Metal-oxide (MOX) chemiresistors

Summary

MOX chemiresistors are heated oxide chips whose resistance falls when a reducing gas reacts with chemisorbed surface oxygen. They are cheap, rugged, give an instant reading with no moving parts, and are the sensor family behind decades of household gas alarms: good for dense, low-cost, always-on leak screening.

They cannot distinguish H2 from methane, CO, alcohol vapor, or most other reducing gases, and the baseline drifts with humidity, temperature, and slow poisoning from siloxanes or H2S. For Hestia, MOX is a useful first-pass screening layer for candidate anomalies, not a standalone quantitative measurement, and not a credible downhole tool.

At a glance — published performance envelope

Detection limit

0.001–50 ppm

Range

1–10,000 ppm

Response t₉₀

1–120 s

Recovery

10–600 s

Selectivity

low

Drift

high

Operating T

−30 … 85 °C

Humidity

5–95 % RH

Power

15–900 mW

Unit cost

$2–$50

TRL

9

Main interferents: O₂, CO, CH₄, ethanol · poisons: siloxanes, H₂S, SO₂. Ranges span the published devices/variants covered below — per-claim citations are in the body; data: src/data/modalities/mox-sensors.yaml.

How it works

A MOX sensor is a heated oxide film (200–450 °C) whose resistance is set by surface redox chemistry. In air, chemisorbed oxygen withdraws conduction-band electrons from the oxide; H2 reacting with that oxygen returns them, collapsing the depletion layer and dropping resistance in n-type material [1][2].

Receptor event. Atmospheric oxygen ionosorbs onto the hot surface, extracting electrons. The dominant species is temperature-dependent [1][6]:

$\mathrm{O_2(gas)} + e^- \rightarrow \mathrm{O_2^-(ads)} \quad (\lesssim 150\,^\circ\mathrm{C})$ $\mathrm{O_2^-(ads)} + e^- \rightarrow 2\,\mathrm{O^-(ads)} \quad (150\text{--}400\,^\circ\mathrm{C})$ $\mathrm{O^-(ads)} + e^- \rightarrow \mathrm{O^{2-}(ads)} \quad (\gtrsim 400\,^\circ\mathrm{C})$

This depletes electrons in a band-bent region at each grain surface, the receptor event in Yamazoe’s framework [1][6]. n-type oxides (SnO2, ZnO, WO3, In2O3, TiO2, the standard H2-sensing materials) show rising resistance from this depletion; p-type oxides (NiO, CuO, Co3O4, Cr2O3) show the inverse via hole accumulation [1].

Transducer function. Current crosses grain-to-grain necks, modeled as two back-to-back Schottky barriers per boundary; total resistance is set by that barrier, not bulk grain resistance [7]. The depletion width is the Debye length,

$\lambda_D = \sqrt{\frac{\varepsilon k_B T}{e^2 N_d}}$

set by permittivity $\varepsilon$, bulk donor density $N_d$, and temperature $T$ [4]; measured values run a few nanometers (~4.2 nm undoped vs. ~2.3 nm at 5% doping, antimony-doped SnO2 films). Sensitivity depends on grain size $D$ relative to $2\lambda_D$ [1][3]:

Regime	Condition	Effect
Grain-boundary control	$D \gg 2\lambda_D$	Only the outer shell depletes; weak response
Neck control	$D \approx 2\lambda_D$	Depletion reaches the grain core; response grows sharply
Grain control	$D \lesssim 2\lambda_D$	Whole crystallite depletes; sensitivity maximized

$D \approx 2\lambda_D$ is why the field moved to nanocrystalline films and, later, single nanowires [1][3].

H2 reaction. H2 reacts with chemisorbed oxygen, releasing water vapor and returning the electron:

$\mathrm{H_2} + \mathrm{O^-(ads)} \rightarrow \mathrm{H_2O(gas)} + e^-$

collapsing the depletion layer and lowering resistance in n-type oxides [1][2]. Some In2O3/ITO and ZnO materials instead show hydrogen-induced metallization: H2 dissociates catalytically, and atomic hydrogen diffuses in as a shallow donor, building a surface accumulation layer rather than modulating an existing depletion layer. Both pathways lower resistance but arise from different physics [2].

Power law. Conductance follows $G \propto P_{H_2}^{\,n}$ (equivalently $R \propto P_{H_2}^{\,-n}$), formalized by Yamazoe and Shimanoe [5]. Limiting cases are $n \approx 1/2$ when O⁻ dominates the surface and a different fractional value (variously ≈1/4 or ≈1, depending on derivation) when O²⁻ dominates [1][3][4]; treat $n$ as material- and regime-dependent, not universal. Commercial datasheets quote the empirical decade-per-decade slope directly from calibration-gas exposures rather than deriving $n$, typically 0.6–1.2 on SnO2 bead parts [8][9][13][14].

Kinetics. Response and recovery are each Arrhenius-limited, $\tau \propto \exp(E_A/2k_BT)$ [4]: readings drift with the sensor’s actual internal temperature, not just the hotplate set-point, and different reducing gases carry different effective $E_A$, the physical basis for temperature-cycled pseudo-selectivity (below).

Noble-metal sensitization. Pd and/or Pt additions work through two additive mechanisms [1][2][6]: electronic sensitization, where Pd/PdO pins the local Fermi level and adds nanoscale depletion regions before any gas arrives; and chemical sensitization (spillover), where the metal catalyzes O2/H2 dissociation and the atomic species diffuse onto the oxide, speeding the surface reaction. Chemical sensitization fails once the noble-metal surface is poisoned; electronic sensitization is more structural. This differs from palladium-resistive sensors, where Pd’s own bulk hydride chemistry does the sensing rather than riding on an oxide host.

Device classes

Base oxide sets the chemistry, morphology sets sensitivity, and doping plus operating mode add selectivity. Commercial H2 parts combine several of these choices at once.

Base oxide.

Oxide	Notes	Ref
SnO2	Historical and still-dominant workhorse (n-type, ~3.6 eV gap); the material Taguchi first commercialized	[10]
ZnO	Wurtzite, n-type; some room-temperature H2 response in nanostructured form	—
WO3	Used mainly for H2S/NO2, but Pd-doped WO3 shows very high room-temperature H2 response (~10⁴)	[2]
In2O3	Small electron effective mass gives dispersive conduction bands; usually an NO2/O3 platform, but doped nanowire FETs show H2 response	[1]
TiO2 / composites	Nanotube TiO2 and mixed oxides (SnO2-Co3O4, In2O3-ZnO, ZnO-PdO) trade process complexity for lower operating temperature or better selectivity	[1][2]

Doping. Undoped oxide is cheapest and least selective. Noble-metal doping (Pd, Pt, sometimes Au) adds the sensitization mechanisms above. Heterojunction pairs (n-n or p-n, e.g. SnO2-Co3O4) add a second internal barrier that responds differently by gas, a further selectivity lever [1][36].

Morphology.

• Sintered thick-film beads with an embedded wire-coil heater: the original Taguchi/Figaro/Nissha/Winsen architecture, still sold today [8][13][16].
• Planar thick/thin film on alumina or silicon with a printed heater (Nissha SP series, Figaro TO-5) [9][14].
• MEMS micro-hotplates: a suspended membrane a few hundred µm across carrying heater and oxide film, lower power and faster thermal cycling than a bead, the platform behind nearly all modern digital parts (Bosch BME688, Renesas ZMOD, Amphenol/SGX MiCS) [19][22][24].
• Nanostructured 1-D materials (nanowires, nanorods, nanotubes) push feature size below $2\lambda_D$ and produce most sub-ppm/ppb laboratory results, but remain largely pre-commercial for H2 [1][2].
• Membrane overlayers: cellulose nanocrystal membrane coatings filter interferents and raise H2 selectivity [35].

Operating mode.

• Isothermal DC (fixed set-point): simplest, least selective, the default on legacy bead/TO-5 parts.
• Temperature-cycled operation (TCO): a repeating hotplate waveform (~100–400 °C) sampled at multiple points; because CO/H2S react fastest at 150–250 °C, alcohols/ketones at intermediate temperatures, and alkanes like CH4 only above ~400 °C, the multi-point signature carries pseudo-selective information a single isothermal reading cannot [34].
• Pulsed/duty-cycled heating: Figaro’s TGS8100 drops average power to ~15 mW versus several hundred mW for a continuously-heated bead of similar sensitivity [11].

Packaging. Bare resistive elements need an external load resistor, ADC, and user-supplied calibration curve (classic Figaro/Nissha/Winsen parts) [8][9][13][14][16]. Pre-integrated digital modules add an onboard ASIC/MCU with I2C/UART output and vendor compensation or trained-model post-processing (Bosch BME688+BSEC, Renesas ZMOD+AI) [22][24].

Vendor-labeling caveat

Not every part marketed under an “H2 sensor” line is a MOX chemiresistor. Winsen’s ZE03-H2, sold beside the genuinely MOX MQ-8 in the same product category, is a three-electrode electrochemical cell [16][18]. Figaro sells catalytic-bead (TGS6812) and electrochemical (FECS-H20) H2 sensors alongside its true MOX line (TGS2616) under one combined datasheet [10]. Check the sensing principle in the datasheet body, not the product family name.

No single MOX element is gas-selective on its own, so the other lever is arrays: run several oxides, dopants, or temperatures in parallel and apply pattern recognition (PCA, LDA, ML classifiers) to the multi-channel signature [34]. See Sensor arrays & machine learning.

Performance envelope

Ranges vary by product generation and material system; “(mkt.)” marks a market-listing figure rather than a datasheet value.

Parameter	Typical / range	Notes	Ref
Detection floor (rated)	1–50 ppm H2	TGS8100/TGS2600 rated from 1 ppm; TGS821 floor 50 ppm; TGS2616-C01 floor 30 ppm; SB-19-00 floor ~10 ppm (mkt.)	[8][9][11][12][13]
Laboratory/demonstrated LOD (research devices)	sub-ppm to low-ppb	Pd-decorated SnO2 nanowires, RT operation, response ~10³–10⁴ at 10,000 ppm H2 with <2 s response; individual ZnO nanowires ~1.5 sensitivity at 100 ppm RT	[2]
Broadband “gas-scan” claimed floor (uncalibrated)	ppb-range	Bosch BME688 spec sheet states VOC/VSC/CO/H2 detection “in the ppb range,” but output requires AI-Studio training and is not an absolute, metrologically traceable H2 concentration	[24]
Rated upper range	1,000–10,000+ ppm (device-dependent)	MiCS-5914 H2 range 1–1000 ppm; TGS2616-C01 to 3,000 ppm; TGS821 to ~4,000 ppm; all well below the 40,000 ppm (4 vol%) LEL	[8][9][19]
Response time (t90-class)	~2 s to ~100 s	Sol-gel SnO2 <10 s at 100 °C; Pd-SnO2 nanowires <2 s at RT; Pd-doped WO3 <100 s at RT	[2]
Recovery time	tens of seconds to minutes	Generally slower than response; motivates TCO-based system modeling approaches to “overcome the slow recovery” of MOX sensors	[1][2]
Sensitivity slope (decade/decade)	0.6–1.2 (SnO2 bead parts, 100→1000 ppm H2)	Directly specified on Figaro TGS821, Nissha SB-19-00/SP-19-01 datasheets	[8][13][14]
Sensor resistance at 100 ppm H2	0.2–10 kΩ (bead/TO-5 parts); 10–1,500 kΩ (MEMS, in clean air)	Wide spread by architecture and generation	[8][9][13][14][19]
Operating (hotplate) temperature	200–450 °C internal	Material- and gas-dependent optimum (e.g., ~300 °C TiO2, ~400 °C ZnO reported for H2)	[1][34]
Ambient/package operating temperature	−30 °C to +85 °C (MEMS parts); −10 °C to +50 °C (classic bead parts)	SB-19-00: −10 to 50 °C; MiCS-2714/5914: −30 to 85 °C	[13][19][21]
Relative humidity range (rated)	<95% RH non-condensing	Explicit ceiling on Nissha datasheets; response magnitude varies strongly within this range (see Cross-sensitivity & environment)	[13][14]
Oxygen dependency	Requires ~18–21% O2 for rated performance	Nissha datasheets state sensitivity “is influenced by the variation in oxygen concentration,” absolute minimum >18% O2	[13][14]
Long-term baseline drift (uncorrected)	25.7–40.8% coefficient of variation over 1 year (array study) vs. 0.38–1.21% short-term CV	One-year field e-nose dataset; short-term repeatability is far tighter than year-scale baseline stability	[28]
Drift after T/RH regression correction	Raw std-dev reduced 46–96% depending on sensor/model	e.g., MiCS-5524: 18.22 kΩ → 1.66 kΩ (91% reduction) after a 3D T/H/response regression correction	[27]
Rated/claimed lifetime	5–10 yr (qualified); vendor claims of 20+ yr in continuous operation	Renesas states ZMOD4510 tested for 10-yr lifetime, with the underlying MOx material demonstrated over 20+ years in actual continuous operation	[23]
Heater power	15 mW (pulsed, TGS8100) to ~660 mW (continuous bead, TGS821)	MEMS microhotplate parts generally 30–150 mW continuous, much lower duty-cycled	[8][11]
Cost class	~$2–20 (bare sensor, small quantity); ~$5–15 (MEMS chip-scale); $9–95 (fully-featured breakout/eval module)	BME688 bare chip $8.99 (DigiKey); MiCS-5914 ~$6.24–6.74 (discontinued generation)	[39][40]

A frontier engineered SnO2-Co3O4 nanocomposite reported as “ultra-selective” for H2 reached only an 11.7:1 H2:CO response ratio using machine-learning post-processing [36], a useful anchor for how far even research-grade selectivity sits from clean single-gas discrimination.

Figures of merit

Universal figures of merit are defined on the Figures of merit page; concentration units follow the concentration units primer. MOX specializes several of them: a power-law response instead of a fixed slope, a heater-driven operating point, and a humidity/oxygen-dependent baseline.

Figures-of-merit table for this modality (12 rows)

Figure of merit	Symbol / formula	Value for this family	How to read it
Sensitivity (response ratio)	$S = R_a/R_g$ (n-type)	Decade/decade slope 0.6–1.2 on commercial SnO2 beads (TGS821, SB-19-00, SP-19-01), 100→1,000 ppm [8][13][14]; Pd-doped WO3 lab devices report response ratios up to ~10⁴ [2]	Valid only at the concentration measured; log-linear, not linear
Power-law exponent $n$	$G \propto P_{H_2}^{\,n}$	Theoretical limits $n\approx1/2$ (O⁻-dominated) to ≈1/4–1 (O²⁻-dominated), material-dependent [1][3][4][5]; datasheets quote the empirical slope directly (0.6–1.2 decade/decade) rather than deriving $n$ [8][9][13][14]	No true linear range; extrapolating past the calibrated span multiplies error
Limit of detection (LOD)	vendor “detection floor”; no datasheet states $k\,\sigma_{blank}/S$ or averaging time	Rated commercial floor 1–50 ppm (TGS8100/TGS2600 from 1 ppm; TGS821 50 ppm; TGS2616-C01 30 ppm) [8][9][11][12][13]; sub-ppm to low-ppb only on research nanowire/nanocomposite devices [2]	Treat as marketing ranges, not an IUPAC/SNR-based LOD (unverified)
Selectivity / cross-sensitivity	$K_{ij}=S_j/S_i$	Best engineered case ~11.7:1 H2:CO (SnO2-Co3O4, ML-assisted) [36]; legacy parts publish no $K_{CO}$ (unverified); CH4 shows non-zero response on Figaro’s own TGS821 chart [8]; H2S is a cumulative poison, not a transient cross-reading [31]	Selectivity is set by the worst interferent (CO), essentially unrejected without array/TCO processing
Response / recovery time	$t_{90}$; recovery not derived from one $\tau$	~2 s (Pd-SnO2 nanowires, RT) to ~100 s (Pd-doped WO3, RT); recovery consistently slower, tens of seconds to minutes [1][2]	Recovery, not response, is usually the field throughput bottleneck
Dynamic range	$N_{\text{decades}}=\log_{10}(C_{\max}/C_{\min})$	~1.5–3 decades per part (TGS2616-C01: 30–3,000 ppm; MiCS-5914: 1–1,000 ppm), all below the 40,000 ppm LEL [8][9][19]	Log-linear device; a linear-axis plot flattens the low-concentration end artificially
Baseline resistance $R_0$ (air)	clean-air reference the response ratio is computed against	0.2–10 kΩ for bead/TO-5 parts, 10–1,500 kΩ for MEMS parts [8][9][13][14][19]	Varies nearly an order of magnitude unit-to-unit; calibrate against the individual unit, not a datasheet nominal
Operating (hotplate) temperature	$T_{op}$, internal to the sensing element	200–450 °C internal, material-dependent optimum (~300 °C TiO2, ~400 °C ZnO reported for H2) [1][34]	Main lever on both selectivity (via differential activation energies) and power draw
Heater power	$P_{\text{heater}}$	15 mW (pulsed, TGS8100) to ~660 mW (continuous bead, TGS821); MEMS microhotplates generally 30–150 mW continuous [8][11]	Duty-cycled operation buys roughly an order of magnitude lower power at the cost of thermal settling time
Humidity coefficient	% resistance change per RH span	Up to ~20% resistance drop from 40%→100% RH even on an engineered, hydroxyl-minimized SnO2 sensor, implying larger swings on conventional sol-gel material (unverified generalization) [26]; rated ceiling <95% RH non-condensing on commercial parts [13][14]	A raw reading without a co-logged RH channel reflects both H2 and an unknown humidity state
Drift / long-term stability	$R_0(t) = R_0(0) + d_Z t$	25.7–40.8% baseline coefficient of variation over one year, uncorrected, vs. 0.38–1.21% short-term CV [28]; reducible 46–96% via T/RH regression correction [27]; rated lifetime 5–10 yr qualified, vendor claims of 20+ yr continuous operation [23]	Excellent short-term repeatability can mask a year-scale baseline wander large enough to mimic a genuine geochemical trend
Operating envelope	compensated range ⊂ operating range ⊂ storage range	Ambient −30 °C to +85 °C (MEMS parts) or −10 °C to +50 °C (bead parts); <95% RH non-condensing; ≥18–21% O2 required for rated performance [13][14][19][21]	The O2 floor is MOX-specific: hypoxic soils near active seeps give an undocumented multiplicative error unless O2 is co-logged

Selectivity and drift, not raw sensitivity, decide whether a MOX reading is usable for geologic H2: the element is already more sensitive than exploration work needs, but unrejected cross-response and long-term baseline drift undercut it. The fixes that help selectivity (TCO, arrays) add the power and firmware complexity that MOX’s low-cost case depends on avoiding.

Cross-sensitivity & environment

Covariate effects are often the same order of magnitude as the target signal at Hestia-relevant, near-atmospheric-background concentrations [38], so a field reading needs companion channels, not just a raw ppm number.

Factor	Effect	Design response	Ref
Temperature	$R_s/R_0$ spans ~0.65–1.6 across −10 to +40 °C at fixed 100 ppm H2 on TGS821, roughly a 2× baseline swing from ambient T alone; kinetics also slow in the cold	Co-log ambient T; don’t trust absolute sensitivity from the hotplate set-point alone	[8]
Humidity	Surface hydroxylation lowers baseline resistance and blocks receptor sites; the dominant covariate	Co-log RH; see detail below	[1][3][26]
Oxygen	Receptor mechanism needs ambient O2; Nissha rates standard test at 21%±1%, minimum >18% for rated performance	Co-log O2, especially in hypoxic soils near seeps	[13][14]
CO	Same surface redox class as H2 at similar activation windows; hardest interferent to reject	Assume overlapping CO/H2 response without array/TCO processing	[36]
CH4	Needs higher activation energy (~400–500 °C); shows non-zero response on unmodified parts at normal operating temperature	High-temperature dwell or TCO to suppress CH4 cross-response	[8][34]
H2S	WO3 parts are themselves used as H2S sensors (~200 °C); on SnO2 it is a poison, forming sulfite then sulfate that resists desorption below ~500 °C	Expect a permanent baseline shift near sulfide-rich seeps, not a transient cross-reading	[1][31]
CO2	Not directly redox-active on generic SnO2; acts indirectly as a diluent shifting local O2 partial pressure	Low direct concern; log O2/CO2 together where headspace CO2 is high	—
Pressure	No datasheet rates operation above ~1 atm; response should track H2 partial pressure, but diffusion into the porous film and local O2 also depend on total pressure	Treat elevated-pressure (downhole) use as unvalidated	—
Helium	Mechanistically inert to the redox cycle at any practical sensor temperature; no paper directly confirms zero response	Bench-test before assuming (unverified, see Open questions)	—
Siloxanes	Cyclic/linear siloxanes (D4, HMDSO) deposit an insulating silica-like layer; poisoning accumulates and does not self-heal	Audit silicone sealants/coatings in any enclosure design	[30]

Siloxane poisoning. A TCO study found 33 ppm·hour cumulative HMDSO exposure, roughly 0.5–1 year of typical indoor off-gassing, cut surface reaction rate constants 40–60% across the gases tested. By 99 ppm·h cumulative dose, CO and acetone signal had fallen to near zero (~95% loss) while H2 had fallen by about 70%; by 684 ppm·h, H2 retained only about 13% of its original value. H2 degrades more slowly than CO/acetone, which paradoxically raises relative H2 selectivity as the sensor is progressively poisoned, but no regeneration was demonstrated [30].

Long-term drift. Baseline resistance drifts even at constant gas/T/RH conditions, attributed to slow annealing of non-equilibrium oxygen-vacancy donor states. A one-year e-nose field dataset found 25.7–40.8% baseline coefficient of variation versus 0.38–1.21% short-term CV in the same sensors [28]; published MOX benchmark datasets show comparable drift, which limits their reuse for classifier training without recalibration [29]. Mandatory periodic re-baselining (zero-air/span-gas cycling) is the practical answer for any unattended deployment.

Use for geologic H2

Field assessment

MOX is cheap, instant-reading, and a reasonable dense first-pass screening layer, but its temperature/humidity/oxygen/CO/CH4/H2S covariate structure is large relative to the Hestia-relevant signal, and no primary geologic-H2 field study reviewed uses it as the quantitative tool of record; the field has settled on electrochemical cells for that role. MOX is not a credible downhole candidate: it needs local active heating and local signal-conditioning electronics, working against a passive downhole design philosophy.

Soil-gas surveys. Low cost, instant readout, and ruggedness are undercut by the covariate structure at Hestia’s target range (roughly 1 ppm anomaly threshold up to percent-level): a ~2× T/RH-driven baseline swing [8], an oxygen dependency rarely logged in the field [13][14], and un-rejected CO/CH4/H2S cross-response [31][34][36]. A raw MOX ppm reading needs simultaneous T/RH/O2 logging and, ideally, confirmatory measurement to be defensible. Every geologic-H2 soil-gas field study reviewed here, MONHyTOR [32] and the Perth Basin comparative survey (handheld iBrid Mx6/GA5000, the WHALI autonomous unit) [33], uses an electrochemical sensor as its primary channel, not MOX; MOX-class elements (MQ-8/TGS-type) appear only in industrial/HVAC leak-check contexts, not calibrated exploration geochemistry.

Continuous monitoring stations. The core liability is drift compounded with poisoning: uncorrected annual baseline drift [28], siloxane exposure below a year of typical off-gassing already degrading sensitivity substantially [30], and H2S co-exposure risking a shift that only reverses near 500 °C [31]. None of this is visible from a single deployment-time calibration; it surfaces as slow baseline creep easily mistaken for a real trend unless the station re-zeroes periodically against a reference gas or sensor. TCO-based self-compensation exists in the literature for this [30] but adds firmware and power budget.

Downhole/borehole logging. A poor fit beyond ordinary hardening. Every part reviewed needs a continuously or duty-cycle heated element at 200–450 °C, at minimum a load resistor and ADC, and is validated only near ambient pressure and up to about 85 °C package temperature (some bead variants only to 50 °C) [13][19][21], far below the 100–230+ °C, tens-of-MPa envelope of deep H2-bearing formations; downhole electronics today are broadly qualified only to about 175 °C, with active materials work still needed for 250–400 °C [37]. A continuously hot element is also an intrinsic-safety burden against a 4 vol% LEL analyte, and the local-heating, local-electronics requirement runs counter to the passive design preference behind fiber-optic sensors. Treat MOX as surface/near-surface only.

Where MOX fits. A cheap, dense, disposable first-pass screening layer, e.g. a wide grid of low-cost, ideally TCO-capable nodes flagging candidate anomaly zones, with any location-specific quantitative claim resting on a co-located or follow-up reference measurement (electrochemical cell, TDLAS/optical technique, or lab GC on a bagged sample), matching how the reviewed field literature already operates [32][33]. See why H2 is hard and geologic context.

Vendors & products

Manufacturer	Product line	Model/SKU	Sensing tech	H2 range (as specified)	Power	Notable specs / caveats	Price class	Ref
Figaro Engineering	”Special” H2 sensor (legacy bead)	TGS821	SnO2 thick-film bead, TO-can	50–4,000 ppm (rated floor 50 ppm; sold as detecting 10–4,000 ppm in some distributor listings)	660 mW (heater, continuous)	Sensitivity slope 0.6–1.2 decade/decade (100→1,000 ppm); own datasheet chart shows CH4/EtOH/CO cross-response and ~2× T/RH baseline swing	Low ($10–20 class)	[8]
Figaro Engineering	MOS H2 sensor	TGS2616-C00 / -C01	SnO2 MOS, TO-5	30–3,000 ppm	280 mW	C01 variant specifically reduces acetylene interference for transformer-oil dissolved-gas-analysis use; sensor resistance 0.30–30 kΩ in 100 ppm H2	Low	[9][10]
Figaro Engineering	Low-power air-contaminant sensor	TGS8100	SnO2 MOS, pulsed-heater TO-5	~1–30 ppm (H2/ethanol class)	15 mW (pulsed duty cycle)	Designed for battery-powered equipment; ~92% power reduction vs. continuously-heated equivalents	Low	[11]
Figaro Engineering	Air-contaminant sensor	TGS2600	SnO2 MOS, TO-5	~1–30 ppm H2 (broad multi-gas, not H2-selective)	~210 mW	General air-quality part; not marketed as H2-selective	Low	[12]
Figaro Engineering	Catalytic / electrochemical H2 lines (not MOX, included for contrast)	TGS6812-D00 (catalytic); FECS-H20 (electrochemical, mfr. New Cosmos)	Catalytic bead / 3-electrode EC cell	Up to 100% LEL (catalytic); 0–4 vol% (FECS-H20)	—	Figaro explicitly recommends catalytic/EC over MOX for safety-critical, near-LEL, or FCEV (GTR-13) applications; MOX recommended instead for low-O2 environments and long-term low-level leak monitoring	—	[10]
Nissha FIS	SB (semiconductor bead) series	SB-19-00	SnO2 bead, embedded Pt heater coil	~10–2,000 ppm (market listing); rated $R_s$ 0.2–2 kΩ at 100 ppm	120 mW (0.9 V, 130 mA heater)	Explicit O2-dependency warning, min. >18% O2; ambient op. range −10 to 50 °C, <95% RH	Low	[13]
Nissha FIS	SP (planar) series	SP-19-01	SnO2 thick film on alumina, printed heater	Optimized for low-concentration H2; $R_s$ 0.5–5 kΩ at 100 ppm	315 mW (5 V, 80 mA heater)	Same O2-dependency and RH ceiling caveats as SB series	Low	[14]
Nissha FIS	Catalytic H2 line (not MOX)	FH2-HY04/05/06/11	Catalytic combustion	0–4 vol%	—	Sold under Nissha’s dedicated “hydrogen detector” product page, distinct from the SB/SP MOX line; siloxane-resistant; used in Toyota Mirai, Mercedes-Benz FCEVs	—	[15]
Winsen Electronics	MQ series	MQ-8	SnO2 bead (Taguchi-type)	~100–1,000 ppm	Bead-type heater class (typically several hundred mW)	Extremely widely used in hobbyist/industrial leak-alarm products; “anti-interference to other gases” claimed but not independently characterized here	Very low ($2–5 class)	[16][17]
Winsen Electronics	ZE03 module family (ZE03-H2 is electrochemical, not MOX)	ZE03-H2	3-electrode electrochemical cell + onboard µC	Family range 0–1,000 ppm (gas-dependent)	<5 mA @ 3.7–9 V	Flagged explicitly: marketed in Winsen’s “H2 sensor” category alongside the genuinely-MOX MQ-8, but a different sensing principle entirely	Low–mid	[18]
Amphenol/SGX Sensortech	MiCS-59xx	MiCS-5914	MEMS MOX (ammonia-primary; multi-gas)	H2: 1–1,000 ppm	mW-class MEMS heater	$R_0$ (air) 10–1,500 kΩ; sensitivity factor 1.5–15; precautions against organic-solvent/silicone/smoke exposure explicit in datasheet	~$6.24–6.74 (sourced; discontinued generation)	[19][40]
Amphenol/SGX Sensortech	MiCS-55xx	MiCS-5524	MEMS MOX	H2 secondary channel (CO/methane-primary part)	mW-class	Indoor CO/gas-leak combo part; used in T/RH drift-correction study above	Low	[20][27]
Amphenol/SGX Sensortech	MiCS-27xx	MiCS-2714	MEMS MOX	H2/NO2; 50 ppb–10 ppm	mW-class	−30 to 85 °C rated	Low	[21]
Renesas Electronics	ZMOD44xx	ZMOD4410	MEMS MOx microhotplate + CMOS ASIC, digital output	TVOC/IAQ-focused (mfr. does not publish a standalone H2-ppm spec; underlying MOx layer is broadband-reactive)	mW-class, duty-cycled	10-year qualified device lifetime; mfr. states underlying MOx material demonstrated 20+ years in actual continuous operation	~$5–10 class	[22]
Renesas Electronics	ZMOD45xx	ZMOD4510	MEMS MOx + ASIC	O3/NOx-focused (oxidizing-gas channel, paired with ZMOD4410 for cross-compensation)	mW-class	10-yr tested lifetime	~$5–10 class	[23]
Bosch Sensortec	BME6xy	BME688	MEMS MOx “gas-scan” element + integrated T/RH/P sensors	Broadband, ppb-range sensitivity to VOC/VSC/CO/H2, but not independently calibrated to H2 ppm without a customer-trained BME AI-Studio model	3.9 mA (standard gas-scan mode); <0.1 mA (ULP mode, 300 s update)	Requires BSEC middleware for any usable IAQ/bVOC-equivalent output; explicitly a relative/index output, not a metrological concentration	$8.99 bare chip (DigiKey)	[24][39]
Sensirion AG	SGP4x (context reference, not H2-specific)	—	MEMS MOx, VOC/NOx index	Not H2-targeted	—	Cited here for the vendor’s own plain-language explainer of generic MOX limitations (semiquantitative, broadband, drift-prone); useful onboarding reference	—	[25]

Open questions

1. No source reviewed tests MOX in Hestia-relevant matrix conditions (hypoxic headspace, elevated CO2, high co-located CH4/H2S). Commission a bench study on the actual expected gas matrix.
2. He blindness is inferred, not measured. A cheap bench test (pure He, He/H2 mixtures) would confirm it, since He is already a Hestia co-tracer gas.
3. No datasheet specifies behavior above ~1 atm. Contact Figaro/Nissha/Winsen/Amphenol directly before ruling out any elevated-pressure surface use.
4. Audit silicone sealants, conformal coatings, or RTV adhesives in Hestia’s own enclosures against the siloxane-poisoning data above [30] before any MOX-based station design.
5. Decide build-vs-buy on TCO firmware: in-house temperature-cycling on a bare bead/TO-5 part, or a pre-integrated vendor-trained module (Bosch BME688+BSEC, Renesas ZMOD+AI).
6. Write a mandatory covariate-logging SOP (temperature, RH, ideally O2); a bare MOX ppm number should not enter the decision pipeline without it.
7. Field-trial array/TCO selectivity against GC ground truth in an actual target basin; the ~11.7:1 H2:CO lab benchmark [36] is likely an upper bound for field hardware, not a floor.
8. State explicitly what role MOX plays in the program (none, screening-only, or otherwise). Published field practice (MONHyTOR [32], WHALI/Perth Basin [33]) has already converged on electrochemical for primary quantitative work.
9. Get firm unit-cost numbers across the vendor table; several entries rely on a spot price or market listing, and cost-per-node matters for the dense-grid screening case.

Related pages

• Why H2 is hard
• Geologic context
• Palladium-resistive sensors
• Work-function devices
• Thermal-conductivity & acoustic sensors
• Catalytic pellistors
• Electrochemical cells
• Fiber-optic sensors
• Sensor arrays & machine learning

References

Full reference list (40 entries)

1. “Road Map of Semiconductor Metal-Oxide-Based Sensors: A Review.” PMC10422562. https://pmc.ncbi.nlm.nih.gov/articles/PMC10422562/
2. “Hydrogen Gas Sensors Based on Semiconductor Oxide Nanostructures.” PMC3386698. https://pmc.ncbi.nlm.nih.gov/articles/PMC3386698/
3. “Metal Oxide Gas Sensors: Sensitivity and Influencing Factors.” PMC3264469. https://pmc.ncbi.nlm.nih.gov/articles/PMC3264469/
4. “Modeling Interfacial Interaction between Gas Molecules and Semiconductor Metal Oxides: A New View Angle on Gas Sensing.” PMC9685467. https://pmc.ncbi.nlm.nih.gov/articles/PMC9685467/
5. Yamazoe, N.; Shimanoe, K. “Theory of power laws for semiconductor gas sensors.” Sensors and Actuators B: Chemical 128(2), 566–573 (2008). DOI: 10.1016/j.snb.2007.07.036. https://www.sciencedirect.com/science/article/abs/pii/S0925400507005175
6. Yamazoe, N. “Receptor Function and Response of Semiconductor Gas Sensor.” Journal of Sensors 2009, Article 875704. DOI: 10.1155/2009/875704. https://onlinelibrary.wiley.com/doi/10.1155/2009/875704
7. Barsan, N.; Weimar, U. “Conduction Model of Metal Oxide Gas Sensors.” Journal of Electroceramics 7, 143–167 (2001). https://link.springer.com/article/10.1023/A:1014405811371
8. Figaro Engineering Inc. “TGS 821 – Special Sensor for Hydrogen Gas,” product datasheet. https://www.figarosensor.com/product/docs/TGS821.pdf
9. Figaro Engineering Inc. “TGS 2616-C01 – for the detection of Hydrogen,” product information sheet. https://www.figarosensor.com/product/docs/tgs2616-c01_product%20information(fusa)rev00.pdf
10. Figaro Engineering Inc. “TGS 6812 and FECS-H20 Catalytic and TGS 2616 MOS Hydrogen sensors and their application.” https://www.sensor-test.de/assets/Fairs/2025/ProductNews/PDFs/Figaro-Engineering-hydrogen-sensors-and-their-applications.pdf
11. Figaro Engineering Inc. “TGS8100” product datasheet. https://www.figarosensor.com/product/docs/TGS8100(0914).pdf
12. Figaro Engineering Inc. “TGS 2600 – for the detection of Air Contaminants,” product information sheet. https://www.figaro.co.jp/en/product/docs/tgs2600_product_information_rev02.pdf
13. Nissha FIS, Inc. “FIS Gas Sensor SB-19-00 for Hydrogen Detection,” specifications (SP1900E). https://www.fisinc.co.jp/en/common/pdf/SB1900E.pdf
14. Nissha FIS, Inc. “FIS Gas Sensor SP-19-01 for Hydrogen Detection,” specifications (SP1901E_P). https://www.fisinc.co.jp/en/common/pdf/SP1901E_P.pdf
15. Nissha FIS, Inc. Hydrogen sensor (catalytic-combustion, FH2-HY series) product page. https://connect.nissha.com/gassensor/en/product/hydrogendetector/
16. Winsen Electronics. “MQ-8 Hydrogen Sensor,” product page. https://www.winsen-sensor.com/sensors/h2-sensor/mq8.html
17. Winsen Electronics. “Flammable Gas Sensor (Model: MQ-8),” manual v1.3. https://cdn.sparkfun.com/datasheets/Sensors/Biometric/MQ-8%20Ver1.3%20-%20Manual.pdf
18. Winsen Electronics. “ZE03 Electrochemical H2 Gas Sensor Module,” product page. https://www.winsen-sensor.com/sensors/h2-sensor/ze03-h2.html
19. Amphenol SGX Sensortech. “MiCS-5914” datasheet, 1108 rev 4. https://www.sgxsensortech.com/content/uploads/2014/07/1108_Datasheet-MiCS-5914.pdf
20. Amphenol SGX Sensortech. “MiCS-5524” datasheet, 1084 rev 8. https://sgxsensortech.com/content/uploads/2014/07/1084_Datasheet-MiCS-5524-rev-8.pdf
21. Amphenol SGX Sensortech. “MiCS-2714” datasheet. https://www.sgxsensortech.com/content/uploads/2014/08/1107_Datasheet-MiCS-2714.pdf
22. Renesas Electronics. “ZMOD4410 Gas Sensor Module for TVOC and Indoor Air Quality,” datasheet and product page. https://www.renesas.com/en/document/dst/zmod4410-datasheet
23. Renesas Electronics. “ZMOD4510 Gas Sensor Module for Nitrogen Dioxide and Ozone,” datasheet. https://www.renesas.com/en/document/dst/zmod4510-datasheet
24. Bosch Sensortec GmbH. “BME688 4-in-1 environmental sensor unit with AI,” product flyer BST-BME688-FL000. https://www.bosch-sensortec.com/media/boschsensortec/downloads/product_flyer/bst-bme688-fl000.pdf
25. Sensirion AG. “What is a metal oxide (MOX) sensor?” Information note. https://sensirion.com/media/documents/0083CDF4/6294DFEA/Info_Note_MOX_sensor.pdf
26. “Reducing Humidity Response of Gas Sensors for Medical Applications: Use of Spark Discharge Synthesis of Metal Oxide Nanoparticles.” PMC6111941. https://pmc.ncbi.nlm.nih.gov/articles/PMC6111941/
27. “Correction Model for Metal Oxide Sensor Drift Caused by Ambient Temperature and Humidity.” PMC9101743. https://pmc.ncbi.nlm.nih.gov/articles/PMC9101743/
28. “Long-term drift behavior in metal oxide gas sensor arrays: a one-year dataset from an electronic nose.” Scientific Data (2025). https://www.nature.com/articles/s41597-025-05993-8
29. Dennler, N.; Rastogi, S.; Fonollosa, J.; van Schaik, A.; Schmuker, M. “Drift in a Popular Metal Oxide Sensor Dataset Reveals Limitations for Gas Classification Benchmarks.” arXiv:2108.08793. https://arxiv.org/abs/2108.08793
30. “Siloxane treatment of metal oxide semiconductor gas sensors in temperature-cycled operation – sensitivity and selectivity.” Journal of Sensors and Sensor Systems 9, 283–292 (2020). https://jsss.copernicus.org/articles/9/283/2020/
31. Yang, et al. “H2S Sensing with SnO2-Based Gas Sensors: Sulfur Poisoning Mechanism Revealed by Operando DRIFTS and DFT Calculations.” Angewandte Chemie International Edition (2025). PMC12124429. https://pmc.ncbi.nlm.nih.gov/articles/PMC12124429/
32. “In-soil hydrogen concentration monitoring using MONHyTOR: application for natural hydrogen exploration.” Advances in Geosciences 67, 57 (2025). https://adgeo.copernicus.org/articles/67/57/2025/
33. “Natural hydrogen soil gas emissions near Harvey, Perth Basin: a comparative study of survey methods.” Science and Technology for Energy Transition (2025). https://www.stet-review.org/articles/stet/full_html/2025/01/stet20250031/stet20250031.html
34. “Metal Oxide Gas Sensors, a Survey of Selectivity Issues Addressed at the SENSOR Lab, Brescia (Italy).” PMC5421674. https://pmc.ncbi.nlm.nih.gov/articles/PMC5421674/
35. “Improving Hydrogen Selectivity in Semiconductor Metal Oxide Gas Sensors with Cellulose Nanocrystal Membranes.” PMC12670992. https://pmc.ncbi.nlm.nih.gov/articles/PMC12670992/
36. “SnO2-Co3O4 nanocomposite sensor: Achieving ultra-selective hydrogen detection in mixed gas environments.” Sensors and Actuators B: Chemical (2024). https://www.sciencedirect.com/science/article/abs/pii/S0925400524012516
37. “Sensing Electronics for Downhole Drilling in Extreme Environments.” Technology listing (NREL-associated). https://labpartnering.org/technologies/3f72efda-2142-450a-aa56-547d82b5ed4b
38. “H2 in Antarctic firn air: Atmospheric reconstructions and implications for anthropogenic emissions.” Proceedings of the National Academy of Sciences 118 (2021). https://www.pnas.org/doi/10.1073/pnas.2103335118
39. DigiKey Electronics. Bosch Sensortec BME688 product/pricing listing. https://www.digikey.com/en/products/detail/bosch-sensortec/BME688/13681261
40. Newark/element14. Amphenol SGX Sensortech MiCS-5914 product/pricing listing. https://www.newark.com/amphenol-sgx-sensortech/mics-5914/mos-sensor-ammonia-300ppm-5-1v/dp/71AC1528