₂ thin film towards 100 ppm of various gases at 360 °C; n Calibration plot of AgInO₂ thin film towards different concentrations of H₂ at 360 °C²¹³. Copyright 2023, Elsevier

Rational design of MOS heterostructures demonstrates enhanced sensitivity, excellent selectivity, ppb level LOD, and acceptable t_res/t_rec for H₂ sensors, making them suitable for detection of trace H₂ in early stage of TR. Moreover, since p-MOSs have a higher tolerance toward humidity than n-MOSs, it is possible to design humidity-resistance H₂ sensor for practical monitoring LIB safety by optimizing the p-n or p-p MOS heterostructures^181,203.

H₂ sensors based on noble metal decorated MOSs

Noble metals are widely recognized for their exceptional catalytic properties. They can effectively reduce the E_ads of gases on the surface of sensing materials, thereby enhancing their performance. Additionally, the modification of MOSs with noble metal such as Pt, Pd and Au introduces more active sites on the MOSs surface, facilitating the preferential adsorption of target gases^215,216,217. Furthermore, due to their inherent catalytic characteristics, noble metals provide reaction pathways for H₂ dissociation, which can lower the activation energy and subsequently enhance the response speed, sensitivity, selectivity and reliability of the sensors²¹⁸

Moreover, the noble metals also create spillover effect in H₂ sensing when deposited on the surface of MOSs. For instance, in the context of H₂ sensing, the spillover effect refers to a process where H₂ molecules absorb onto the surface of noble metals and are subsequently dissociate into H atoms. To elucidate the mechanism underlying this spillover reaction, operando techniques are employed to reveal the dynamic reaction process, such as in-situ Raman²¹⁹, in-situ infrared spectroscopy (IR)²²⁰, in-situ electron energy loss spectroscopy^220,221, and in-situ X-ray photoelectron spectroscopy (XPS)^222,223. Li et al.²²³ used Pt-γ-WO₃ as a model sample and employed near ambient pressure XPS (NAP XPS), DFT calculations, and microkinetic model to study the dynamic evolution of surface states, during H₂ sensing, as illustrated in Fig. 14a, b. The activation barrier height (E_a) of H₂ dissociation on the surface of Pt-γ-WO₃ is much lower than that on the surface of pure WO₃. As a result, H₂ preferentially dissociates on the surface of Pt-γ-WO₃. W 4f and O 1s spectra from Pt-γ-WO₃ were collected at RT under 5 mTorr H₂ gas (Fig. 14b). The NAP XPS analysis demonstrated several dynamic changes when exposed to H₂, including the presence of W⁵⁺, the formation of adsorbed water (H₂O_ads), and an increase in undercoordinated oxygen species, revealing that H₂ spillover through proton-coupled electron transfer. By studying the effect of temperature via NAP XPS and theoretical calculations, Li et al.²²³ concluded that H₂ spillover involves multiple pathways after dissociation into H atoms on the surface of Pt: (Ⅰ) at low temperatures (25–50 °C), H inserts into the bulk oxide lattice (H_bulk); (Ⅱ) reverse H spillover is observed when the temperature increases to 50–150 °C; and (Ⅲ) an accumulation of O_V is detected due to the facilitation of H₂O_ads desorption with further increases in temperature (150–400 °C) (Fig. 14c). Thus, the incorporation of noble metals reduces the reaction activation energy of MOSs, leading to faster t_res/t_rec and lower operation temperature^224,225,226. Meng et al.²²⁷ reported that the operation temperature of SnO₂ decreased from 300 °C to 125 °C after Pd loading. Figure 14d demonstrates the TEM image of Pd/SnO₂. Figure 14e shows that the sensing response of 0.5 at% Pd/SnO₂ to 5000 ppm achieved a value of 2727, which is approximately 47.15 times greater than that of SnO₂. Figure 14f, g presents a RT H₂ sensor based on Pd/PdO₂ composites that achieves detection of 4% H₂ within ~1 s ascribing to the rapid reduction kinetics associated with PdO₂²²⁸. The fabrication of Pd/PdO_x composites was conducted via an electrophoretic deposition method, during which the Pd⁰ content gradually decreased from 43% to 29% as the electrode potential increased from 1 to 10 V. The presence of metallic Pd allows rapid dissociation of H₂ into adsorbed H atoms and coupled with ultrafast reduction kinetics of PdO_x, enabling 1 s detection of 4% H₂ at RT. However, it is important to note that since the reduction process for PdO_x is irreversible at RT, the Pd/PdO_x composites do not recover after exposure to H₂. In recent years, to further improve the catalytic effect of noble metals, atomically dispersed single-atom catalysts (SACs) have been proposed to maximize the exposure of metal atoms and enhance their utilization efficiency^229,230. To highlight the unique role of SACs, the sensing performance of Pd SA/Co₃O₄ and Pd NPs/Co₃O₄ were studied to provide a deeper insight and understanding of the effect of Pd SACs (Fig. 14h-j)²³¹. Different concentrations of Pd SAs were obtained by increasing the at% of Pd/Co from 1 to 5%. The sensing response of Pd/Co₃O₄ to H₂ gradually increased with rising concentrations of Pd SAs (1–5% Pd). As the Pd/Co ratio was further increased (7.5–20% Pd), Pd NPs formed due to the aggregation process and the response value of sensor subsequently decreased. Because all of the Pd SAs can serve as electron doners compared with Pd NPs to sensitize the sensor performance. However, the high surface energy of SACs presents a significant challenge in fabrication of SACs on sensing materials with high stability²³². Thus, defect engineering has been proposed as a strategy to stabilize SAs, given that both the unsaturated coordination conditions and the high diffusion barrier at the defects can facilitate anchoring of SAs^233,234. By construction of O_V, Pt SAs were steadily anchored on the surface of Fe₂O₃ NSs. enabling ultrafast H₂ sensing (t_res = ~2 s), as shown in Fig. 14k-n²³⁵. According to DFT results, Pt SAs are positioned in the vacancy position sites and coordinated with the adjacent Fe atoms. The Pt-Fe spacing in the Pt-Fe₂O₃-O_V is smaller than that in Pt-Fe₂O₃, indicating enhanced stability of the former structure (Fig. 14l). Consequently, the attenuation rate of the sensor based on Pt-Fe₂O₃-O_V was merely 7.5% over a testing period of 30 days, significantly lower than the 15.1% observed in the Pt-Fe₂O₃-based sensor (Fig. 14n). In addition, owing to the lower E_ads of H₂ on the surface of Pt-Fe₂O₃-O_V, this configuration demonstrates higher sensitivity towards H₂ while maintaining improved stability. At the device level, Guo et al.²³⁶ developed Pd-catalyzed dual-gate TeSeO FET H₂ sensors, achieving a theoretical LOD of 35 ppb. The dual-gate modulation offers practical strategies for the rational design of high-performance H₂ sensors based on noble metal-decorated MOSs.

Fig. 14: Noble metal mechanisms for enhanced H₂ sensing in MOS devices.

a Schematic H₂ dissociation on the monoclinic WO₃ (001) surface w/wo Pt metal cluster based on DFT results; b Operando monitoring of H₂ spillover on WO₃ surface at RT using NAP XPS; c Schematic illustration of the reaction mechanism at varied temperatures²²³. Copyright 2025, American Chemical Society. d TEM image of 0.5 at.% Pd/SnO₂ NPs and e the dynamic response curves of the SnO₂ sensors with various Pd concentration ratios exposure to 100-10000 ppm H₂²²⁷. Copyright 2022, Elsevier. f Schematic preparation of Pd/PdO_x sensors by electrophoretic deposition and the SEM image of prepared Pd/PdO_x; g Dynamic sensing curves of Pd/PdO_x prepared under different voltages to 4% H₂²²⁸. Copyright 2023, Spring Nature. h STEM images of Co₃O₄ and Pd/Co₃O₄ with different Pd contents; i, j Hydrogen sensing properties of NP films of Co₃O₄ and Pd-Co₃O₄ with Pd contents of 1–20%²³¹. Copyright 2020, American Chemical Society. l STEM image of Pt-Fe₂O₃-O_V nanosheet; m DFT simulations of Pt-Fe₂O₃ and Pt-Fe₂O₃-O_V after adsorption of H₂ molecules; m Dynamic sensing curves and n long-term stability of sensors to H₂²³⁵. Copyright 2024, American Chemical Society

To further enhance the sensing performance of noble metal-modified MOSs, researchers preferentially design more complicated sensing materials by introducing bimetallic catalysts or constructing ternary heterojunctions^{237,238,239,240}. Bimetallic catalysts exhibit superior catalytic performance compared to their monometallic counterparts, primarily due to their synergistic effects. For instance, since Pd demonstrates excellent reactivity and durability to H₂ and Pt exhibits outstanding catalyst effect for both hydrogenation and oxygen reaction, PdPt bimetallic NPs are usually utilized for highly sensitive H₂ detection^241,242. Zhou et al.²⁴³ reported a humidity immunity H₂ sensor based on PdRh-sensitized Fe₂O₃, the sensing response of PdRh/Fe₂O₃ (105.9) to 10 ppm H₂ at 230 °C is significantly higher than that of Fe₂O₃ (36.3), Rh/Fe₂O₃ (64), and Pd/Fe₂O₃ (89.4). The enhancement mechanism can be attributed to the fact that Pd promotes the adsorption and dissociation of H₂, while Rh improves the dissociation of adsorbed O₂. To date various bimetallic catalysts have been reported in MOS-based H₂ sensor such as PdCo²⁴⁴, PdPt²⁴⁵, PdAu^246,247, PdRu²⁴³, PdAg²⁴⁸, NiPt²⁴⁹, AgCu²⁵⁰, AuSn²⁵¹, and so on. However, MOS-based sensors typically operate at high temperatures, which greatly compromises their stability due to the deactivation or deterioration of noble NPs²⁵². To investigate the stability mechanism underlying noble metal catalysts, in-situ TEM is employed to capture the dynamic evolution of PdAg NPs under operational conditions, as illustrated in Fig. 15a-c²⁴⁸. Based on in-situ TEM observations, two failure mechanisms have been identified: coalescence of PdAg NPs at 300 °C and phase segregation at 500 °C. At 300 °C, a decrease in centroid-to-centroid distance between two adjacent NPs is noted, accompanied by atom migration until two NPs merge into a larger NP (Fig. 15b). When the operation temperature increases to 500 °C, the Ag preferentially segregates from the alloy phase due to its lower Tamman temperature (~345 °C) compared to Pd (640 °C) (Fig. 15c). To avoid particle coalescence and phase segregation, it is essential to maintain a low density of PdAg NPs on the surface of ZnO, with an operational temperature kept below the Tamman temperature of Ag. Furthermore, the elemental ratios of Pd/Ag must also be taken into account to optimize sensing performance. Therefore, the optimized PdAg/ZnO-based sensor demonstrates a high sensing response along with satisfactory long-term stability, as shown in Fig. 15d-g. Similar to MOS-based sensors with different exposed facts, the noble NPs with different exposed facets can also significantly influence sensor sensitivity. Meng et al.²⁵³ fabricated a series of PdPt/SnO₂-based H₂ sensors as shown in Fig. 15h. The sensing results illustrate that both the exposed facts of PdPt NPs and their content ratios have a considerable influence on sensor performance (Fig. 15i-l). The PdPt nano-octahedrons (NOs) modified SnO₂-based sensor shows the highest sensing response owing to the higher catalytic of (111) facet. As discussed above, both construction of MOSs heterojunctions and the decoration of MOSs with noble metals are two efficient strategies to promote the sensor performance. Therefore, researchers are committed to integrating these strategies together to fabricate ternary heterojunctions for H₂ sensing^239,240,254. Hu et al.²⁵⁵ reported a remarkably enhancement in sensing response toward H₂ by employing ternary heterojunction as the response order is Pd/ZnO-SnO₂ > ZnO-SnO₂ > SnO₂. Cai et al.²⁵⁴ presented a highly sensitive H₂ sensor based on Pd-NiO/SnO₂ with fast t_res/t_rec (1/4.9 s) to 1 ppm H₂. The improved sensing performance is mainly due to (Ⅰ) the hollow NiO/SnO₂ heterojunction provides sufficient reactive sites; (Ⅱ) kinetically promoted H₂ spillover effect by Pd NPs (Fig. 15m-o). Although sub-second t_res has been achieved in aforementioned works, the realization of comparably fast t_rec remains challenging. Xing et al.²⁵⁶ addressed this issue by developing a Pd-SnO₂/ZnO-based sensor that exhibits ultrafast recovery kinetics (Fig. 15p). The rapid sensing mechanism relies on redox reaction acceleration, catalytic promotion by noble metals, and synergistic heterojunction effects among PdO, SnO₂, and ZnO. The sensor demonstrates notable performance in sensitivity and selectivity, combined with sub-second t_res/t_rec of 0.8/0.8 s toward 50 ppm H₂, surpassing most reported H₂ sensors with similar structural configurations (Fig. 15q-s).

Fig. 15: High performance MOS based H₂ sensors via bimetallic catalyst modification or construction of ternary heterojunctions.

a Schematic showing the coalescence sintering of PdAg NPs under the gas cell; Morphological evolution of PdAg NPs b at 300 °C and c at 500 °C in air; d–g Sensing results of the ZnO-based H₂ sensor with PdAg NPs as the catalyst²⁴⁸. Copyright 2022, American Chemical Society. h TEM and EDS mapping images of PdPt NPs with different shapes and exposure facets; i–l Sensing results of the PdPt/SnO₂-based sensors²⁵³. Copyright 2023, Elsevier. m Schematic diagram of hollow nanoreactor driven kinetics; n, o Dynamic sensing results of four samples such as NiO/SnO₂ NPs, Pd-NiO/SnO₂ NPs, NiO/SnO₂ YS NPs, and Pd-NiO/SnO₂ YS NPs²⁵⁴. Copyright 2023, John Wiley and Sons. p SEM and TEM images of Pd-SnO₂/ZnO nanofibers; q–s Sensing results of Pd-SnO₂/ZnO-based H₂ sensors²⁵⁶. Copyright 2024, Elsevier

Therefore, these results confirm the effectiveness of constructing noble metal-modified MOSs could further enhance the overall sensor performance specifically sensitivity and t_res/t_rec (within 1 s) to fulfill the US DOE standard. We anticipate that the already impressive performance could be further enhanced, particularly in terms of reliability via material-level optimizations discussed in prior Sections “H₂ sensors based on single MOSs” and “Hetero MOS-based H₂ sensor”.

Table 3 displayed sensing properties of MOS-based H₂ sensors. To summarize, MOS-based H₂ sensors not only exhibit ultrahigh H₂ response with a low LOD but also have demonstrate high thermal and chemical stability. However, the majority of MOS-based sensors operate at elevated temperatures (over 200 °C) to facilitate H₂ reactions with chemisorbed oxygen species or the MOSs themselves. To lower the energy consumption, decoration of MOSs with noble metals and employing PHM method have been reported as effective strategies to lower the energy consumption. The other strategy is to employing materials with the capability to detect gases at RT such as carbon materials and 2D materials as discussed in the following sections.

Table 3 Summary of sensing properties of MOS-based H₂ sensors

Carbon material-based chemiresistive H₂ sensor

H₂ sensors based on pure carbon materials

Carbon-based materials have garnered significant attention as promising candidates for chemiresistive gas sensors due to their high specific surface area and tunable electrical properties^257,258,259. Carbon nanomaterials exhibit a wide range of structures (e.g., 0D nanodiamonds (NDs) and active nanocarbon NPs, 1D CNTs, 2D graphene (Gr) and graphdiyne (GDY), as well as 3D fullerenes) that arise from different bonding configurations such as sp, sp², or sp³ hybrid orbitals²⁶⁰. For sensors based on pure carbon materials, changes in the resistance of carbon materials can be measured, contributing to the electron transfer once exposed to target gases, as shown in Fig. 16. In inert atmospheres, H₂ directly donates electrons when adsorbed onto the surface of carbon materials. This causes a decrease/increase in resistance for n-type/p-type carbon materials²⁶¹. In air, H₂ reacts with oxygen species adsorbed on the surface of carbon materials and releases electrons into the conduction band of these materials, leading to variations in resistance. However, pristine carbon materials suffer from poor H₂ sensitivity due to weak adsorption of H₂ molecules on material surface^{262,263,264,265}. To activate sensing properties, several strategies have been employed, including increasing adsorption sites and utilization of external heat or light energy²⁶⁶.

Fig. 16: The fundamental H₂ sensing mechanism based on carbon materials.

Schematic diagram of H₂ adsorption and electron transfer on carbon materials with various dimensions

Construction of porous sensing materials is an effective strategy to provide more active adsorption sites²⁶⁷. Guo et al.²⁶⁸ modulated the density of CNT to enhance H₂ accessibility to the underlying active sites of porous CNT films, which optimized H₂ sensing performance. Besides, the application of carbon materials with inherent abundant adsorption sites, can enhance the gas sensing performance as well. For instance, N-doped nanodiamonds (N-NGs), consisting disordered carbon with sp²/sp³ bonds, possess plenty of grain boundaries and defects, can enhance H₂ sensing response when combined with other carbon materials^{264,269,270,271}. In addition, 2D porous carbon material-GDY, has aroused attention in gas sensing application^{272,273,274,275} owing to its uniform nanopore structure and highly reactive triple bonds, which exhibits excellent electron capture capability^276,277. Nam et al.²⁷⁸ reported a H₂ sensor based on hydrogen-substituted GDY (HsGDY) with rapid and reversible H₂ sensing. Figure 17a depicts the schematic diagram of HsGDY structure and its synthetic route. Different from other 2D materials such as Gr, transition metal dichalcogenides (TMDs), and MXenes, the HsGDY with uniform 1.63 Å nanopores facilitate the diffusion of target molecules through these pores, thereby providing a significantly accelerated reaction pathway. Additionally, sp carbon in alkynyl linkage can serve as an effective binding site for the molecules. The HsGDY-based H₂ sensor showed a wide detection range (0.1–10,000 ppm) with good reproducibility (n = 3 experimental replicates), as displayed in Fig. 17b. Compared with conventional 2D materials such as GO, MoS₂ and MXene, the HsGDY showed around 2 orders higher sensing response without any baseline drift and 4/6 times faster t_res/t_rec (8/38 s) to 1% H₂, as shown in Fig. 17c. Among previously reported 2D materials (before 2023), this is the maximum detection speed without any dopants or functionalization⁵². Both theoretical and experimental approaches were employed to study the sensing mechanism of HsGDY, as illustrated in Fig. 17d-f. Based on the DFT results, H₂ is observed to adsorb within the pocket sites formed by the horizontal slip between the A and B layers of bulk HsGDY, resulting in an expansion in the distance between two acetylenic bonds within the pocket (Fig. 17d). Furthermore, the chemisorption of H₂ would take place with each H atom chemisorbing to the top and bottom acetylene group of the identified adsorption site (Fig. 17e). This chemisorb interactions between acetylenic bonds and H atoms were verified by in-situ Raman characterization in Fig. 17f.

Fig. 17: H₂ sensors based on pure carbon materials.

a Schematic illustration of the molecular sensing behavior and synthesis process of HsGDY; b Dynamic sensing response of HsGDY-based sensors to 0.1–10,000 ppm H₂; c Real-time response of HsGDY, and three other 2D materials upon H₂ exposure (100–1000 ppm); d Side-view image of the physisorbed configuration of the H₂ molecule in the interstitial pocket site between layers of HsGDY. e Charge density difference plot of HsGDY and physisorbed H₂ molecule; f In situ Raman spectra of the HsGDY film under N₂ and H₂ exposure²⁷⁸. Copyright 2023, American Chemical Society. g SEM image of Gr-based sensor; the effect of UV irradiation on h the surface potential and i sensing response of Gr²⁹³. Copyright 2025, Elsevier

In addition to designing sensing materials characterized by a porous structure and abundant defect sites, which significantly enhance the number of active sites, the application of external energy (e.g., light and heat) is regarded as another effective strategy for improving sensor performance^{279,280,281,282,283}. Specifically, elevating the temperature can optimize sensor response by activating the surface for chemisorption of target gas molecules²⁸⁴. Also, it ensures the complete desorption of gas molecules and accelerates the recovery process^{285,286,287,288}. Park et al.²⁸⁸ used polystyrene substrate as a template to create a wrinkled structure in multi-wall CNT (MWCNT) films via a thermal shrinkage effect. By increasing the applied voltage via electrothermal conversion, the operating temperature of the MWCNT films was raised, resulting in an improved sensing response to 10% H₂. However, additional heat supply increases power consumption and may bring about safety hazards when detecting flammable explosive H₂^289,290,291. Applying light irradiation is a reliable alternative to enhance sensing ability which can modulate the concentration of photocarriers to promote charge transfer process, and provide more active sites for gas adsorption²⁹². Tang et al.²⁹³ applied UV irradiation to achieve the detection of H₂ down to 5 ppm based on pure Gr without modification (the sensor structure is shown in Fig. 17j). Prior to this, Gr alone was rarely reported as a H₂ sensor due to its lack of surface defects or functional groups, which limited gas adsorption^294,295. In this case, UV light promotes the desorption of the pre-adsorbed water and oxygen on Gr surface, while also generating photogenerated carriers that enhance H₂ sensing capabilities²⁹³. Moreover, the change in surface potential illustrated in Fig. 17k revealed that UV light can alter electronic states of Gr into n-type doping, thereby accelerating desorption process. As shown in Fig. 17l, under illuminated conditions, the sensor exhibited a rapid and reversible response to 10 ppm H₂. Whereas, it showed no response in the absence of light, which indicates the critical role of UV light for proper sensing.

Despite the achievements in pure carbon materials for H₂ detection, most of them still suffer from unsatisfactory sensing performance owing to the weak physical adsorption on their surface. The limited intrinsic selectivity of pure carbon-based materials toward H₂ brings about challenges in the complex gas environment of TR process, ultimately compromising detection accuracy. Therefore, to increase sensor sensitivity and selectivity, various methods have been proposed, including surface treatments to introduce functional chemical groups, decoration of noble metal NPs and design of hybrid structure with MOSs.

H₂ sensors based on carbon materials with functional groups

During H₂ adsorption process, functional groups of carbon materials exert catalytic effects, promoting dissociation of H₂ into H atoms²⁹⁶. To date, surface treatment with liquid chemicals is a primary strategy to introducing of functional groups in CNT and Gr^{296,297,298,299,300}. For CNT, strong acids (e.g., HNO₃, H₂SO₄) and strong oxidants (e.g., H₂O₂) are generally used to introduce functional groups (e.g., -OH, -N-H-, -C=O-) to the side walls and ends of CNTs^301,302. Dhall et al.²⁹⁶ developed a H₂ gas sensor based on multiwall CNTs (MWCNTs) with amount of functional groups (-COOH, -OH, C-O) (Fig. 18a and b). For functionalized MWCNTs (F-MWCNTs), there exist many H₂ adsorption sites, including outer surface, external grooves, interstitial channels, and inner pores, as shown in Fig. 18c. Additionally, these functional groups act as catalysts that accelerate the adsorption/desorption rate^296,297. Compared with raw MWCNTs (R-MWCNTs), the F-MWCNTs showed greatly improved sensitivity for H₂ sensing, as illustrated in Fig. 18d. After acid treatment, the response value of sensor increased from ~1.5% to 8%, and the t_rec was shortened from 190s to 100s. In terms of Gr, its derivatives, such as graphene oxide (GO) and reduced graphene oxide (rGO), possess a wealth of functional groups that serve as active sites for gas adsorption^303,304. Figure 18e depicts a schematic diagram of an H₂ sensor based on GO NSs by using a porous PU foam as a substrate to form a dual-honeycomb structure. This porous device structure significantly enlarged the specific surface area and improved H₂ adsorption capacity. Figure 18f, g exhibits the concentration-dependent H₂-sensing performance and show that the sensor can detect H₂ in a range of 2–100%. Nevertheless, for effective warning of TR process, sensitive detection of low-concentration H₂ is critical³⁰⁵. Therefore, compared to GO, rGO is promising to achieve a lower LOD ascribing to existing abundant active defect sites, enhanced electrical conductivity, and Gr-like low noise features^306,307,308. Since the density of functional groups, defect sites and conductivity of the rGO greatly depends on its reducing degree, modulating the degree of reduction presents an effective approach for optimizing its gas sensing capability^{309,310,311,312}. To systematically investigate the effect of its reduction degree, Schipani et al.³⁰⁰ fabricated a series of rGO-based sensors by varying the thermal reducing temperatures, as displayed in Fig. 18h. Raman spectra presented in Fig. 18i reveals that an increase in treatment temperature leads to a higher degree of reduction, which correlates with an elevated content of sp² carbon within the rGO structure³⁰⁰. This reduction process provides a higher concentration of unsaturated carbon atoms that promote the current transport²⁹⁸. Figure 18j and k show that a higher degree of reduction in rGO enhances H₂ sensing response. Moreover, the higher degree of reduction lowered the content of oxygenated functional groups that may interact with H₂O^313,314,315, thereby improving the hydrophobic properties of rGO. The optimized rGO-based H₂ sensor exhibited a relatively good humidity tolerance, achieving a theoretical LOD down to ~2.5 ppm at high humidity^300,316. Since the humidity in the environment undergoes dynamic and irregular fluctuations, employing carbon materials with hydrophobic properties to design sensing materials may diminish the impact of humidity on sensors in certain contexts.

Fig. 18: Introduction of functional groups to improve sensing performance.

HR-TEM images of a P-MWCNTs and b F-MWCNTs; c Schematic diagram mechanism for H₂ adsorption; d Dynamic response curve of P-MWCNTs and F-MWCNTs film at different concentrations of H₂²⁹⁶. Copyright 2013, Elsevier. e Schematic illustration of the porous GO structure and its sensing mechanism; f, g Linear gas response to various H₂ concentration (2–100%)²⁹⁹. Copyright 2022, American Chemical Society. h Schematic graph of rGO structure and the thermal process to prepare rGO sensors; I Raman spectra of rGO samples annealed at different temperatures; Response to H₂ concentrations from 100–1000 ppm for rGO-based sensors j in the absence of humidity and k under 90% RH³⁰⁰. Copyright 2023, American Chemical Society

H₂ sensors based on carbon materials decorated with hetero-materials

Another strategy to improve H₂ sensitivity of carbon materials is to construct heterostructures, including decoration of noble metals and combination of MOSs. Catalytic metals such as Pd^{317,318,319,320,321,322,323,324}, Pt^{249,265,325,326} and Au³²⁷ have been decorated on carbon materials to fabricate high-performance H₂ sensors mainly through aqueous reducing methods and physical deposition techniques^{249,323,328,329,330,331}. In this system, carbon materials mainly act as channels for fast electron transportation as well as provide extra adsorption sites. Catalytic noble metals, such as Pd, Pt, Au etc., contribute to dissociation of H₂, and some of them can react with H₂ to form hydride as mentioned before^61,223,332. Since intrinsic CNT is not sensitive to H₂, Girma et al.³²⁴ fabricated Pd-SWCNTs films with Pd NPs evenly distributed on SWCNTs (Fig. 19a-c). The response curve in Fig. 19d revealed that the sensor achieved a wide H₂ detection range (0.002–2%) and a fast t_res/t_rec (10/3 s) to 1% H₂. The decoration of Pd NPs is critical to enhance H₂ sensing through reducing the adsorption energy of H₂ as well as lowering the activation energy of reactions between H₂ and adsorbed oxygen species³²⁴. For this kind of composites, the content of decorated noble NPs plays a significant role in H₂ sensing performance, as shown in Fig. 19e. On one hand, excessively thin Pd film fails to generate sufficient effective sites to interact with H₂. Conversely, excess surface coverage of Pd introduces additional Pd conduction pathways that are less sensitive to electron transfer than semiconducting channels^323,324,333. Moreover, to further enhance the sensing performance of noble metal-decorated carbon materials, bimetallic NPs are used to modify carbon materials such as NiPd³³⁴, PdPt³³⁵, NiPt^249,336, AuPt³³⁷. It has been reported that bimetallic catalysts can reduce the hysteresis of single noble metals and lower the adsorption energy of H₂^105,249,320, thus exhibit better sensitivity and reversibility than monometallic counterparts³³⁸.

Fig. 19: High performance carbon based H₂ sensors via noble metal decoration or construction of heterojunctions.

a Fabrication of Pd decorated SWCNT-based H₂ sensors and the corresponding b TEM and c AFM images; d Sensing responses toward various H₂ concentrations (2 ppm–1%); e Effect of the Pd thickness on sensor response under 1% H₂³²⁴. Copyright 2023, John Wiley and Sons. F Schematic illustration and TEM image of rGO-SnO₂-ZnO nanocomposites; g Dynamic resistance change curves and h the response versus H₂ concentration from 1 to 1000 ppm at 175 °C³⁴⁶. Copyright 2023, Elsevier. i Schematic graph of structure and sensing mechanism of Pd-doped rGO/ZnO-SnO₂ nanocomposite; j Dynamic response curves of samples towards 50–500 ppm H₂ at 380 °C (NC0, NC1, NC2, NC3, NC4, and NC5 represent samples with the weight ratios (wt%) of the GO and ZnO-SnO₂ being 0, 1, 2, 3, 4, and 5 wt%, respectively); k Response transients of the sensor to 50 ppb H₂³⁵². Copyright 2022, Springer Nature

The second method to promote H₂ sensing ability is the addition of MOSs to carbon materials. MOSs have received widespread attention in the development of commercial gas sensors in regard of its simplicity of use and low cost, and acceptable sensitivity³³⁹. Incorporation of MOSs into carbon materials generates heterostructures and induces more adsorption sites, which facilitates the adsorption of H₂ molecules and promotes electron transfer process^207,340. Thus, the design of MOSs nanostructures in the hybrid is critical for modulating H₂ sensing performance^{98,339,340,341,342,343,344,345}. Li et al.³⁴⁶ fabricated rGO-SnO₂-ZnO nanocomposites, where 1D ZnO was distributed on 2D rGO-SnO₂ film to form heterostructures, as shown in Fig. 19f. The construction of heterojunction interfaces results in a greater variation in resistance decrease upon H₂ exposure and thus amplify the sensing response³⁴⁶. The fabricated rGO-SnO₂-ZnO-based H₂ sensor showed a wide detection range of 1–1000 ppm (Fig. 19g, h) and with a high response value of 19 to 500 ppm H₂, which surpassed the sensors based on rGO-SnO₂ (10.5) and rGO-ZnO (5.6).

Furthermore, the construction of multi-component systems, by incorporation of both noble metals and MOSs into carbon materials, can further improve H₂ sensing performance^{347,348,349,350,351,352,353}. Typically, Zhang et al.³⁵² synthesized Pd-decorated rGO/ZnO-SnO₂ to fabricate H₂ sensors with good sensing properties (Fig. 19i). In this hybrid system, the high surface area of rGO facilitated H₂ adsorption and the formed a p-n-n heterojunction of rGO/ZnO-SnO₂ could amplify the changes in resistance, leading to enhanced sensing response, as shown in Fig. 19j³⁵⁴. Moreover, the selective reaction of Pd with H₂, coupled with its spillover effect, greatly contributed to the high sensitivity and accelerated response/recovery rate. From the sensing results in Fig. 19j, k, the Pd-decorated rGO/ZnO-SnO₂ nanocomposites can detect H₂ concentration from 0.05 to 500 ppm at 380 °C and shows the high response of 9.4 to 100 ppm H₂ within few seconds (t_res/t_rec = 4/8 s). The greatly enhanced sensor performance can be attributed to synergetic effects including good electronic properties and high specific surface area of rGO, electron transfer modulation of formed heterojunctions, as well as catalytic effects of Pd NPs. This highlights the importance of constructing hierarchical structures for H₂ sensing applications. The functionalization with noble metals and MOSs enhances the material’s selective response to H₂, as well as yields a remarkable improvement in sensitivity by orders of magnitude, pushing LOD to the ppb level. It also enhances the sensor’s selectivity toward H₂, which is crucial for reliable detection against gas interference inside battery packs. The resulting sensors achieved a rapid t_res close to 1 s, nearly satisfying the US DOE benchmark. This strategy of constructing heterostructures establishes a robust platform facilitating adaptive sensing capabilities and ensuring battery safety.

Sensing properties of carbon-based H₂ sensors have been summarized in Table 4. In general, carbon-based materials, such as CNTs and Gr, featured with high specific surface area, high carrier mobility and low operation temperature have been regarded as promising candidates for RT gas sensing applications. However, most pristine carbon materials suffer from poor sensitivity to H₂ because of weak adsorption. Although construction of carbo-based composites can lower the LOD and enhance their sensitivity, high operation temperature is still required for composites containing MOSs in most cases.

Table 4 Summary of sensing properties of carbon-based H₂ sensors

2D material-based chemiresistive H₂ sensor

H₂ sensors based on pure 2D materials

With the advancement of nanomaterial technology, 2D van der Waals (vdW) materials are attracting attention in gas sensing realm regarding to their high surface-to-volume ratio, unique electrical and optical properties³⁵⁵. These features facilitate electron transfer via molecular adsorption at RT. 2D materials, especially TMDs such as MoS₂, MoSe₂, WS₂, and SnS₂, have been widely investigated as sensing materials for H₂ detection^{356,357,358,359,360,361}. However, bare TMDs-based H₂ gas sensors suffered from long t_res/t_rec and low sensitivity. To address these problems, several strategies have been proposed including design of porous nanostructures, modulation of vacancies, and construction of heterojunctions. The design of porous nanostructures, such as self-assembly of TMDs NSs to form hollow tubular, spherical and aerogel nanostructures, presents an effective method to enhance the exposed area of 2D materials and provide efficient pathways for gas diffusion^362,363,364. Apart from porous nanostructures, Agrawal et al.³⁶⁵ reported a pyramid MoS₂ structure for H₂ sensing, which was formed by layer to layer growth of monolayer MoS₂, as shown in Fig. 20a. The sensing performance of pyramid MoS₂ is exhibited in Fig. 20b, c. Compared with traditional MOS-based and other bare MoS₂-based H₂ sensor reported in literatures, the fabricated sensor showed superior performance by exhibiting a high sensitivity (69.1%) and a short t_res (32.9 s)^366,367,368. Because this unique stacked in-plane monolayer of MoS₂ pyramid increases favorable adsorption sites on MoS₂, including top of hexagon (H), top of Mo atoms (T_M) and top of S atoms (T_S)³⁶⁹. Furthermore, modulation of vacancies is also an promising approach to increase H₂ sensitivity^370,371. For TMDs, vacancies like S vacancies, Se vacancies serve as the major gas adsorption centers owing to their high catalytic activity^{372,373,374,375}. They tend to induce greater electron transfer and stronger interaction with target gases according to theoretical calculations^376,377. Rezende et al.³⁷⁰ investigated the role of S vacancies in MoS₂ for H₂ sensing and the relevant optical image of the MoS₂-based H₂ sensor is shown in Fig. 20d. The concentration of S vacancies was precisely regulated through the deposition of Al₂O₃ on MoS₂. This is attributed to the preferential growth of Al₂O₃ on surfaces with localized defects, which effectively passivates the S vacancies. Figure 20e presents the H₂ sensing mechanism, the dissociative adsorption of H₂ is facilitated by the catalytic effect of S vacancies^378,379. In Fig. 20f, the MoS₂-based sensor before Al₂O₃ deposition displayed obvious responses to H₂ from 0.5% to 50%. While after Al₂O₃ deposition, the sensor exhibited lower sensitivity and a longer t_res/t_rec (Fig. 20g), ascribing to the reduced S vacancies. Another method to improve H₂ diffusion and absorption is hybridizing TMDs to form heterojunctions, since EDL formed at the hetero-interface can tailor the electron transfer and actively tune the sensing performance^380,381. Kalita et al.³⁸² synthesized MoSe₂-WSe₂ NSs for H₂ sensing through a liquid phase exfoliation method. As shown in Fig. 20h, i, during the exfoliation process, functional groups, vacancies, and structural defects are formed within the NSs of MoSe₂-WSe₂, leading to increased active sites for gas adsorption³⁶⁰. Compared to pure WSe₂, the formation of a built-in potential at the hetero-interface between WSe₂ and MoSe₂ facilitates adsorption of oxygen onto the surface of MoSe₂-WSe₂, promoting the reaction with H₂ and electron transfer (Fig. 20i, j)³⁸². As illustrated in Fig. 20k, the MoSe₂-WSe₂-based showed a response of 59.57% to 25 ppm H₂, which is higher than that of WSe₂-(~38%) based sensors. Furthermore, owing to the enhanced efficiency of electron transfer, the t_res of MoSe₂-WSe₂-based sensor obviously shorted (Fig. 20l). However, sensors based on pristine 2D materials face limitations in detecting low H₂ concentrations and exhibit cross-sensitivity to polar gases (e.g., NH₃ and NO₂)^45,281,383. Consequently, achieving highly sensitive and selective H₂ sensing with pure 2D materials remains challenging, which calls for effective methods, such as combining with noble metals or MOSs to improve their H₂ sensing properties.

Fig. 20: Different approaches for enhancing H₂ sensing response of pristine 2D materials.

a Schematic illustration of adsorption sites and interaction with H₂ on MoS₂ pyramids; b Dynamic response for different concentrations of H₂ and c response time of 1% H₂ at 150 °C³⁶⁵. Copyright 2020, Elsevier. d Optical image of the MoS₂ sensor and e representation of H₂ adsorption on V_s; f Real-time current change for different H₂ concentrations (0.5–50%); g Sensor response toward 20% H₂ of MoS₂ before and after Al₂O₃ adsorption³⁷⁰. Copyright 2019, John Wiley and Sons. Schematic diagram of h MoSe₂ and WSe₂, and i MoSe₂-WSe₂ nanocomposites; j Sensing mechanism of MoSe₂-WSe₂; k Correlation curve between sensor response and gas concentrations (5–25 ppm H₂) of MoSe₂-WSe₂; l Response time curve of MoSe₂-WSe₂ and WSe₂³⁸². Copyright 2025, Elsevier

Noble metal decorated 2D materials

To increase H₂ selectivity, decoration of TMDs with noble metal NPs (Pd^{359,384,385,386,387,388,389,390,391,392,393}, Pt^{358,394,395,396,397}, Au^59,391, Ce³⁹⁸ etc.) or metal alloys^399,400 is one common approach due to their catalytic properties and spillover effects^389,395,397. Suh et al.³⁹¹ have found that modifying the MoS₂ surface with Pd leads to improved H₂ sensing performance. The underlying mechanism involves significant charge transfer upon H₂ exposure, facilitated by the spillover effect and the formation of PdH_x. which reduces the hole concentration and increases the resistance of the p-type MoS₂. Kim et al.⁵⁹ decorated different noble metal NPs on MoS₂ flakes (Fig. 21a), and investigated their respective role on the selectivity change by experimential and theoretical methods. Through DFT theoretical calculations, they observed a significant increase in binding energy for Pd (2.47 eV) and Pt (1.94 eV) when compared to pristine MoS₂ (0.15 eV), with reactive adsorption oxygen on the surface identified as the favorable binding sites (Fig. 21b, c). This finding is consistent with the experimental results in Fig. 21d. Therefore, apart from facilitating electron transfer, the decoration of Pd, Pt provides strong chemisorption sites for enhanced H₂ adsorption. Many reports have also observed similar performance improvements by noble metal decoration^{386,389,391,392,394}. Moreover, assembling 2D materials into 3D hollow structures can further improve sensing response^{358,359,366,396}. For instance, Park et al.³⁹⁶, designed Pt-decorated ultrathin MoS₂ hollow spheres, resulting in a significant enhancement of the H₂ sensing performance, including response value and t_res/t_rec, compared with non-hollow structure. Wu et al.³⁸⁴ synthesized self-assembled MoS₂ NSs with Pd decoration and investigated the H₂ sensing performance. Figure 21e illustrates the evolution of MoS₂ nanoflowers with varying concentrations of Pd NPs. In the case of Pd (1:1)-MoS₂, Pd NPs were deposited along the edges of MoS₂ nanosheets while preserving their porous structure. Conversely, in the Pd (2:1)-MoS₂ composites, the high coverage of Pd NPs resulted in a reduction in material porosity, consequently leading to fewer accessible active sites. Thereby, the Pd (1:1)-MoS₂-based sensor showed the highest sensing response of 8.7 towards 100 ppm H₂, attributed to the catalytic effect of Pd NPs and the preservation of its porous structure (as illustrated in Fig. 21f, g). Similar to carbon materials, utilization of external heat or light energy during the sensing measurement is also an effective approach to improve the sensor performance^{279,280,281,282,283}. Mai et al.³⁹⁰ reported a light-induced H₂ sensor based on monolayer MoS₂ with Pd nanoclusters evenly distributed on MoS₂ surface (Fig. 21h). The sensing mechanism in Fig. 21i revealed that light irradiation generated additional charge carriers and reduced the background resistance which in turn amplified the sensing responsivity³⁹⁰. The sensor performance was studied under dark and light conditions at different working temperatures. Figure 21j, k demonstrate that sensing response is greatly enhanced under light illumination. Moreover, as the temperature increased from RT (Fig. 21g) to 100 °C (Fig. 21h), t_res was shortened from 351 s to 47 s. Therefore, the results verified the positive role of heat input and light irradiation in terms of the enhancement of sensor performance. Although decoration of noble metals and regulation of test conditions could enhance the sensitivity of 2D materials toward H₂, the adsorption/desorption kinetics remain relatively slow (hundred seconds of t_res/t_rec) that poses challenges for real-time monitoring. Thus, further methods are needed to address these hurdles to facilitate early failure detection in LIBs.

Fig. 21: Decoration of noble metal catalysts for improved H₂ sensitivity of 2D material based sensors.

a SEM image of pristine MoS₂ and MoS₂ decorated with Pd, Pt, Au NPs; b Binding structure of H₂ molecule on different noble metal NPs; c H₂ binding energy and d the corresponding sensing values on various systems⁵⁹. Copyright 2023, American Chemical Society. e SEM images of Pd (1:1)- MoS₂, and Pd (2:1)-MoS₂ at 200 nm scale; f Gas sensing response of MoS₂ sensors with varying concentrations of Pd NPs: g Dynamic response curve for different H₂ concentrations (25–100 ppm) of Pd (1:1)-MoS₂³⁸⁴. Copyright 2025, Elsevier. h High-magnification TEM image and I light activated sensing mechanism of MoS₂ decorated with Pd nanoclusters; Sensing response under dark and light conditions at j 25 °C and k 100 °C³⁹⁰. Copyright 2021, American Chemical Society

H₂ sensor based on MOSs modified 2D materials

Another promising strategy to enhance the sensing performance of 2D materials involves hybridizing them with MOSs to form heterojunctions. The gas sensing performance of these heterostructures can be tailored by controlling their composition, architecture, and surface properties. Yang et al.⁴⁰¹ utilized a hydrothermal method to prepare 2D MoS₂-decorated Zn-doped MoO₃ nanoribbons and the SEM/TEM images are shown in Fig. 22a. The composites showed enhanced sensitivity with a wide H₂ detection range (5–1500 ppm) and a good linear response (Fig. 22b, c). Beyond binary composites, the construction of more complex hybrids, such as synthesis of ternary composites and functionalization of noble metals, can further enhance the gas sensing response. Bai et al.⁴⁰² prepared the heterostructural CdS/PbS/SnO₂ composites which exhibited a response of 1125.2% to 100 ppm H₂ that is an order of magnitude greater than that of CdS/SnO₂ and PbS/SnO₂ at 200 °C. Meng et al.³⁶¹ used the hydrothermal method to prepare SnS₂/SnO₂ nanocomposites, with agglomerated SnO₂ NPs wrapped in the SnS₂ flakes, as shown in Fig. 22d. The composite was later modified with Pd NPs and demonstrated improved H₂ sensing performance due to Pd decoration by promoting the adsorption/dissociation of H₂ molecules. Figure 22e shows that 1.0 At. % Pd/SnS₂/SnO₂ sensor exhibited high response (95) and rapid t_res/t_rec (1/9 s) to 500 ppm H₂ at 300 °C. However, as shown in Fig. 22f, the response of the Pd/SnS₂/SnO₂ composites exhibited a slight fluctuating downward trend over 30 days. This instability may be due to partial oxidation of the 2D materials at high temperature under air atmosphere⁴⁰³. Furthermore, even at mild operating temperatures, long-term signal drift problems still occur due to slow oxidation of TMDs^356,404. Therefore, for TMDs-MOSs composites with direct exposure of the TMDs, often fail to maintain good long-term stability³⁶¹. In contrast, in-situ oxidation can form a dense of passivating MOS layer on the surface of 2D material and thereby inhibiting further oxidation of 2D material.³⁵⁶. Moreover, by adjusting oxidation times and temperatures, the thickness of MOS layer can be tuned to match the Debye length for improvement of sensing performance^405,406. Various types of MOS/TMDs heterostructures have been synthesized including MoO₃/MoS₂³⁸⁷, WO₃/WS₂^356,407, SnO₂/SnSe₂⁴⁰⁸, and In₂O₃/In₂Se₃⁴⁰⁹. For instance, Paolucci et al.⁴⁰⁸ obtained the amorphous SnO₂/SnSe₂ heterostructure by partially oxidation of SnSe₂, with the inner SnSe₂ uniformly covered by a self-terminating SnO₂ layer. After annealing the SnSe₂ in air at 200 °C for 48 h, a uniformly amorphous SnO₂ layer was formed on the surface of SnSe₂, as shown in HRTEM images (Fig. 22g). Attributing to the passivation effect of the formed SnO₂ layer, the inner SnSe₂ was prevented from further oxidation even when extending the annealing time to 170 h 220 °C (Fig. 22g). The obtained SnO₂/SnSe₂-based H₂ sensor exhibited a low LOD (5 ppm) and a short t_res/t_rec (3/19 s), as shown in Fig. 22b. Due to the chemical stability of surface SnO₂, the baseline resistance of the SnO₂/SnSe₂ showed no substantial differences as the operating temperature was cycled from RT to 150 °C and reversed back to RT (Fig. 22i). The response of the SnO₂/SnSe₂-based sensor to 100 ppm H₂ remained almost the same over a period of one year (Fig. 22j). The good long-term stability of In₂O₃/In₂Se₃ materials prepared via in-situ oxidation is also observed⁴⁰⁹. Figure 22k and l reveal the sensing properties of In₂O₃/In₂Se₃ and its good reproducibility of baseline resistance at different working temperatures even after one year, which further proves that the oxide skin layer effectively passivated the underlying TMD layer from spontaneous degradation⁴⁰⁹. Therefore, compared to merely construction MOS/TMD composites, the introduction of MOS passivation layer can significant extent the device lifetime and provide robust operation across an RH range of 0–80%. By further functionalization with noble metals, the detection of H₂ can be cut down to only 1 s, satisfying the requirements of US DOE for practical application.

Fig. 22: Construction of heterojunctions to improve stability in 2D material based H₂ sensors.

a TEM image of 2D MoS₂-decorated MoO₃; b The relationships among the sensor response, the BET specific surface area and the content of the adsorptive oxygen; c Sensing response towards 5–1800 ppm H₂⁴⁰¹. Copyright 2022, Elsevier. d SEM images of SnO₂, SnS₂/SnO₂ and Pd/SnS₂/SnO₂; e Response curves of Pd/SnS₂/SnO₂ towards 100–10,000 ppm H₂; f Long-term stability of 1.0 at% Pd/SnS₂/SnO₂ to 5000 ppm H₂ at 300 °C³⁶¹. Copyright 2022, Elsevier. g HRTEM images and related SAED patterns of the 48 h and 170 h annealed SnSe₂ flakes in air at 200 °C; h Dynamic electrical responses of SnO₂/SnSe₂ at 100 °C to H₂ (5–100 ppm); i Baseline resistances change of SnO₂/SnSe₂ with different annealing times by modulating the working temperature in 25–150–25 °C range; j Long-term stability of SnO₂/SnSe₂ sample to100 ppm H₂ over one year⁴⁰⁸. Copyright 2022, Elsevier. k Dynamic electrical responses of In₂O₃/In₂Se₃ at 100 °C to H₂ (5–100 ppm); l Baseline resistance variations of In₂O₃/In₂Se₃ by modulating the working temperature in 25–150–25 °C range over different time periods⁴⁰⁹. Copyright 2023, American Chemical Society

2D MXene-based H₂ sensors

MXene is a class of carbide, nitride and carbon nitride 2D materials and has attracted wide attention due to its unique physiochemical properties and high electrical conductivity^410,411,412. Generally, MXenes are prepared by selective etching of metal atoms of MAX phases in fluoride-contained acidic solution, leading to formation of -O, -OH, and/or -F functional groups^413,414. Its chemical formula can be expressed as M_n+1X_nT_x, where M is the transition metal, X is carbon and/or nitrogen, T_x represents the surface groups. Different to most 2D materials, such as Gr, WS₂, and MoS₂, the abundance of functional groups present on the surface of 2D MXenes facilitates an increase in interlayer spacing within assembled MXene NSs. This expansion enhances the active surface area and creates nanosized interlayer channels that allow analytes to diffuse between the layers of stacked MXenes⁴¹⁵. Lee et al.⁴¹⁶, synthesized single-/few-layer 2D V₂CT_x by etching V₂AlC MAX in 50% HF solution for 92 h at RT, as shown in Fig. 23a. After etching process, the interlayer space of V₂CT_x expands and the surface of it is terminated with -O, -OH and -F functional groups. The presence of the oxygen containing functional groups, indicating partial surface oxidation of V₂CT_x into VO_x. Attributing to a higher selectivity of vanadium oxides (VO₂, V₂O₅) and V-doped MOSs toward H₂^417,418,419, the V₂CT_x-based sensor demonstrated a significantly higher response to H₂ compared to other gases (Fig. 23b). The sensor also achieved detection of H₂ from 2 to 100 ppm with good linearity at RT under ambient humidity (Fig. 23c). Similar to other materials, modification of noble metal NPs and hybridization of with MOSs are accessible approaches to improve response and recovery speed of MXene materials. Noble metal promotes the adsorption and dissociation of H₂, and some of them can selectively react with H₂ to form hydride as mentioned before^{417,420,421,422}. Moreover, it has been reported that bimetals can achieve higher reactivity than monometallic counterparts^110,249,338. Compared to bimetallic alloys, where the electrochemical potential of the alloy presents a compromise between the two constituent metals, researchers have reported that the spatially separated bimetals NPs exhibit enhanced catalytic activity in the realm of electrocatalysis^423,424. However, such separated bimetals have been rarely explored for gas sensing applications. Recently, Wang et al.⁴²⁵ synthesized spatially separated Pt and Pd modified Ti₃C₂T_x (Pt=Pd/Ti₃C₂T_x) via a two-step loading process of metal NPs (Fig. 23d). They systematically studied the H₂ sensing performance of Pt=Pd/Ti₃C₂T_x in comparison with monometallic Ti₃C₂T_x (Pd-Ti₃C₂T_x and Pt-Ti₃C₂T_x) as well as PdPt alloys modified Ti₃C₂T_x. As illustrated in Fig. 23e and f, the sensing performances of Pt=Pd/Ti₃C₂T_x to H₂ are superior to that of other materials, including sensing response and t_res/t_rec. Researchers assumed that the improved sensing performance is ascribed to the enhanced electron transfer between the Pt and Pd metals via supported Ti₃C₂T_x, while the PtPd alloy formation may weaken the electron transfer⁴²⁶. Another strategy for enhancement of sensor performances is fabricating MXene-MOS composites by incorporation of the properties of MXenes and MOSs⁴²⁷. This approach enables tailoring of MXene/MOS interfacial structures and active sites, optimizing electron transfer dynamics and gas adsorption capabilities^428,429. Chen et al.⁴²⁷, reported H₂ sensors based on Ti₃C₂T_x-SnO₂ nanocomposite films (Fig. 23g). The SEM image of the composite in Fig. 23h unveiled that lamellar-structured MXenes were closely connected with hexagonal SnO₂ NSs, forming heterojunctions. Some of SnO₂ NSs were inserted between Ti₃C₂T_x layers to expand the interlayer spacing of Ti₃C₂T_x, resulting in increasing active sites and enhancing gas adsorption kinetics⁴³⁰. From the response curve in Fig. 23i and j, by optimizing the amount of added MXene, t_res/t_rec were shortened to 11/13 s, and the sensor was able to detect a wide range of H₂ concentrations (10–2000 ppm). Leveraging the synergistic effect of metal NPs and MOSs, Zhang et al.⁴³¹ proposed a novel heterostructure composed of SnO₂-TiO₂/MXene decorated with Pd NPs. The gas sensor demonstrated a low LOD (200 ppb) and rapid t_res/t_rec (6/9 s) with a good linear response to H₂. In addition, machine learning techniques were adopted to further enhance the sensor’s selectivity, which can improve the accuracy of gas identification in complex environments.

Fig. 23: The strategies to enhance the sensing performance of Mxene based H₂ sensors.

a Schematic illustration of the synthesis and delamination of V₂CT_X MXene; b Gas response toward 100 ppm H₂ and five other interfering gases at RT; c Relationship between response and concentration of H₂ (6–200 ppm)⁴¹⁶. Copyright 2019, American Chemical Society. d Schematic diagram of the synthesis procedure of Pt=Pd/Ti₃C₂T_X; Sensing properties of four samples: e response/recovery time at 200 ppm H₂ and f response versus H₂ concentration (200–1000 ppm)⁴²⁵. Copyright 2023, Elsevier. g Heterostructural illustration and h SEM image of SnO₂-MXene composite nanomaterials; i Transient response curves to 600 ppm H₂ and j fitting curve of response to 10–2000 ppm H₂ of MXene/SnO₂ composite sensors with various ratios⁴²⁷. Copyright 2024, Elsevier

Table 5 summarized the sensing characteristics of recently reported 2D material-based H₂ sensors. In conclusion, 2D materials such as TMDs and MXenes, exhibit RT sensing performance due to their high surface-to-volume ratio and tunable electrical properties. However, the reliability and stability of pristine monolayer or few-layer 2D materials in practical settings remains constrained by their vulnerability to oxidative and humid conditions. This limitation becomes particularly pronounced within the harsh environment within a battery pack during Stage II of TR, where temperatures soar rapidly above 100 °C⁵⁰. Therefore, more efforts are demanded for enhancing the stability of 2D material-based sensors within battery packs. Alternatively, employing wide-bandgap semiconductors presents another promising route. Compared to other materials, III-V semiconductors exhibit potential for H₂ sensing applications owing to their stable crystal structure and chemical bonding, which contribute to excellent chemical stability and resistance to degradation. This will be discussed in the following section.

Table 5 Summary of sensing properties of 2D material-based H₂ sensors

Wide band-gap semiconductors-based chemiresistive H₂ sensor

Wide band-gap semiconductors, such as silicon carbide (SiC)⁴³² or III-nitride materials (GaN, InGaZnO, etc.)⁴³³ have been widely used in silicon-based electronic devices ascribing to their chemical stability, high temperature/high power durability⁴³⁴, and high electron mobility⁴³⁵. Through structure design and surface modification, wide band-gap semiconductors also possess application potential in the field of gas sensing. For instance, to increase specific surface area and adsorption sites, porous GaN honeycomb networks and GaN NW films have been developed for H₂ detection^{436,437,438,439}. Shafa et al.⁴³⁶ fabricated H₂ sensor based on porous GaN and investigated the effect of pore radius and density on H₂ sensing (Fig. 24a). According to the dynamic response curves in Fig. 24b, c, the higher porosity promoted H₂ diffusion, leading to enhanced sensitivity and improved utilization of inner sensitizer. However, similar to other sensing materials, pure wide band-gap materials also exhibit poor selectivity to H₂. To address this issue, there are two typical strategies: (i) noble metal functionalization; (ii) covering of the top surface of sensing materials with physical or chemical filters. The method of decoration of noble metals, such as Pd and Pt, has been reported in many works^{436,438,440,441}. Li et al.⁴⁴¹ fabricated Pd NPs decorated GaN NWs for enhanced H₂ sensing, as shown in Fig. 24d, e. In air, the decoration of Pd NPs promotes the adsorption of oxygen on the GaN NWs surface and the formation of Schottky barrier at Pd-GaN interface generates EDL with high resistance baseline. In H₂, a large resistance variation is induced via two mechanisms (Fig. 24f): the reaction of H₂ with adsorbed oxygen that releases electrons back to GaN, and electron transfer between PdH_x (formed by H₂ absorption in Pd) and GaN. Thus, the sensing response of Pd-GaN-based sensor to H₂ in a range of (0.1–10,000 ppm) was significantly enhanced via optimizing the thickness of deposited Pd film (Fig. 24g, h). Due to the high selectivity of Pd for H₂, the fabricated Pd/GaN sensor demonstrated a much higher response to H₂ than other interfering gases (Fig. 24i). However, it has been reported that CO, NO_x, and sulfuric compounds can poison noble metal catalysts (e.g., Pd)^46,442,443. Therefore, filter layers with well-defined microporous structures are designed for selective H₂ detection by effectively blocking the diffusion of interfering gases owing to the different kinetic diameters between H₂ and other molecules^444,445. To date, microporous polymer^46,324 polymer-derived ceramics⁴⁴⁴, and metal-organic framework (MOF) membrane^321,446 have been used as filter materials. Huang et al.⁴⁴⁶ fabricated amorphous IGZO (a-IGZO) films covered with a ZIF-8 membrane for H₂ sensing. Insets in the dynamic sensing curves of a-IGZO-based sensors without and with a ZIF-8 membrane are shown in Fig. 24j-l. Compared to a-IGZO based sensor, the sensing response of ZIF-8/a-IGZO shows a slightly decrease. Whereas, the sensor’s responses to CO (100 ppm), O₃ (5 ppm), and NO₂ (5 ppm) were significantly suppressed after introducing a ZIF-8 membrane. This molecular sieving effect stems from the inherent properties of the ZIF-8 membrane, whose small pore size (0.34 nm) restricts the diffusion of gas molecules with larger kinetic diameters, such as CO (0.376 nm), O₃ (0.380 nm), and NO₂ (0.364 nm)^447,448. Also, the sensor with a ZIF-8 membrane experiences a slight decrease of 8% in response after 600 cycles of operation and showed a long-term stability over a month. Moreover, due to their excellent compatibility with MEMS technology, many field-effect-transistor chemiresistive gas sensors based on III-V semiconductors have been reported to enhance their gas-sensing performance via gate modulation^449,450,451. In general, though III–V semiconductors-based sensors did not exhibit ultrahigh sensitive toward H₂ compared with other conventional types of sensing material, they are still worthy investigation owing to their high compatibility with MEMS technology processes (Fig. 25a).

Fig. 24: H₂ sensors based on wide band-gap semiconductors.

a SEM image of porous GaN structure; Effect of porosity on H₂ response at RT b versus time and c versus the porosity⁴³⁶. Copyright 2019, American Chemical Society. d, e HRTEM images and F the schematic diagram of H₂ sensing mechanism of Pd-GaN NW; Dynamic resistance g and response h curves of Pd-GaN to various H₂ concentrations (1–10,000 ppm); i Selectivity of Pd-GaN⁴⁴¹. Copyright 2024, Elsevier. Dynamic response curves of a-IGZO films j without and k with a ZIF-8 film, and the insets show SEM images of corresponding materials; l Comparison between H₂ response of a-IGZO films without and with a ZIF-8 film toward H₂ and interfering gases⁴⁴⁶. Copyright 2024, Elsevier

Fig. 25: Key factors and optimization strategies for gas sensing performance: an overview.

Schematic graph of several strategies to improve sensor performance, including sensitivity, stability, selectivity, t_res/t_rec, broad linear detection range, and accuracy

Challenges, outlook, and conclusions

The safety concerns associated with LIBs are garnering increasing attention. During TR process, a lot of heat and flammable/explosive gases are released. This can lead to LIB failure and even severe accidents, including fires and explosions. To date, various reports have confirmed that H₂ is the first gas released in comparison to other indicator gases such as CO, CO₂ and CH₄^39,40. Hence, developing high-performance H₂ sensors for in-time and accurate detecting of the LIB status presents an effective strategy for early safety warning. In addition, with the development of green energy technologies, H₂ sensors play a crucial role in areas including H₂ transportation and storage. For practical applications, US DOE has established objectives for the advancement of H₂ sensor, which include achieving the t_res/t_rec within 1 s for H₂ ≥ 1%, a broad detection concentration range of 0.1–10%, and a lifespan exceeding 10 years⁴⁴. Although some progress has been made in H₂ detection technology, current research still encounters some challenges that require further addressing in chemiresistive H₂ sensor:

• (1) The sensing performance of H₂ need to be addressed. Given that H₂ gas can serve as an indicator for the formation of trace Li dendrites, it is essential to expand the detection range of H₂ to ppm levels or even ppb levels in order to accurately assess the conditions of LIB during their early stages⁴³. While many H₂ sensors are capable of detecting H₂ at ppm levels, their sensing speed and selectivity often degrade for H₂ < 0.1%. To our knowledge, such ultrafast sensing and recovery speed (t_res/t_rec ≤ 1 s) has rarely been accomplished in the field of chemiresistive H₂ sensor. Regarding selectivity, although decorating the sensing materials with Pd is an effective strategy, it may experience catalyst poisoning, which greatly impacts the sensitivity, selectivity and stability of sensors. Furthermore, the sensing behavior of H₂ sensors in various humidity levels (0–95% RH) and temperatures (−40–85 °C) should be studied to ensure reliable operation in ambient conditions. To date, several reports have demonstrated that covering sensing materials with a layer of polymer or MOF, such as PMMA and ZIF-8, can provide protection against interfering gases and humidity due to molecular sieving effect^46,321. However, the current methods do not consistently yield optimal results because the morphologies and operating temperatures of sensing materials vary significantly. Therefore, there is a pressing need for more reliable and universal strategies for designing sensing materials aimed at enhancing sensor selectivity for commercialization.

• (2) The long-term stability of H₂ sensor is one of the most parameters for practical applications. It requires that H₂ sensors operate reliable and stably without significant single drift over time. Currently, researchers primarily focus on enhancing sensitivity, linear detection range, response-recovery speed, and anti-interference ability of H₂ sensors. However, there is a relative lack of research concerning their long-term stability in real-world applications. Although several strategies have been proposed to improve sensor stability-such as use of Pd alloys¹¹⁴, decrease of catalyst density²⁴⁸, and passivation of the underlying TMDs via in-situ oxidization⁴⁰⁸—a comprehensive analysis of the failure mechanism remains insufficient. Thus, a deeper understanding of the failure mechanism of gas sensors is essential to provide a guidance for design of highly stable H₂ sensors.

• (3) Exploring novel testing and analyzing strategies is also important for optimizing sensor performance. Conventional electronic chemiresistive gas sensors are designed to provide variations in resistance or current as an output in response to changes in gas concentrations. However, the non-linearity exhibited by chemiresistive sensors has been regarded as an intrinsic challenge, attributed to the power law that governs their direct current resistance response^452,453 This non-linear behavior adversely impacts the sensitivity of sensors at elevated gas concentrations and necessitates additional sensor calibration, thereby increasing costs⁴⁵⁴. To address this problem, Potyrailo et al.⁴⁵⁵ first proposed using an impedance measurement based on a dielectric excitation technique to detect signals from MOS-based sensors 2020. By varying frequencies, the impedance of SnO₂-based sensor showed a linear sensing behavior up to 10,000 ppm methane while its resistance response progressively saturated. Additionally, this approach significantly enhanced the stability of sensors against environmental conditions such as ambient humidity and temperature fluctuations. Similarly, using this measurement method can also enhance the performances of TMDs-based gas sensors, including linear detection range, high sensitivity, and stable baseline⁴⁵⁶. Recently, Zhang et al.⁴⁵⁷ proposed a chemiresistive-potentiometric multivariate sensor aimed at improving its capability for discrimination gases by providing more signal information. Therefore, alongside designing sensing materials, innovative testing approaches may play an important role in enhancing H₂ sensing performance.

• (4) In addition to emphasizing the performance of H₂ sensors themselves, the development of complementary technologies is equally crucial for practical TR monitoring. It is undeniable that the performance of H₂ sensors presents a vital factor in TR monitoring. However, nearly all reported H₂ sensors are currently evaluated under specific experimental conditions, which limits their practicality for accurate prediction and safety warning in real-world TR detection. Thus, the complicate internal conditions within LIB packs further exacerbate this challenge, making it difficult to assess LIB status accurately with data from a single H₂ under simulated conditions. Therefore, there is an urgent need for development of intelligent gas sensor arrays capable of processing data from sensor arrays using Artificial Intelligence technology to enhance predication accuracy^{320,440,455,458}. In addition, although gas sensors offer superior response speed and stronger sensing abilities compared to temperature, pressure, and voltage sensors³², the incorporation of multiple signals remains essential for presenting a comprehensive view of the LIB status, thereby improving safety warning systems.

In this review, the authors stress the importance of developing high-performance H₂ sensors with high sensitivity, fast response and recovery speed, high stability, and good selectivity to address the safety concerns of LIBs. To facilitate a comprehensive discussion on the current status of H₂ sensor development, we classified these sensors based on their sensing materials such as metals (e.g., Pd and Pd alloys), MOSs (single MOSs and MOS-based composites), carbons (CNTs, Gr, and GDY), 2D materials (TMDs and MXenes), and wide band-gap semiconductors. Within each material category, various strategies have been discussed to improve sensor performance. The primary methods for optimizing sensing properties include regulation of morphologies, doping/modification with metals, construction of heterojunctions, and designing of composites. These approaches effectively alter grain size, porosity, specific surface area, and reactivity of the sensing materials. Furthermore, the formed heterojunctions can enhance electron transfer characteristics between gases and materials, and increase the number of active adsorption sites. Moreover, owing to different underlying sensing mechanisms, there are several unique approaches to enhance their sensing properties. For instance, owing to the lattice expansion mechanism observed in Pd-based sensors, construction of nanogaps can greatly enhance their sensitivity and reduce their LOD with optimized width of nanogaps. Additionally, considering that slow oxidation may occur under mild operational conditions for 2D materials such as TMDs and MXenes, in-situ oxidation emerges as an effective method for constructing heterojunctions with MOSs while passivating the surface of 2D materials. This method improves both sensitivity and stability in H₂ sensors. To achieve a broad linear detection range, several strategies have been reported. These include the introduction of a buffer layer between the substrate and Pd to suppress phase transitions, as well as the application of heat to eliminate hysteresis. Furthermore, the development of novel measurement techniques has demonstrated significant potential for facilitating linear detection behavior. Therefore, to advance the development of high-performance H₂ sensors, it is imperative to establish a comprehensive understanding of the intrinsic response mechanisms and failure modes inherent in diverse sensing material systems. This understanding is vital for addressing key challenges in optimizing sensor sensitivity, stability, selectivity, and t_res/t_res for practical applications. Figure 25 summarizes some commonly used strategies in existing reports for enhancing sensor performance. Beside the inherent sensor performance, multi-signal information and Artificial Intelligence are needed to provide a reliable and accurate prediction of TR in the future.