rus
Issue #1/2023
E.S.Tyunterov, V.S.Abrukov, V.A.Mukin, A.V.Smirnov, D.V.Petrov, N.I.Petrov, N.V.Aleksandrova, I.Y.Semenova
A METHODOLOGY FOR DEVELOPMENT OF THIN-FILM SYSTEMS WITH DEFINED GAS SENSITIVITY REQUIRING NO POWER SOURCES AND INTENDED FOR CHEMORESISTIVE GAS SENSING
A METHODOLOGY FOR DEVELOPMENT OF THIN-FILM SYSTEMS WITH DEFINED GAS SENSITIVITY REQUIRING NO POWER SOURCES AND INTENDED FOR CHEMORESISTIVE GAS SENSING
DOI: https://doi.org/10.22184/1993-8578.2022.16.1.22.28
A methodology for development of thin-film systems is presented, which will make it possible to obtain chemoresistive sensors with a given gas sensitivity. The methodology provides for conducting a series of experiments on the thin-film systems synthesis using various synthesis technologies (different starting materials, parameters of thin films synthesis, various methods of thin films synthesis: thermoresistive evaporation, air-drop sputtering, annealing in oxygen atmosphere), measuring their properties and characteristics, creating a database of experimental results; generalization of the dependencies contained in the experimental data using artificial neural networks in the form of multifactorial computational models. These models will make it possible to solve direct and inverse problems, extrapolate the dependencies revealed in the experimental data, and conduct virtual experiments.
A methodology for development of thin-film systems is presented, which will make it possible to obtain chemoresistive sensors with a given gas sensitivity. The methodology provides for conducting a series of experiments on the thin-film systems synthesis using various synthesis technologies (different starting materials, parameters of thin films synthesis, various methods of thin films synthesis: thermoresistive evaporation, air-drop sputtering, annealing in oxygen atmosphere), measuring their properties and characteristics, creating a database of experimental results; generalization of the dependencies contained in the experimental data using artificial neural networks in the form of multifactorial computational models. These models will make it possible to solve direct and inverse problems, extrapolate the dependencies revealed in the experimental data, and conduct virtual experiments.
Теги: chemoresistive sensors development methodology gas sensitivity metal oxides multivariate computational models neural networks single-walled carbon nanotubes thin-film systems газочувствительность методология разработки многофакторные вычислительные модели нейронные сети однослойные углеродные нанотрубки оксиды металла тонкопленочные системы хеморезистивные сенсоры
INTRODUCTION
Nowadays, gas sensing is being actively developed as it is in demand in industry at manufacturing plants, in medicine, in public safety, etc. [1, 2]. Nanomaterials are promising materials for gas sensors as they have good gas-sensitive properties.
This paper presents a new methodology for developing sensors with predetermined gas sensitivity based on thin film systems: metal oxides (MOs) + single layer carbon nanotubes (SLCNTs). Preliminary experiments [3] prove that such systems can have good gas sensing properties.
The development methodology consists of an experimental part and a part related to the generalisation of the results obtained using artificial neural networks in the form of multifactor computational models which will allow solving direct and inverse problems of the experiment, extrapolating the dependencies revealed in the experimental data and conducting virtual experiments.
As an example of the methodology application, in this paper we shall consider thin film systems based on SnO2, SnO2/SLCNT. These systems are interesting in that they can operate as sensors without a power supply, as they do not require heating. They can be used to detect such gases as methanol and ethanol.
RESEARCH METHODS
In order to generalise all obtained experimental data (not only those described above), a set of multifactor computational models has been developed using artificial neural networks (ANNs). The basics of ANNs and methods of using ANNs in modeling experimental data are presented and described in [3]. Neural networks included in the analytical platform Deductor Academic 5.3 Build 0.88 (www.basegroup.ru) – free academic version - were used to develop the models.
RESULTS
The first step in producing a thin-film system based on MO and SLCNT will be to spray a thin film of MO on the surface of a glass substrate by thermoresistive evaporation in a vacuum apparatus VUP-4. The essence of the method of thermo-resistive evaporation is as follows: MO granules are placed on the tungsten evaporator, then the evaporator is heated, the MO atoms and clusters convert into the gaseous phase and condense on the glass substrate surface in the form of a film.
The following quantities will be varied and measured: spray time, evaporator temperature, total mass of MO pellets and average pellet size. Thickness of the resulting MO film will be recorded.
The second step will be to apply SLCNT films on the thin MO film using the air-spray method. A ready-to-use SLCNT solution, an engraver and an oven will be used. The ready-to-use SLCNT solution will be pretreated in an ultrasonic bath to achieve a homogeneous solution composition which will ensure a more homogeneous application of the SLCNT film. Then the substrate with the MO thin film sprayed on it is fixed on the engraver, the engraver with the film is brought in rotation and placed into the muffle furnace heated to 100 °С. The SLCNT is sprayed onto the thin MO film using an air-drop spraying method from the prepared solution. Rotation of the graver is necessary to homogeneously spray the SLCNT onto the MO film.
The concentration of the SLCNT solution will be varied and measured. The thickness of the resulting system: MO film + SLCNT will be recorded.
The third step in obtaining an MO-based thin-film system will be annealing of the obtained samples in a MIMP-VM furnace.
During the annealing process, the following values will be varied and measured: annealing temperature, and annealing time.
The fourth stage of the research will be to measure characteristics of the obtained samples of the MO + SLCNT composite system. Topographical (surface profile), electrophysical (resistance and specific surface resistivity) and gas-sensitive characteristics will be measured: change of sample resistance while holding it in a certain gas vapour, response time and recovery time. From the surface profile (to be measured with an atomic force microscope) it will be possible to know the mutual arrangement of MO and SLCNT clusters on the surface (film solidity, maximum and minimum "height" of the surface and the standard deviation of the "height of the surface").
Based on the results of these four stages, a database of experiments conducted, including all the above parameters of system synthesis technologies and all measured characteristics of the created systems, will be generated and, using artificial neural networks (ANN), a multifactor computational model will be created, capable of identifying all dependencies contained in the experimental data, solving direct and inverse problems of experiment, extrapolating the dependencies identified in the experimental data and conducting virtual experiments.
Elaboration of such an ANN model will make it possible to develop technologies for production of sensors based on MO + SLNT systems with predetermined sensitivity.
The methodology for developing such ANN models is described in detail in [4].
Examples of the application of ANN models for solving direct and inverse problems of the experiment, extrapolation of the dependencies revealed in the experimental data and conducting virtual experiments are given in [5–13].
DISCUSSIONS
We used feed forward neural networks with one input layer (the number of neurons was determined by the number of factors), one hidden layer with different number of hidden neurons (from 5 to 8), and one output layer (with one neuron). The target functions of the models were transparency coefficient and apparent and real parts of dielectric permittivity. The model factors were as follows: temperature dependences of electrical resistivity (resistance values as a function of substrate heating temperature 25...90°C) with deposited semiconductor SnO2 and SnO2/ SLCNT films – Fig.1, 2; electrical resistivity for gas-sensitive SnO2 and SnO2/ SLCNT films as a function of soaking time in an analyte gas (ethanol vapour) – Fig.3, 4.
The resulting models were peculiar calculators (obtained by training neural networks) based on real gas-sensor measurements which allow of any set of factor values in order to determine the target function of a particular model.
It should be noted that activation energy (electric width of forbidden zone) and conductivity type are determined from graphs (Fig.1, 2), and according to these parameters as well as optical width of forbidden zone it is possible to estimate the doping degree of semiconductor films (p- and n-type conductivity) [14]. The degree of doping, respectively, influences the gas sensitivity of the sensor structures.
CONCLUSIONS
A methodology for the development of technologies for synthesis of chemoresistive thin film systems based on metal oxides and carbon nanotubes intended for producing gas sensors with a given gas sensitivity is presented. The methodology includes five steps, four of which involve experiments, and the fifth step involves the use of artificial neural networks to develop a multifactor computational model capable of identifying all dependencies contained in the experimental data, solving direct and inverse problems of the experiment, extrapolating the dependencies identified in the experimental data, and conducting virtual experiments. This methodology can be applied in the development of technologies for synthesis of thin-film systems of various applications, as well as in development of new nanomaterials. In the future, such methodology should be applied to development of the "Genome of nanomaterials" [5].
PEER REVIEW INFO
Editorial board thanks the anonymous reviewer(s) for their contribution to the peer review of this work. It is also grateful for their consent to publish papers on the journal’s website and SEL eLibrary eLIBRARY.RU.
Declaration of Competing Interest. The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Nowadays, gas sensing is being actively developed as it is in demand in industry at manufacturing plants, in medicine, in public safety, etc. [1, 2]. Nanomaterials are promising materials for gas sensors as they have good gas-sensitive properties.
This paper presents a new methodology for developing sensors with predetermined gas sensitivity based on thin film systems: metal oxides (MOs) + single layer carbon nanotubes (SLCNTs). Preliminary experiments [3] prove that such systems can have good gas sensing properties.
The development methodology consists of an experimental part and a part related to the generalisation of the results obtained using artificial neural networks in the form of multifactor computational models which will allow solving direct and inverse problems of the experiment, extrapolating the dependencies revealed in the experimental data and conducting virtual experiments.
As an example of the methodology application, in this paper we shall consider thin film systems based on SnO2, SnO2/SLCNT. These systems are interesting in that they can operate as sensors without a power supply, as they do not require heating. They can be used to detect such gases as methanol and ethanol.
RESEARCH METHODS
In order to generalise all obtained experimental data (not only those described above), a set of multifactor computational models has been developed using artificial neural networks (ANNs). The basics of ANNs and methods of using ANNs in modeling experimental data are presented and described in [3]. Neural networks included in the analytical platform Deductor Academic 5.3 Build 0.88 (www.basegroup.ru) – free academic version - were used to develop the models.
RESULTS
The first step in producing a thin-film system based on MO and SLCNT will be to spray a thin film of MO on the surface of a glass substrate by thermoresistive evaporation in a vacuum apparatus VUP-4. The essence of the method of thermo-resistive evaporation is as follows: MO granules are placed on the tungsten evaporator, then the evaporator is heated, the MO atoms and clusters convert into the gaseous phase and condense on the glass substrate surface in the form of a film.
The following quantities will be varied and measured: spray time, evaporator temperature, total mass of MO pellets and average pellet size. Thickness of the resulting MO film will be recorded.
The second step will be to apply SLCNT films on the thin MO film using the air-spray method. A ready-to-use SLCNT solution, an engraver and an oven will be used. The ready-to-use SLCNT solution will be pretreated in an ultrasonic bath to achieve a homogeneous solution composition which will ensure a more homogeneous application of the SLCNT film. Then the substrate with the MO thin film sprayed on it is fixed on the engraver, the engraver with the film is brought in rotation and placed into the muffle furnace heated to 100 °С. The SLCNT is sprayed onto the thin MO film using an air-drop spraying method from the prepared solution. Rotation of the graver is necessary to homogeneously spray the SLCNT onto the MO film.
The concentration of the SLCNT solution will be varied and measured. The thickness of the resulting system: MO film + SLCNT will be recorded.
The third step in obtaining an MO-based thin-film system will be annealing of the obtained samples in a MIMP-VM furnace.
During the annealing process, the following values will be varied and measured: annealing temperature, and annealing time.
The fourth stage of the research will be to measure characteristics of the obtained samples of the MO + SLCNT composite system. Topographical (surface profile), electrophysical (resistance and specific surface resistivity) and gas-sensitive characteristics will be measured: change of sample resistance while holding it in a certain gas vapour, response time and recovery time. From the surface profile (to be measured with an atomic force microscope) it will be possible to know the mutual arrangement of MO and SLCNT clusters on the surface (film solidity, maximum and minimum "height" of the surface and the standard deviation of the "height of the surface").
Based on the results of these four stages, a database of experiments conducted, including all the above parameters of system synthesis technologies and all measured characteristics of the created systems, will be generated and, using artificial neural networks (ANN), a multifactor computational model will be created, capable of identifying all dependencies contained in the experimental data, solving direct and inverse problems of experiment, extrapolating the dependencies identified in the experimental data and conducting virtual experiments.
Elaboration of such an ANN model will make it possible to develop technologies for production of sensors based on MO + SLNT systems with predetermined sensitivity.
The methodology for developing such ANN models is described in detail in [4].
Examples of the application of ANN models for solving direct and inverse problems of the experiment, extrapolation of the dependencies revealed in the experimental data and conducting virtual experiments are given in [5–13].
DISCUSSIONS
We used feed forward neural networks with one input layer (the number of neurons was determined by the number of factors), one hidden layer with different number of hidden neurons (from 5 to 8), and one output layer (with one neuron). The target functions of the models were transparency coefficient and apparent and real parts of dielectric permittivity. The model factors were as follows: temperature dependences of electrical resistivity (resistance values as a function of substrate heating temperature 25...90°C) with deposited semiconductor SnO2 and SnO2/ SLCNT films – Fig.1, 2; electrical resistivity for gas-sensitive SnO2 and SnO2/ SLCNT films as a function of soaking time in an analyte gas (ethanol vapour) – Fig.3, 4.
The resulting models were peculiar calculators (obtained by training neural networks) based on real gas-sensor measurements which allow of any set of factor values in order to determine the target function of a particular model.
It should be noted that activation energy (electric width of forbidden zone) and conductivity type are determined from graphs (Fig.1, 2), and according to these parameters as well as optical width of forbidden zone it is possible to estimate the doping degree of semiconductor films (p- and n-type conductivity) [14]. The degree of doping, respectively, influences the gas sensitivity of the sensor structures.
CONCLUSIONS
A methodology for the development of technologies for synthesis of chemoresistive thin film systems based on metal oxides and carbon nanotubes intended for producing gas sensors with a given gas sensitivity is presented. The methodology includes five steps, four of which involve experiments, and the fifth step involves the use of artificial neural networks to develop a multifactor computational model capable of identifying all dependencies contained in the experimental data, solving direct and inverse problems of the experiment, extrapolating the dependencies identified in the experimental data, and conducting virtual experiments. This methodology can be applied in the development of technologies for synthesis of thin-film systems of various applications, as well as in development of new nanomaterials. In the future, such methodology should be applied to development of the "Genome of nanomaterials" [5].
PEER REVIEW INFO
Editorial board thanks the anonymous reviewer(s) for their contribution to the peer review of this work. It is also grateful for their consent to publish papers on the journal’s website and SEL eLibrary eLIBRARY.RU.
Declaration of Competing Interest. The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Readers feedback
