System and method for the identification of gases and gas mixtures in the air

Abstract

The present invention relates to a system for recognizing and identifying gases and gas mixtures, hereinafter indicated as "electronic nose". (FIGURE 1)

Description

[0001] SYSTEM AND METHOD FOR THE IDENTIFICATION OF GASES AND GAS MIXTURES IN THE AIR

[0002] DESCRIPTION

[0003] The present invention relates to a system and a method for recognizing and identifying gases and gas mixtures, hereinafter also designated as “electronic nose”.

[0004] Background

[0005] Despite the growing diffusion, stimulated by the diffusion of IOT (Internet of Things) technologies, the electronic noses still have a series of problems limiting the use thereof. The current systems provide the use of several sensors to discriminate the gases (it is possible to get up to 30 sensors in a commercial electronic nose) and then, in relation to this numerousness, the main problems with the current electronic noses are the physical sizes of the device, the development of the electronics managing the numerous sensors equipping it, reproducibility, stability over time and calibration of each one of said sensors and, at last, the energy consumption in reference to possible applications of “wearable” nature. Currently there are proposals for miniaturization of the sensors, even with low energy consumption, but they require still complex measurement techniques which can be implemented in laboratory only, with expensive and sophisticated apparatuses.

[0006] Object of the invention

[0007] The aim of the present invention is to solve the problems left open by the known art.

[0008] Such aim is reached by a sensor as defined in claim Nr. 1.

[0009] An additional object of the present invention is a measurement system as defined in claim Nr. 13. An additional object of the present invention is a measurement system as defined in claim Nr.16.

[0010] The developed system is capable of providing a simple and stable method to implement an electronic nose; moreover, the system can be easily adapted depending upon needs. In fact, the sensible material can be easily functionalized and it can be used even in air not necessarily in closed and controlled environment.

[0011] Another advantage with respect to the know art is that one succeeds in developing an electronic nose by using only one reading channel, by simplifying the driving electronics and the relative electric circuits and then by reducing considerably the developing, implementing and maintenance costs (for example periodic calibration) of the device.

[0012] The object of the present invention is capable of discriminating several gases even under not optimum conditions (not only in analysis chamber or in controlled environment), by replicating the operation of the electronic nose by using only one reading channel.

[0013] This makes that, in order to discriminate among several gases in a complex mixture to be analysed, the use of an array of sensors is not required, but only one suitably driven sensor is sufficient. Moreover, the sensitive layer of the sensor is easy to be functionalized for optimum performances in the different application areas.

[0014] Additional features of the present invention are defined in the corresponding depending claims.

[0015] The advantages, together with the features and the use modes of the present invention, will result evident from the following detailed description of its preferred embodiments, shown by way of example and not for limitative purposes.

[0016] Brief description of figures

[0017] Hereinafter in this description the drawings shown in the enclosed figures will be referred to, wherein:

[0018] Figure 1 is a block diagram exemplifying an apparatus according to the present invention; Figures 2A and 2B show schematically the structure of a FET transistor according to the invention;

[0019] Figure 3A describes the principle for using the apparatus as electronic nose;

[0020] Figure 3B shows an example of the gate voltage values applied over time during an exposure to gases or during a calibration measurement;

[0021] Figures 4A and 4B illustrate the dependence between gate voltages and the source-drain current;

[0022] Figure 5 illustrates an example for collecting data after exposure to a polluting gas (ethanol), for three different concentration values of ethanol in air);

[0023] Figure 6 illustrates the result of a t-SNE analysis on the classification capability of the device after exposure to different gases in air;

[0024] Figure 7 illustrates the accuracy in classifying a gas with respect to the number of characteristics, that is “features” extracted from the measurements of sensor exposure to gas mixtures, (a) Results starting from data classified with t-SNE algorithms; (b) Results starting from data classified with ll-MAP algorithms. The accuracy is measured based upon a neural network whose first layer of neurons corresponds, each time, to the number of features taken into consideration;

[0025] Figure 8 illustrates a comparison between the relative response for the same sensible layer of a NiPc-Graphene-Si3N4-Si sensor obtained in the chemiresistive configuration (a) and in the one of GFET (b) as a function of the ammonia concentration (NH3) thereto the sensor is exposed. At 10 ppm of exposure the response in (b) is equal to about 20 times the one measured in (a).

[0026] Detailed description of the invention

[0027] The present invention will be described hereinafter by making reference to the above-mentioned figures.

[0028] By way of introduction, a field effect transistor is a device having three terminals: source, drain and gate. Source and drain are connected by a channel, whereas the gate is isolated from the channel through a dielectric. These devices allow to apply a potential difference both between the source and drain contacts and between the source and gate contacts. By applying a potential difference between source and gate a current is generated flowing in the transistor channel, and the gate voltage is used to modify the density of electrons and then to modulate the resistance of the transistor channel.

[0029] Figure 2A shows the sequence of the different layers of the transistor: from bottom upwards: gate electrode, n-type silicon wafer constituting even the mechanical substrate of the apparatus, Si3N4insulating layer, graphene layer, source and drain electrodes.

[0030] Figure 2B shows the sequence of the different layers of the functionalized graphene-based transistor: from bottom upwards: gate electrode, silicon wafer, Si3N4insulating layer, functionalized graphene layer, source and drain electrodes.

[0031] With reference to Figures 2A and 2B, the developed system consists of a substrate having thickness of 700 pm which divides into two levels, a first layer is in contact with the gate and it consists of (n-Si) n-doped monocrystalline silicon, the second layer consists of silicon nitride ( Si3N4) and preferably it has a thickness of about 200 nm. The Si3N4layer can be deposited for example with PECVD techniques.

[0032] A third layer, forming the transistor channel, preferably consists of graphene. Source and drain of the FET transistor are formed on said third layer. In the herein described embodiment the used graphene was deposited by CVD (Chemical Vapour Deposition) technique on copper and then transferred on the silicon nitride layer. The silicon nitride has an important role because, thanks to its dielectric properties and to the suitable selection of its thickness, decouples the graphene from the gate, by allowing to polarize the source-drain channel with a control voltage. As it will be illustrated hereinafter, this voltage allows to drive the FET like an electronic nose.

[0033] As it can be deduced by the data shown in Figures 7-8, the authors of the present invention found that the use of silicon nitride ( Si3N4) as second dielectric layer is particularly advantageous since it allows reducing possible micro-short circuits, by providing more reliability and stability in reading the signal, as well as the possibility of applying - in the reading phase - wider voltage ramps (up to a maximum of 100 V).

[0034] Other dielectric layers can be used between silicon wafer and graphene. The most common one is silicon dioxide (SiO2), which can be obtained, in the simplest way, by oxidation in controlled atmosphere of the same silicon wafer. Even a layer of AI2O3 alumina can be used, deposited by radio frequency sputtering (RF sputtering) or by ALD (Atomic Layer Deposition). Another possible selection is to use a layer of SiC silicon carbide, deposited above the n-Si wafer. In this case the graphene can be grown directly on the SiC, or transferred on the same as in case of Si3N4.

[0035] Potentially even the graphene could be replaced by another 2D material. However, it is necessary to select the 2D material so that it could guarantee a low electric resistance along the whole channel. The graphene (or other 2D material with high mobility peof the electric charges (up to pe~ 2 x 1 o5cm2V-1s-1at 300 K) is then functionalized with phthalocyanine powders, preferably with nickel phthalocyanine powders (NiPc).

[0036] The functionalization can be performed through dropcasting, that is through the deposition of a solution of phthalocyanines in suitable solvent (for example 0.1 mM acetone), which once evaporated leaves as residue a layer of phthalocyanines directly on the solvent. This technique is very practical and allows to prepare quickly different layers of phthalocyanines for the prototyping phase of the device. There are also other techniques called PVD (physical vapor deposition) which allow to deposit the functionalizing layer under controlled conditions of high vacuum or ultra-high vacuum (at least 10'7mbar) through electronic bombardment or powder sublimation.

[0037] There are also alternatives to the Ni phthalocyanines as far as the functionalization is concerned: phthalocyanines of other metals (for example Co, Fe, Cu, with suitable selection of the solvent), other molecules with macrocycle structure (for example porphyrins), molecules which form covalent bonds with graphene (for example diazonium salts), or metal nanoparticles (Pt, Au, Pd among the most widespread ones) can be used.

[0038] The graphene functionalized with phthalocyanines constitutes the sensitive part of the system: the interaction with the gases to be monitored or with the atmosphere at which the device is exposed determine a variation in the channel conductivity after charge transfer phenomena among gases and the sensitive layer or polarization of the sensitive layer after events both of physisorption of molecules and chemisorption of molecules. In this perspective, the system is configured like a chemirestore wherein the current flowing from source to drain can be modulated by the gate potential to activate the FET functionality.

[0039] The source and drain contacts can be implemented by silver paste (silver paint or conductive paint), that is through deposition on the sides of the device of two opposite layers of a commercial preparation based on Ag powders diluted in suitable solvent. Alternatively, the contacts can be implemented through PVD techniques (sublimation, electronic bombardment, or radiofrequency sputtering) from target of metals or metal alloys.

[0040] The gate preferably consists of a thin layer of gold, deposited on the PCB (printed circuit board) board whereon the device electronics are housed, which is placed in direct contact with the first level of the device, that is the n-Si wafer.

[0041] The transistor has lateral sizes no larger than 1cm x 1cm. The lateral sizes are important for an effective collection of the currents generated after interaction with gases. The reduction of such sizes in most cases determines an increase in the system effectiveness. The contact with the gate can be optimized by depositing with PVD techniques, even on the n-Si wafer, a metal layer which, in turn, will be placed in contact with the layer of gold arranged on the PCB board.

[0042] The collected source-drain current data are analysed through machine learning algorithms. Such algorithms allow to analyse the acquired data with the purpose of determining the presence of specific molecules in the gaseous sample being analysed or to classify according to predefined parameters the sample as belonging to one or more classes. In terms of machine learning analysis, the device performances are optimized by a phase for training the algorithms, preceding the one for identify ing / classifying the data.

[0043] PCA and / or t-SNE algorithms can be further used as algorithms to reduce the sizes and different clustering algorithms (clustering for dendograms, decision trees, k-nearest neighbour) and neural networks.

[0044] For descriptive completeness a description of the main usable algorithms is reported. PCA (Principal component Analysis) is the size-reducing algorithm most used among those reported in literature. At first one has to detect a hyperplane onto which the data have to be projected. Usually, one selects the projection which preserves the maximum variance. Then, PCA identifies the hyperplane which preserves the greatest variance portion, then one selects other hyperplanes perpendicular to the first selected (as many as the dimensions of the space). The i-th hyperplane is called main i-th component. In order to be able to select the best projection hyperplane the data are firstly centred around the origin by subtracting the average. Through this process, then, the data can be represented through a matrix M mapping the data in the plane through the following formula: wherein x is the average valu e:

[0045] If we are using a q-dimensional space M is a matrix q x m. The problem to find the best hyperplane to project the data, then, reduces to a problem to the eigenvalues. Let’s consider the matrix M: wherein are the normalized eigenvectors of the covariance matrix of the data: for q eigenvalues

[0046] The vectors are called main components. A good practice is standardization, which is obtained through the transformation: Wherein and are the average value and the standard deviation of the attribute X, respectively.

[0047] The second size-reducing algorithm is t-SNE (t-stochastic neighbour embedding), this algorithm was never applied to the electric noses in literature. The operation principle of this algorithm can be summarized in these five steps:

[0048] 1 . As first step it is necessary to convert the distances between the points collected in the starting space in conditional probabilities. For example, let’s take two points, xtand xj, in the original space and let’s suppose to wish to calculate the probability (pij) that xjis selected as neighbour of xj. By using a Gaussian curve, it is possible to define the probability that it will be greater for the neighbouring points and smaller for the distant points:

[0049] Wherein:

[0050] 2. In order to be able to control the selection rate of the neighbours, the width of the Gaussian curve has to be controlled, the wider the curve is, the more far points are taken.

[0051] 3. By projecting the points in a space with smaller size it is possible even to calculate the probability qij) by using the projections of xiand of xjwhich are called yiand yj, respectively.

[0052] 4. This algorithm tries to minimize the differences between pijand qijwhile it positions the data. The function used for minimization is a cost function called Kullback-Leibler divergence:

[0053] In order to minimize C an algorithm called Gradient descent is used. Sometimes the Student’s t-distribution is used instead of the Gaussian curve in the space with smaller size (

[0054] With reference to figure 3A, this shows a possible calibration form of the apparatus which provides the application of a gate voltage (for example from -10

[0055] V to +10 V, with AV = 1 V) by performing a measurement of the source-drain current in the pre-exposure phase. This step is recommended to have a reference “zero” for the subsequent measurements.

[0056] As previously mentioned, the use of silicon nitride as second dielectric layer in the sensor allows to be able to apply a higher gate voltage, for example from -50

[0057] V to + 50V, up to values from -100 V to +100V, with AV = 1 V. This advantageously allows to be able to measure, directly or indirectly, several physical quantities and then to be able to extract several features for each gas detection performed with the sensor, so as to allow a more discrimination thereof.

[0058] After exposure to the atmosphere to be monitored, the same sequence of measurements is performed as the gate potential varies.

[0059] Figure 3B shows an example of the gate voltage values applied overtime during an exposure to gases or during a calibration measurement.

[0060] For the different voltage values the difference in the measured value of current between the phase of gate voltage equal to zero and the phase with applied voltage (FIGURE 4A) is determined. From the difference between such current values the l-V graph is reconstructed with the current values obtained for the different, positive and negative, applied voltages. This procedure is repeated for each exposure.

[0061] Figure 5 illustrates an example for collecting data after exposure to a polluting gas (ethanol, for three different concentration values in air). For each exposure the current data constitute a line in the table of data. Each voltage value identifies a column. The data table will be subject to multivariate statistical analysis.

[0062] With such data tables are generated which are subject to multivariate analysis or, more generally, to machine learning approaches. In these tables the collected data are organized by lines (1 line = 1 exposure, in the specific case a line for each ethanol concentration is provided) and columns (1 column = 1 value of gate voltage; in the specific case there will be 21 columns for the 21 voltage values used in the range -10, +10 volt). A table example is shown in the lower panel of Figure 5.

[0063] Figure 6 shows the result of a t-SNE analysis of the classification capability of the device after exposure to different gases in air. The t-SNE analysis was performed on a table of data collected and organized as described previously. It can be noted that different gases have neighbouring coordinates (V2, V1) in the space of the two variables V1 and V2. Such closeness determines the formation of clusters (highlighted with ellipses, in the graph). If the clusters do not overlap, the device is said to be capable of classifying the different gases thereto it was exposed.

[0064] Figure 7 shows the accuracy obtainable by using a sensor according to the present invention, comprising silicon nitride as second dielectric layer and a third layer of graphene functionalized with nickel phthalocyanine powders (NiPc- Graphene-Si3N4-Si). The data show that the specific configuration of the sensor devised by the authors of the present invention allows, by reading the current between the source-drain terminals as a function of the gate potential, to extract an arbitrarily high number of physical quantities associated to a specific gas. In particular, it was found that the sensor comprising silicon nitride as second dielectric layer allows to be able to extract between 15 and 60 different features for each type of gas, by using one single reading channel. This number can be optimized even based upon the type of gas to be monitored and discriminated, that is based upon the specific application field. This allows, depending upon the application type, to be able to select the number of features which maximize the accuracy of the results in classifying different gases inside an even complex mixture.

[0065] The data reported in Figure 8 show the response superiority of the NiPc- Graphene-SisN4-Si sensor in GFET mode with respect to a chemiresistive reading. In fact, the response provided by the sensor after exposure to 10 ppm of ammonia results to be about 20 times higher in GFET configuration (Figure 8b) with respect to a chemiresistive configuration (Figure 8a).

[0066] The present invention has been sofar described with reference to preferred embodiments thereof. It is to be meant that each one of the technical solutions implemented in the preferred embodiments, herein described by way of example, can advantageously be combined, differently from what described, with the other ones, to create additional embodiments, which belong to the same inventive core and however all within the protective scope of the herebelow reported claims.

Claims

CLAIMS1. A sensor for detecting a plurality of gases in the air in an environment, comprising a field effect transistor (FET) comprising a first layer of n-doped monocrystalline silicon n (n-Si) in contact with the gate of the FET, and a second dielectric layer superimposed on the first one, wherein the FET channel comprises a third layer made of 2D material with high mobilityof the electric charges formed on said second layer, said third layer being functionalized with phthalocyanine powders, wherein source and drain of the transistor (FET) are formed on said third layer.

2. The sensor according to claim 1 , wherein said first layer has a thickness of about 500 pm.

3. The sensor according to claim 1 or 2, wherein said second layer has a thickness of about 200 pm.

4. The sensor according to any one of the preceding claims, wherein said second layer is a layer of silicon nitride (Si3N4, silicon dioxide (SiO2) or silicon carbide (SiC).

5. The sensor according to any one of claims 1 to 4, wherein said second layer is a layer of silicon nitride (Si3N4) .

6. The sensor according to any one of the preceding claims, wherein said third layer made of 2D material has a mobility of the electric chargesat 300 K.

7. The sensor according to any one of the preceding claims, wherein said third layer made of 2D material is a graphene layer.

8. The sensor according to any one of the preceding claims, wherein said phthalocyanines are nickel phthalocyanines (NiPc).

9. The sensor according to any one of claims 1 to 7, wherein said phthalocyanines are phthalocyanines of a metal selected from: cobalt, iron, copper.

10. The sensor according to any one of the preceding claims, wherein said gate is made from a layer of gold.

11. The sensor according to any one of the preceding claims, wherein said transistor (FET) has maximum dimensions of 1cm2.

12. The sensor according to any one of the preceding claims, further comprising a metal layer deposited on the face of said first layer in contact with the gate of said transistor (FET).

13. A system for detecting a plurality of gases in the air in an environment, comprising a sensor according to any one of the preceding claims and a transistor (FET) gate voltage control apparatus, for acquiring and processing data of sourcedrain current of the transistor (FET) relative to each voltage applied to the gate; said control apparatus being programmed to process said data by means of machine learning algorithms configured and trained to recognize a plurality of gases as a function of said data.

14. The system according to the preceding claim, wherein said control apparatus is configured to apply in sequence to the gate of the transistor (FET) an increasing voltage in the range -100V I +100V, with a step of 1V.

15. The system according to any one of claims 13 or 14, wherein said control apparatus is configured to apply in sequence to the gate of the transistor (FET) an increasing voltage in the range -10V I +10V or in the range -50V I +50V, with a step of 1V.

16. A method for detecting a plurality of gases in the air in an environment by means of a system according to any one of claims 13 to 15, comprising the following steps: - applying to the gate of the transistor (FET) an increasing voltage in a predefined range, with a predefined step; at each step measuring and acquiring the value of the source-drain current of the transistor (FET). processing said data by means of machine learning algorithms configured and trained to recognize a plurality of gases as a function of said data.

17. The method according to the preceding claim, wherein said predefined range is between -100V and + 100V, in particular between -10V and +10V or between -50V and + 50V.

18. The method according to claim 16 or 17, wherein said predefined step is 1V.

Images