The use of filters (typically activated carbons) to protect gas sensors is well documented. Filters remove the possibility of poisons reaching the sensor and can also remove a response from potential interferent substances. The interferent is adsorbed on the filter and does not diffuse through it, to the sensor. When filters are used with metal oxide semiconductor (MOS) sensors, which operate at elevated temperature (300-500°C), they generally experience a temperature profile, due to the close proximity of the base of the filter to the sensor element (typically a few millimetres to-15mm). The capacity of the filter is finite and decreases as the temperature increases. Calculation of capacity depends on interference gas and temperature as well as the nature of the carbon and can be determined by the empirical Dubinin-Radushkevitch equation. This allows available capacity for a given carbon and temperature to be estimated from a knowledge of the molecule, its vapour pressure at that temperature and its threat concentration. The capacity decreases with decreasing threat concentration and increasing temperature.

Conventional gas sensor filters have two disadvantages. Both are associated with their finite capacity. Once full, the interfering substance diffuses through the filter without adsorption. In the vocabulary of molecular filters, this is known as"breakthrough". The greater the capacity of the carbon, the longer breakthrough is delayed, for a given threat. The rate of diffusion through a saturated filter is determined by the concentration gradient, but the rate of diffusion through the filter is at a rate reduced approximately by the volumetric filling factor of the filter material. Since this is typically ~70%, the effective diffusion rate is only slightly lower than it would be if the filter were not present.

In practical terms, the saturated filter provides negligible improvement in sensor

specificity. This weakness is evident whether the sensor is operating (and the filter therefore experiences a temperature profile), or when it is cold. In the latter case, the sensor would respond to the interferent immediately on being made active by heating.

With heated MOS sensors, a more subtle disadvantage of a conventional filter is evident when it is partially contaminated with an interfering substance, and particularly when at a lower temperature than that when the sensor is operating normally. When the sensor is powered up, filter capacity decreases as its temperature rises. Desorption of the interferent results in its diffusion both towards and away from the sensor, resulting in a sensor response which typically first increases in time, and then slowly decreases as the filter comes to equilibrium with an otherwise clean atmosphere. This phenomenon is known as "backfilling". A simple calculation shows that when the filter is-2/3rds full with a particular threat, backfilling on power up is inevitable. Paradoxically, a larger capacity filter, or one with a greater dependence of capacity on temperature, can give rise to greater backfilling. In practice, backfilling of a cold filter by a chemical contaminant has been shown to be inevitable, and backfilling can thus be seen as the governing parameter in respect of the overall performance of a carbon filter when employed with a heated MOS sensor.

The invention is defined by the claims hereinafter.

In the drawings, given by way of example, the response of various sensors to a number of gases is shown in the presence or absence of a number of catalytic filters : Figures 1 to 7 show the response (resistance against time) for a chromium titanate sensor, and Figures 8 to 10 show the response for a tin oxide sensor.

We have found, surprisingly, that if a high surface area powder which is a metal oxide of a composition useful because of its gas sensing properties, is used in conjunction with the activated carbon granules or cloth, the capacity of

the resulting filter is dramatically improved. The metal components in the oxide are First Order Transition Elements and/or one or more of La, Ce, Pr, Zr, Ta, Hf, Mo, W, Al, Si, Sn, Pb and Bi. For example Figure 1 illustrates the beneficial effects of both chromium oxide and chromium-titanium oxide"powder-fills"on the ability of the composite filter to withstand the threat posed by 600 ppm ethanol for 8 hours to a Cr-Ti-O based semiconductor gas sensor. Figure 2 illustrates for the case of the chromium-containing oxide fills around a Cr-Ti-O gas sensor, the significant reduction in both the degree and duration of backfilling after the filter has been contaminated, when at room temperature, with 100ppm ethanol for 1 hour. Figures 3 and 4 illustrate in the absence of the activated carbon (similar to saturated carbon) the protection conferred by various oxide"powder-fills"to a CO gas sensor in the presence of flowing air containing 300ppm ethanol over an 8 hour period. In no case is the sensor response to CO effected by the respective"powder-fills". Figures 8,9 and 10 give further support to the protective properties of"powder-fills"for the case of a general hydrocarbon sensor based on a SnO2 sensing element. By the term "high-surface area powder"we mean a powder having a powder density less than 20% of the bulk density and/or a specific surface area of greater than 0.5m2/g. In Figures 1,2,3,8-11 the chromium oxide had a specific surface area of about 40m2/g and for the chromium titanium oxide in Figures 1 and 2, a specific surface area of 5m2/g. Figure 5 demonstrates clearly the reduced protection provided by a"powder-fill"with a lower specific surface area 2m2/g.

Equally, the"powder-fills"can function by enhancing the response of a gas sensor to a target gas, thereby increasing its specificity. Figure 7 shows how the response of a Cr-Ti-O gas sensor to hydrogen relative to CO is enhanced when a CuCr204 and W03 fill is used. Figure 11 illustrates similar benefits for a SnO2 sensor where the response to propane is enhanced in the presence of a Cr203"powder-fill".

It is believed that the mechanism by which these powders exert their

beneficial effect is by chemical reaction with the respective gases rather than by thermal means, in the sense, for example, of reducing the filter temperature or by physical means such as impeding the diffusion of the gas to the sensor surface. Such powders are known to exhibit catalytic activity (and it is indeed this property which, linked to a favorable bulk resistivity change, can give rise to a measurable response). Not all high surface area metal powders provide desirable effects. In Figure 6, results are shown for powder fills which react with the target gas, in this case CO while appearing to provide only physical protection (as indicated by the ascending response curve) against an ethanol threat.

Generally, the activity for heterogeneous catalysis of an oxide powder is a property of its chemical composition, atomic surface condition, particle size and the density of the packed bed used to characterise it. For a given molecular reaction, there is typically a temperature (the"light-off"temperature) at and above which a particular reaction occurs. In the case of a powder filling the gap between a sensor element and a carbon filter, the powder has a distribution of temperature dependent on distance from the heated element, and the temperature of the sensor element (typically 400-500°C). As a consequence, it should be possible for the powder to oxidize and/or crack a range of reducing gases before they are able to diffuse through the powder bed to the sensor element. The products of these reactions are either undetected or weakly detected by the MOS sensor element (CO2, H20, CH4, for a Cr-Ti-O based CO sensor) in which case protection is achieved, or, more easily "sensed"than the parent gas, in which case enhancement is achieved. As mentioned previously, the powder is a catalytically active material which may or may not be useful per se for resistive gas sensing devices.

To provide effective protection, the properties of the powder material and its distribution of temperature should be such as to react with interfering substances, whilst affecting only to a minor degree the response of the sensor

to the molecular species of interest. The powder might oxidize reducing gases such as alcohols, solvents, volatile organics and other organic molecules, or reduce oxidizing gases such as ozone, oxides of nitrogen and chlorine.

Molecules of interest are typically small and relatively stable molecules, such as hydrogen, ammonia, carbon monoxide, carbon dioxide, methane and other light hydrocarbons. It is also clear that the specificity of the response of the gas sensor can be tuned by reference to the target gas, choice of operating temperature, nature of the powder fill material and device geometry.

A further benefit of the presence of the activated carbon outer later of the filter is that it reduces the possibility of poisons reaching the powder fill, and eventually the sensor. The lifetime of both elements is, as a consequence, enhanced.

Features of the invention are: Powder fill between sensor chip and carbon filter, latter protecting chip interfering substances such as ethanol, solvents and VOCs.

Powder fills of chemical compositions useful for gas sensing e. g. chromium oxide, chromium titanium oxide, tin oxide, titanium dioxide.

Powder fill which is catalytically active, but not necessarily useful for gas sensing.

Powder fill chosen to incorporate one or more materials so as to tune the specificity of the response of the gas sensor.

Powder surrounding the sensor chip, heated by it, experiencing temperature gradient so as to cool base of carbon filter.

Temperature gradient includes range of temperatures for which compositions are known to exhibit a gas response or the ability to exhibit a capability for heterogeneous catalysis.

Powder fill surrounding a gas sensor so as to reduce the ingress of substances which might poison the response of the gas sensor. improvement in the specificity of hydrogen, ammonia, carbon monoxide,

carbon dioxide, methane and other light hydrocarbon gas sensors.

The following examples illustrate the invention: Example 1 A semiconductor sensor based on the Cr-Ti-O system as the sensing oxide, as described in WO 95/00836, was used in this example. The Cr-Ti-O system is a p-type material and undergoes an increase in electrical resistance in the presence of reducing gases. The sensor build is described in various Capteur Sensors'product data sheets for the G series of sensors for detecting carbon monoxide, for example GS07 and GL07 data sheets. The unfired thickness of the Cr-Ti-O oxide layer is 40 microns. Prior to applying the filter cap containing activated carbon, the cap was inverted and the deadspace between the brim of the cap and the stainless steel gauze was filled with 2 different oxide powders to 2/3 of the volume. These oxides were Cr203 and Cr-Ti-O, respectively. The sensor base was inverted and clipped on to the filter cap so that on standing upright the oxide powder filled the sensor chip compartment. The sensors were powered up alongside standard sensors to approx. 400°C such that the electrical resistances were 175 kohms in 50% RH clean air. Following stabilisation over a period of 12 hours, the sensors were exposed to 600ppm ethanol for 8 hours in 0% relative humidity. While the unfiltered sensor exhibited a resistance increase of 2500 kohms, the standard filtered sensors increased resistance 500 kohms, as shown in Fig. 1. However, the"powder-fill" sensors showed negligible increases in resistance, i. e. did not sense the presence of ethanol.

Example 2 As for Example 1 with power switched off for 1 hour following sensor stabilization. For the duration of this 1 hour period, the sensors were exposed to 1000ppm ethanol in air at 0% relative humidity. Figure 2 shows the effect of "backfill"of ethanol from the contaminated filter caps onto the sensor chip on subsequent powering up of the sensors. The benefit of the powder-fills in

suppressing"backfill"is clearly demonstrated.

Example 3 Semiconductor sensors as in Example 1 except here, the active sensing layer of Cr-Ti-O was printed to an unfired thickness of 90 microns and the sensors powered up to achieve a sensor resistance of 100 kohms (equivalent to a heater temperature of approx. 400°C). The activated carbon was removed from the filter caps prior to filling with powder fill in the normal way so that protection against the contaminant gas was provided only by the powder-fill. The stabilization period was 20 minutes. The oxides used as"powder-fills", as listed in Table 1, were chosen because of their catalytic activity towards ethanol while being inert to CO. A typical response to 400ppm CO falls within the range 250- 350 kohms. The grades of Cr203 powder used, as shown in Table 1 is different from the grade used in Example 1 and 2.

Table 1 Oxide Surface Area Figure (m2l9) Cr203 71 3,4,5 (a) & (b) Cr203 21 5 (a) & (b) (coarse) Fe203 581 3 (a) & (b) (Cr, Fe) 203 121 3 (a) & (b) CoA1204 111 4 (a) & (b) Sn02 81 4 (a) & (b) La203 21 4 (a) & (b) Pr203 0 9+1 4 && In Figure 5 (a) and (b), the effect of surface area and particle size on the catalytic activity and, hence, protective capability of Cr203 is demonstrated.

Example 4 As for Example 3, except in this case powders unsuitable to improving sensor specificity to CO are used. Fig. 6 (a) highlights the suppressant effect such oxides have on the CO response, via chemical reaction with CO (less pronounced for Hf02). Furthermore, protection against an ethanol threat would appear to be temporary, as indicated by a rising response curve. Such behaviour is consistent with a physical adsorption interaction with the contaminant. A shorter exposure period of 4 hours in 300ppm ethanol was used for the HfO2"powder-fill"sample. Irrespective of the effect on CO, these "powder-fills"would not provide long-term protection against ethanol.

Table 2 Oxide Surface Area Figure (m219) Ti02 552 6 (a) & (b) AI203 200+2 6 (a) & (b) Hf02 141 6 (a) & (b) Example 5 As for Example 3, but here the sensor is used to detect hydrogen and is thus required to be less sensitive to CO and/or more sensitive to hydrogen. Oxide Surface Area Figure (m, ig) CuCr204 61 7 W03 21 7 Example 6 A semiconductor sensor based on Sn02 as the sensing element, as described in Capteur Sensors product data sheet for the WL01 hydrocarbon sensor, was used in this example. Because, the SnO2 system is an n-type

material, a reduction in electrical resistance occurs in the presence of reducing gases. The sensors are set up to a temperature of 475°C which results in a resistance of 15 kohms 5 kohms. For this example, the"powder-fill"was chromium oxide, as used in Figures 1,2,3 and 5. Figures 8,9 10 illustrate the protection bestowed to the sensor against interference from hexamethadisiloxane (HMDS) which is a common sensor poison, NO and SO2 respectively. The response to methane is unaffected by the presence of the "powder-fill".

Example 7 This example serves to demonstrate the enhancing effect of a chromium- oxide"powder-fill"in the detection of propane by the WL01 sensor, as shown in Figure 11.