Your air quality monitor cannot smell anything. It watches oxygen cling to a grain of tin oxide

There is a small metal can inside the LPG leak alarm screwed to a kitchen wall, and a similar one inside almost every indoor air quality monitor sold for under a few thousand rupees. It is doing something much stranger than the product page suggests. It is not detecting gas. It is running a tiny heater at around 350 °C, keeping a speck of tin oxide hot, and continuously measuring how hard it is to push a current through that speck.

Everything else — the number on the display, the colour of the LED, the alarm — is inference from a resistance. Once you know what changes that resistance, the whole family of odd behaviours these devices show stops being mysterious and becomes predictable.

Oxygen is the thing being measured

Start with the sensing element: a porous film of a metal oxide semiconductor, usually tin dioxide (SnO₂), sometimes zinc oxide or tungsten trioxide, sintered into a mass of small connected grains.

In ordinary air, oxygen molecules adsorb onto the surface of those grains. Adsorbed oxygen is electron-hungry: it pulls electrons out of the semiconductor to become surface species such as O₂⁻ and O⁻. Those electrons are now trapped at the surface and no longer available to carry current, so a shell near the surface of every grain is stripped of charge carriers. This is the depletion layer, and it is electrically a poor conductor wrapped around a better one.

Current through the film has to hop from grain to grain, and at every contact it must cross two of those depleted shells. That barrier is what sets the sensor's resistance in clean air — its baseline.

Now let a reducing gas arrive — carbon monoxide, hydrogen, methane, an alcohol, almost any volatile organic compound. At the sensor's operating temperature it reacts with the adsorbed oxygen and carries it away as carbon dioxide and water. The trapped electrons are released back into the grain, the depletion layer thins, the barrier at each grain contact falls, and the resistance drops. The electronics read that drop and convert it into a number.

So the sensor never identifies a gas. It reports that something removed oxygen from its surface. Hold on to that sentence — most of what follows is a consequence of it.

Why grain size is the whole design

Here is where this becomes a nanomaterials story rather than an electronics one.

The depletion layer has a characteristic thickness, set by the material and its carrier concentration, called the Debye length. For tin dioxide at operating temperature it is of the order of a few nanometres.

If a grain is large — say a hundred nanometres or more — the depleted shell is a thin skin on a conductive core. Current takes the easy path through the core, and changes at the surface shift the resistance only modestly. But shrink the grain until its diameter approaches twice the Debye length, and the depletion regions from opposite sides meet: the entire grain is depleted. Now there is no easy interior path at all, and the conductivity of the whole grain — not just its skin — responds to what is happening on the surface.

The result is a step change rather than a gradual improvement. Sensitivity rises sharply once the grain size crosses that threshold, which is why sensor-grade metal oxides are specified in nanometres and why how the powder was synthesised decides how good the device can be. It is one of the cleanest cases anywhere of a nanoscale effect that is not a curiosity but the entire basis of a commercial product.

There is a tension built into it, though. Fine grains sinter and coarsen at operating temperature over months, the grains grow, and the sensitivity that depended on being below the threshold quietly declines. Much of the materials engineering in a good sensor is about keeping small grains small for years at 350 °C.

Why it has to be hot

The heater is not a design flaw, and it is not there to burn anything off. It is there because the chemistry has to be fast and reversible.

Oxygen adsorption, the reaction with the target gas, and the desorption of products all need thermal energy to proceed at a useful rate. Below about 200 °C the surface reactions are too slow and the sensor responds sluggishly if at all; the useful window is typically 300 to 450 °C, and the optimum temperature differs by target gas — which is itself sometimes exploited, by cycling the heater through several temperatures and reading the pattern of responses.

Three practical consequences follow directly.

• It uses real power. A continuously heated sensor draws on the order of tens to a hundred milliwatts, which is why a battery-powered gadget either has a short life or pulses the heater and sleeps between readings — and a pulsed sensor is measuring a surface that has not fully settled.
• It needs a warm-up. After power-on the surface has to reach an equilibrium coverage of adsorbed oxygen. Minutes for a rough reading, and often a day or more of continuous running before the baseline stops drifting; a brand-new sensor may need a longer burn-in still.
• It is a hot element near flammable gas. Certified detectors handle this with enclosure design and flame arrestors, which is a real difference between a certified alarm and a bare module.

A metal-oxide sensor does not answer "is this gas present?". It answers "has the oxygen on my surface been disturbed, and by how much?" — and every limitation these devices have follows from the gap between those two questions.

Selectivity is the weakness, and it is fundamental

Because the mechanism responds to anything that consumes adsorbed oxygen, a plain tin oxide sensor responds to carbon monoxide, hydrogen, methane, LPG, ethanol, acetone, cleaning products, perfume, cooking fumes and the solvent coming out of new furniture. It cannot tell them apart, because the underlying measurement makes no distinction between them.

Manufacturers push back on this in several ways, all partial. Doping the oxide with palladium, platinum or gold sensitises it catalytically and shifts its relative response between gases. Choosing the operating temperature favours some reactions over others. A filter layer — activated carbon, or a zeolite — can block larger molecules from reaching the surface. Sensor arrays with several differently-doped elements, read together by a classifier, get further still, and this is the "electronic nose" line of research.

None of it makes the device specific in the way a laboratory instrument is. That is the honest headline: an inexpensive metal-oxide sensor is a good detector of change and a poor identifier of substance.

This is exactly why the "eCO₂" reading on a cheap indoor monitor deserves scepticism. Many of those devices contain no carbon dioxide sensor at all. They measure total volatile organic compounds with a metal-oxide element and then estimate a carbon dioxide figure from it, on the assumption that human occupancy raises both together. In an empty room with an open bottle of solvent, that assumption is simply false. Real carbon dioxide measurement uses NDIR — infrared absorption at a wavelength CO₂ specifically absorbs — and a monitor with a genuine NDIR cell will say so, because it costs more.

The reason a monitor in a bad room says the air is fine

This is the behaviour that most often makes people distrust these devices, and it is deliberate rather than a fault.

Metal-oxide sensors drift. The baseline resistance in clean air changes over weeks and months as grains coarsen, as contaminants accumulate and as the heater ages. An absolute calibration made at the factory would not survive that. So most consumer modules run automatic baseline correction: the firmware watches the lowest reading seen over a rolling window — commonly a week or two — and assumes that this represents clean air, then reports everything else relative to it.

In a normal home that works, because at some point in any week a window is opened. In a room that is never ventilated, the algorithm takes the least-bad polluted air it has seen and defines that as zero. The monitor will then cheerfully report "good" in air that a reference instrument would not agree about, and it will still respond correctly to a change — spray a solvent and it will spike — while being wrong about the absolute level. A device that behaves suspiciously well in a stuffy room is usually doing exactly what it was programmed to do.

The same logic explains why these monitors are far better at ventilation feedback than at compliance measurement. Watch the trend, act on the spikes, and do not read the absolute number as though it came from a calibrated instrument.

Humidity, poisoning and the other ways they fail

Water vapour adsorbs on the same surface and also donates electrons, so rising humidity lowers the baseline resistance and looks like gas. Good modules compensate using an onboard humidity sensor; cheap ones do not, and their readings track the weather.

Silicone poisoning is the classic irreversible failure. Volatile siloxanes — from sealants, adhesives, some cosmetics, hair products and lubricants — decompose on the hot surface and leave silica behind, permanently blocking active sites. Sensitivity falls and never returns. This is why sensor datasheets warn against silicone exposure in assembly and storage, and why a detector mounted next to freshly applied sealant may quietly stop working.

Sulphur compounds and heavy contamination do similar damage more slowly. And every one of these sensors has a finite life — typically a few years — after which it should be replaced rather than trusted, which is the part of a domestic gas alarm that almost nobody acts on.

When to insist on a different technology

Metal-oxide sensing is cheap, robust, low-maintenance and sensitive across a wide range of gases, which is exactly the right combination for a leak alarm or a ventilation indicator. It is the wrong choice when a specific number matters.

• Carbon monoxide safety is the case for an electrochemical cell: it is genuinely specific to CO, works at room temperature, draws almost no power, and reads in ppm. It also has a defined lifetime, typically 5–10 years, after which the whole unit is replaced.
• Carbon dioxide means NDIR, for the reason above.
• Combustible gas at industrial concentrations is the domain of the catalytic bead sensor, which measures the heat of combustion and reads in percentage of the lower explosive limit.
• VOCs at a quantified level call for a photoionisation detector, which is a field instrument rather than a component.

For a kitchen LPG alarm, a metal-oxide sensor with proper certification is entirely appropriate, and it is what almost every domestic unit sold in India uses. The certification, the enclosure, the alarm circuit and the replacement schedule are what separate a safety device from a module on a breadboard — and it is worth saying plainly that a bare sensor wired to a microcontroller is a learning project, not a gas alarm.

Why it matters for students and researchers

This is one of the best teaching devices in applied nanoscience, because a single controllable variable — grain size relative to the Debye length — produces a large, measurable, commercially decisive effect, and the whole chain from surface chemistry to a number on a screen is short enough to hold in your head at once.

It is also an unusually honest example of engineering trade-offs. Sensitivity, selectivity, stability, power and cost pull against each other, and every commercial sensor is a specific compromise among them rather than an attempt to maximise any one. Students who work through why a manufacturer chose 350 °C, this dopant and that filter have learned more about engineering than any amount of specification-reading conveys.

The open problems are active and accessible. Room-temperature sensing — using semiconducting metal oxide composites, or two-dimensional materials — would remove the heater and the power budget with it, and is one of the more promising uses being explored for materials such as MXenes. Selectivity through sensor arrays and machine learning is a live field with real results and real overclaiming. Long-term stability against grain coarsening remains a materials problem in the ordinary sense. And calibration of dense low-cost sensor networks against reference instruments, in the humidity conditions India actually has, is a question with direct public-health consequences and far too little published work behind it.

Frequently asked questions

Why does my air quality monitor read differently from another one in the same room?

Because both are probably reporting relative to their own automatically-corrected baselines, and those baselines depend on the history each device has seen. Add differences in warm-up state, humidity compensation, sensor age and where in the room each one sits, and disagreement is expected. Compare each device against its own past readings rather than against the other one.

Is the "CO₂" number on my monitor real?

Only if the monitor contains an NDIR sensor, and it will normally say so in the specification because it is the expensive part. Otherwise the figure is "equivalent CO₂", estimated from a volatile organic compound reading on the assumption that people are the main source of both. It tracks occupancy reasonably and fails completely whenever the VOCs come from something other than breathing.

How long does a gas sensor last?

A few years for a metal-oxide element, and it degrades gradually rather than failing outright — which is the dangerous part, because a sensor losing sensitivity still reports numbers. Electrochemical cells have a defined life, usually five to ten years, and expire whether or not they have been used. Any domestic gas or CO alarm has a replacement date printed on it, and that date is the specification, not a suggestion.

Why does the reading go up when I mop the floor or spray deodorant?

Because those release volatile organic compounds, and the sensor responds to anything that consumes oxygen from its surface. It is not malfunctioning; it is doing precisely what it does. This is the same limitation that makes an unqualified "air quality" figure from a single metal-oxide element hard to interpret without knowing what happened in the room.

Can these sensors detect a real gas leak reliably?

A properly certified domestic alarm using this technology is a reasonable and widely used safety device, and its sensitivity to LPG and methane is genuine. What matters is the certification, correct placement — LPG is heavier than air and pools low, methane and biogas rise — a stable power supply, and replacement at the stated interval. A bare sensor module wired up on a bench shares the physics and none of the engineering that makes a certified alarm trustworthy.