A recent study revealed that humans can discriminate among more than a trillion different smells [1
] and the mammalian nose remains the primary “apparatus” used in many applications to evaluate the smell of products. Despite the recent progress in research in the field, the mammalian olfactory system is complex and mechanisms of olfaction are still not fully understood [2
]. Driven by the needs of odour detection for medical applications, environmental monitoring, security or food monitoring, the development of electronic noses has increased over the years [3
]. These sensing technologies operate by mimicking the manner that mammalian noses proceed to discriminate odorant volatile compounds. The first study in this field [10
] reported that a system aiming to mimic the mammalian olfactory system may be composed of two main elements: roughly tuned receptor cells, not selective toward specific odorant molecules and a system capable of performing parallel processing of the output signals. The processing may include qualitative analysis of sensor signal reports, by using, for example, pattern recognition techniques. Since then, various electronic noses have been developed based on different sensor technologies and different identification and classification methods [11
Multi-gas detection using portable systems requires the use of sensors adapted for this specific application, especially in terms of sensitivity, response time and recovery time, selectivity, application to a wide range of gases, simplicity and convenience with respect to the use and replacement of sensors, and equally important, the sensor life-time. To meet this demand, a large number of technologies, based on optical, mechanical and electrical techniques, have been developed for chemical transduction.
Regarding the choice of these sensor techniques, solutions are distinguished by the type of sensitive layer and the principle of transduction. The most used for electronic nose applications are: semiconductor metal oxides gas sensors, conducting polymer sensors [12
], Surface Acoustic Wave (SAW) sensors [13
], Quartz Crystal Microbalance (QCM) [14
] and optical fibre sensors [15
]. Gas sensors differ in size, sensitivity, operating temperature, response and recovery times.
Electronic noses technologies offer a cheaper alternative to existing analytical instruments such as gas chromatography, mass spectrometry or ion mobility spectrometry [16
]. They are supposed to be an alternative mobile or transportable and easy to use. In this context, the size and the number of sensors are important parameters, not only to obtain a smaller instrument but also to promote the use of smaller gas volumes, smaller detection surfaces and shorter detection times.
In the last two decades, advances in the micro-electromechanical systems (MEMS) field have promoted the development of miniaturized sensors that are able to transduce mechanical energy (e.g., gravitational potential energy) to electrical energy. The operation principle of MEMS sensors is that chemical, physical or biological stimuli can be transduced to mechanical stimuli and affect mechanical characteristics of the sensor structure in a manner that these changes can be measured by electrical or optical means. In this context, bio-chemical detection is possible by measuring mass or surface stress changes. In particular, microcantilevers, the simplest MEMS structures, offer the possibility of label-free biochemical detection with very high sensitivity [17
]. The sensitivity of resonant microcantilevers is related to the dimensional scale of these devices.
In this perspective, a microcantilever-based electronic nose can offer highly desirable characteristics, including fast responses, height sensitivity and being able to accommodate a large number of sensors in a small volume. Moreover, they are suitable for mass production, taking advantage of micro-machining techniques and circuit integration.
In this work, we report the fabrication and the development of a silicon and synthetic diamond microcantilever array-based electronic nose. In the MEMS field, silicon is widely used. In fact, processing methods such as etching and photolithography have been thoroughly developed by the electronics industry, and have been easily adapted for MEMS production. As a consequence, the development of new techniques has not been favoured. Nonetheless, diamond is expected to be a very promising alternative in the micro-sensors field. In fact, diamond is a highly suitable material for the manufacture of resonant microcantilevers because of its exceptional mechanical and thermal properties, biocompatibility as well as excellent hardness and robustness. Polycrystalline diamond can be an excellent choice for the fabrication of resonant sensors due to its high elasticity modulus (in the order of
]). Moreover, because of its carbon nature, this material is convenient for stable grafting of a wide range of bio-receptors by covalent C–C binding [19
The originality of the study reported in this paper is related to the development of a complete modular and autonomous system which is designed from sensors (silicon and diamond) to signal processing to be low noise, sensitive and easy-to-operate. In order to increase the sensitivity of our microcantilever sensors to volatile organic compounds detection, a variety of polymer coatings has been used to coat microcantilever surfaces. These sensors present mass resolution down to the ng range. Finally, we report the successful application of this electronic nose approach to discriminate some volatile organic compounds.
In this experiment, the detection of vapors is achieved by the diffusion of the analyte into the polymer layer and also by surface interaction with diamond sensors (bare cantilevers). As the operation mode for this application is the dynamic mode, changes in the polymer layers lead to an increase of sensor total mass, resulting in a negative frequency shift. Shifts of resonance frequency of each microcantilever are specific to the interaction between vapor molecules and the polymer. When we change to a flow of pure nitrogen in the same temperature and flow rate conditions, the trapped analyte starts to diffuse out of the sensitive layer, back to the environment, causing a decrease of mass and a positive resonant frequency shift. Figure 12
presents some examples of cantilever-array responses in dynamic mode. Considering the frequency shift as the system output signal, the signal-to-noise ratio will depend on the sensitive layer and the substance to be detected. However, in any case, the estimated noise level (fluctuation of the output signal when the system receives reference gas) is 4 Hz.
For each vapor tested, we have generated different concentrations in order to verify the linearity of the microcantilever sensors. Each concentration was also repeated at least three times in order to ensure that the response is reversible as shown in Figure 12
This procedure was repeated for 13 substances : Toluene; Styrene; Pentanal; Octanal; Hexanal; Ethanol; 2-Methyl-1-propanol; Butanol; Benzadehyde; Acetone; 6-Methyl-5-hepten-2-one; Phenyl acetate; Isopropanol. Each vapor was generated at 500 ppm and sensor response (resonance frequency shift) was measured. Relative response patterns for 12 of 13 vapors are shown in Figure 13
. The best sensitivity was estimated for the couple “Ppy-Phenyl Acetate” and is about 1.62 Hz/ppm. Considering the noise level of 4 Hz, the limit-of-detection in this case is 7.5 ppm. The relative response of a sensor (coating) is equal to the frequency shift of the sensor divided by the sum of the shifts for all eight sensor coatings. The sum of these scaled responses is the unity for each vapor, which facilitates comparisons. As sensor 2 was a diamond cantilever without coating, it was used as reference. Therefore, the response of this sensor is close to 0 Hz and was not represented in Figure 13
. As we can observe, patterns of vapors are all different from each other. From Figure 13
, we have an indication of the discriminating capability of our sensor array for this group of solvents. Some similarities can be found among the patterns from the same chemical class. For example, response patterns for ethanol and acetone are similar. One can also notice the complexity of the problem of recognizing and discriminating among more than a few vapors.
In order to identify the tested samples and evaluate the selectivity of the system, cantilever-array responses of different concentrations have been used to generate a set of “time-shift” vectors that corresponds to the dynamic development of the detection curves for each analyte. The data set was evaluated using principal component analysis (PCA) techniques, allowing to extract the most dominant deviations in the responses for the various sample vapors. In order to reduce the complexity of the analysis, we have only kept eight analytes, the best represented on the PCA first plane. As shown in Figure 14
, each VOC is comprised of a cluster without any overlap. This result demonstrates the ability of the system to discriminate a large number of VOCs in dynamic mode.