1. Introduction
Minimization of power consumption is one of the main requirements for sensor technology, especially when targeting the development of portable systems. This is even more crucial when the sensor must work at elevated temperatures, such as metal oxide chemical sensors [
1]. Micro Electro-Mechanical System (MEMS) is a high-tech field that combines microelectronics with micromachining technology in order to integrate microcomponents, microsensors, microactuators, signal processing and control circuit [
2]. In the common silicon substrate, micro-hotplates generally consist of a thin dielectric membrane suspended over a silicon substrate. In microsystems technology, hotplates are mainly used for sensor applications [
3] where the active sensing material is deposited onto the membrane integrated with electrical stimuli and readout [
4]. With the development of MEMS technology, that offers a powerful tool to obtain low cost, high efficiency and long-term devices, micro-hotplate has gradually gained wide appreciation for gas sensors [
5], infrared emitters, actuators etc. [
6]. By employing thin films with good thermal insulation as the membrane structural material, the micro-hotplate presents a series of advantages such as miniaturized size, fast response, high sensitivity and low power consumption [
7,
8,
9,
10]. Non-stoichiometric low stress silicon nitride (SiN) grown by Low Pressure Chemical Vapor Deposition (LPCVD) is a very common choice allowing to obtain very flat suspended membranes with high electrical and thermal insulation suitable for the micro-hotplate fabrication as reported by recent literature in this field [
11,
12,
13]. The suspended membrane approach is fundamental to obtain very low heating inertia and high responsive sensors. Heaters and thermometers structures are designed on the SiN membrane to efficiently provide the thermal flux in the sensing region, where thermal driven adsorption or chemical reactions modulate the signal on the sensing elements. Most of the current literature is mainly focused on the development, modelling and optimization of specific and customized micro hot-plates mainly searching straightly uniform heat transfer on the membrane [
3,
7,
9] or, in other cases, exploiting a thermal gradient to obtain multi-sensing array [
14,
15]. Hence the existing literature usually deals with individual structures of one of the two types above mentioned, eventually with minor and limited variations of the proposed layout (e.g., shape or dimensions of the heater). The main novelty of this paper consists in providing the information and conditions for the parallel development, optimization and modelling of two micro hot-plates fabricated according to two different technological and design paradigms (i.e., buried and coplanar). The two structures are developed in parallel and with analogous characteristics since they share similar designs and dimensions, same materials and fabrication process and same heater aspect ratio. The conditions for the design of the two structures were investigated: technological implementation, critical materials properties characterization (Pt Temperature Coefficient of Resistance and TaOx adhesion layer), construction and simulation of a model with a common and widespread tool (Comsol Multiphysics
®, Comsol Inc., Burlington, MA, USA), device characterization (thermal, electrical, time response) and long-term reliability test. The parallel treatment and discussion allow to establish a clear and evident comparison and easy to implement design rules for both the structures according to specific and customized needs. Such customized needs may vary from a uniform temperature distribution (generally a more usual requirement satisfied by the buried approach) to a large stable temperature gradient over the suspended membrane (coplanar approach), thus for instance allowing a single heater serving a multi-sensing platform in which several integrated active nanostructures may sense the environment at different operating temperatures. This is known to represent a possible strategy to improve the selectivity for gas sensing applications. The mentioned structures are designed following two different strategies: (i) a coplanar approach where heaters and thermometers are integrated on the same layer as the sensing electrodes, thus resulting on the same geometrical plane; and (ii) a buried approach where they are embedded into an insulating layer, therefore the resistor is placed beneath the sensing electrodes. These two strategies correspond also to different sensing methods: the coplanar approach provides a thermal gradient on the membrane that can be exploited to let the sensor operate at different temperatures at the same instant, while employing a buried approach a very accurate and uniform heat distribution is obtained, thus avoiding any false positive from highly sensitive sensor. From the point of view of fabrication the coplanar approach provides great simplification in terms of process steps and consequent costs with respect to the buried one, since all the processes required to deposit insulation materials on the heater and to electrically separate it from the sensing electrode are not necessary.
In this paper Finite Element Method (FEM) models are set up in order to predict the electrical and thermal behavior of the micro-hotplates following respectively the coplanar and buried approach. The models are based on a multiphysics approach involving electrical, thermal and fluidics aspects. The multiphysics approach has become the leading method to deal with high complex systems that cannot be uniformed on a single physics, and this is particularly evident for MEMS [
16,
17,
18], where electrical, thermal and, sometimes, mechanical behaviors are responsible, at the same time, of the good functionality of the device The geometries, materials and physics are derived from the actual devices to obtain the best possible accuracy. In particular the Pt characteristics were investigated to obtain the experimental Temperature Coefficient of Resistance (TCR) to be included as an input in the models. The two types of devices were fabricated through standard MEMS processes and then characterized. Besides morphological, electrical and thermal characterizations, this work includes reliability tests in static and dynamic modes.
3. Results and Discussion
In micromechanics SiN is very often used as a structural material. LPCVD SiN is particularly relevant because it is characterized by a high fracture toughness and a good thermal shock resistance that allow the membrane to withstand high working temperature. Moreover, its low thermal conductivity involves a low heater power consumption.
The materials generally used as Resistance Temperature Detectors (RTDs) are platinum, copper, nickel and nickel iron alloys. However, Pt is the primary choice for most industrial, commercial, laboratory and other critical temperature measurements because of its resistance to oxidation, best accuracy and chemical and thermal stability also at high temperatures [
23,
24,
25]. Similarly intrinsic properties like high conductivity, chemical and thermal stability, and suitability for wire bonding make Au the most favourable material in order to realize sensing electrodes and contact pads in a single fabrication step.
According to the described process conditions, Ta was used as adhesion layer because it has proven to be better than other materials like Ti. In fact, the Ta layer is able to withstand annealing in an oxidizing atmosphere at 600 °C. This is in contrast to the optimal solution on the SiO
2 barrier layer, where the TiO
2/Ti is by far the best choice [
26]. The evidence of the deterioration of the Ti/Pt bilayer in oxidizing environments at high temperatures is fully explained. Annealing in an oxidizing atmosphere causes diffusion of O
2 through the Pt columnar grain boundaries, rapid oxidation of the underlying Ti layers and its migration into Pt. These phenomena generate significant problems in terms of loss of adhesion and degradation of Pt resistive properties [
27]. Such adverse reactions could not be prevented despite the deposition of Pt at high temperatures for achieving effective densification in the Pt layers. Improved stability is usually obtained by stabilizing the interface chemistry by introducing a well reacted oxide layer at the interface between Pt and the Si-based substrate. In this view, Ta plays an important role in the diffusion phenomena because during the heat treatment it forms a stable oxide increasing its thickness [
23,
28]. In particular, Ta oxidizes because oxygen diffuses through the Pt grain boundaries and, in contrast to the case of TiO
x, the underlying relaxation mechanism is a recrystallization process rather than a diffusion one [
23]. As Ta naturally oxidizes, the introduction of the oxide barrier is not performed through a deposition process, resulting in a simplification of the micro-hotplate fabrication process. Such recrystallization processes may affect the device functionality also through the formation of hillocks that increment the roughness of Pt surface [
23,
29,
30,
31,
32].
The morphological analysis performed with a Field Emission Scanning Electron Microscope confirms the expectations of the literature survey: the surface of the resistor is full of hillocks with different sizes. In the top view reported in
Figure 6a, hillocks appear as white islands that seem out of focus with respect to the Pt surface. Actually, hillocks are raised grains characterized by multiple dimensions and a different crystalline orientation with respect to the orientation of thin films [
32].
Figure 6b shows a cross-section view of the resistor in which the Pt columnar grain structure is remarkable as well as the presence of hillocks on the top.
Figure 6c is a tilted view that attests the distribution of hillocks on the whole Pt surface. Regarding the buried approach, heat treatment plays a key role in the evaluation of insulating layer thickness because of its correlation with the formation of hillocks which increase in number as well as in size with temperature, inducing a significant surface roughness [
32]. As a consequence, the SiO
2 thickness was tailored in order to properly embed the resistor and get the electrical isolation. The depth profile analysis performed by XPS is able to provide information about the atomic composition of the resistor. The atomic concentration of the resistor is reported as a function of the sputter time in
Figure 7. The point of interest is situated at the interface between the resistor and the SiN of the substrate, where percentages of Ta and O
2 are clearly detected by XPS. Such an evidence fully agrees with the aforementioned discussion about the oxidation of Ta adhesion layer due to the heat treatment. As a consequence the resistor, initially composed by a room temperature sputtered Ta/Pt bilayer, finally involves a thin layer of tantalum oxide. As TaO
x is an insulator, the current injected in the resistor flows exclusively through the Pt layer which accordingly requires a proper design in order to tailor the resistive properties.
The aim of the resistor is to heat the membrane and contextually measure the temperature reached on the membrane exploiting the Joule effect. Knowing the behavior of the electrical resistance as a function of the temperature is fundamental in order to properly use the resistor as heater and thermometer. In this view, the TCR describes the relative change of a resistance associated with a temperature change and is defined by Equation (7) in which α has the dimension of an inverse temperature (°C
−1 or K
−1). If the temperature coefficient itself does not vary too much with temperature, a linear approximation (Equations (7) and (8)) can be used to determine the value of the resistance
R at a temperature
T, using the Callendar–Van Dusen equation [
33], given its value
R0 at a reference temperature
T0 (usually room temperature):
Therefore, evaluating the TCR of the resistor is mandatory in order to univocally associate the resistance value to the real temperature reached on the resistor. It is known that the resistive properties of a bulk material are quite different with respect to the properties of a thin film. Thus, the TCR of TaO
x/Pt bilayer, that is far different from that of Pt bulk (3.92 × 10
−3 °C
−1) [
23], has to be experimentally evaluated. Measuring the variation of resistance as a function of temperature over the range 20–70 °C (59.75 Ω @ 20 °C and 65.15 Ω @ 70 °C) and using the Equations (8) and (9), the average TCR for the fabricated resistor is equal to 1.81 × 10
−3 °C
−1. Knowing the TCR allows one to tailor the supplying voltage and the related power consumption necessary to reach a temperature set point. Using this experimental value it is possible to evaluate the temperature distribution on the SiN membrane (
Figure 8 and
Figure 9) by the FEM models. The plots reported on
Figure 8 and
Figure 9 demonstrate that the predictive models are fundamental tools to tailor the design of a sensor, and in general of a microdevice, since they provide information impossible to be extracted from a simple characterization. The hypothesis beyond this assertion is that data from experimental analysis should match the model results, but, of course, an improvement of the same is always possible by changing parameters or introducing correction factors. An evaluation of the SiO
2 film on the buried design was also carried out, but there is no significant contribution from this layer to the device performance or temperature distribution since a difference of the order of +0.1 °C was found on the same probe point on the bottom with respect to the top of the SiO
2 surface.
The virtual models results demonstrate that the working temperature (400 °C) of the sensor [
11] is reached with 5.7 V applied to the thin film heater.
Figure 8a shows a linear temperature gradient formation on the membrane that can vary from 400 °C in the region next to the Pt resistor to about 150 °C close to the pads of the sensing electrodes. In order to quantify the linear gradient, the variation of temperature in the sensing region along the A-A section was evaluated: a linear coefficient between the temperature and the x coordinate of 0.2 °C/μm was found. This means that a linear variation from 25 °C to 400 °C that increases of 0.2 °C every micron, can be stably maintained along the A-A section (
Figure 8a). As expected, the temperature distribution of the buried design model is actually uniform especially in correspondence of the sensing region, where the set value of temperature can be spatially maintained with a maximum deviation of 7% with respect to the maximum temperature reached at the center (
Figure 8b). Moreover if an area of 200 × 200 μm
2 around the membrane center is considered, the temperature maximum deviation with respect to the maximum decreases to 2% (
Figure 8b).
The micro-hotplate die (
Figure 10a), having a square shape sized 6 × 6 mm
2, is compatible with the standard TO-8 package. The wire bonding is realized through gold ball bonding and the die remains floating and well anchored to the electrical connections. Such a set-up is used to carry out the infrared thermography analysis. The actual device characterization confirms the simulated behavior: (i) the device reaches 400 °C with 5.7 V applied to the thin film heater; (ii) the temperature distribution on SiN membrane is corroborates by the thermography analysis on the sensor as reported in the comparative plots in
Figure 11. Using this results it is possible to predict very accurately the working temperature on the SiN membrane and then use this data to calibrate the final device selectivity [
11].
Using the simulated behavior of the micro-hotplates it is possible to obtain the characteristic temperature curve of the resistors at different voltages (
Figure 11). The linear trend is a typical property of the Pt film, but this assertion is verified only if a proper thermal annealing at a sufficient higher temperature is performed. For this reason the devices have been thermally treated at 600 °C during the fabrication process as already described. In the coplanar design the U-Shape thermography replicates the design of the Pt resistor very well. This suggests that the temperature gradient can be geometrically designed as a function of the resistor geometry. Therefore, the predictive model can be easily employed to check the geometry variation influence on the temperature gradient. The meandering design of the buried resistor ensures the desired temperature uniformity in the sensing area, thus providing the ideal solution for every micro-hotplate based sensor in which the active material is very sensitive to the temperature variation and needs a high heat flux to work properly.
The response time of the models and the actual device have been investigated in order to compare the ramp up velocity (
Figure 12). The response time was defined as the time to reach the plateau temperature within an error of ±5% with respect to the plateau nominal temperature. In the case of the buried design, for the selected voltage of 5.7 V, it was expected a temperature plateau of 407 °C on the resistor. The simulations results indicate a delay of the order of tens of milliseconds between the driving voltage ramp up and temperature plateau and then a resulting response time of 70 ms. For the actual device a response time of 75 ms was obtained. These results confirm the model prediction and can be used to obtain a very fast microhotplate based sensor design, in particular, to evaluate the ideal response time of the sensor, to tune the driving voltage waveform, to define the plateau temperature at a specific voltage, etc. The same result in terms of time response was obtained on the coplanar design. A typical temperature drift due to the rapid rising up of the voltage is shown (
Figure 12) for the actual device.
The actual resistor is able to warm up to 400 °C with a power consumption of about 250 mW. Under these conditions, the properties of the resistor have been tested for a period of three weeks during which no appreciable changes have been observed (
Figure 13) already used This means that the Pt film has a very high stability in terms of conductivity and, even if directly exposed to a not clean environment without any package, it does not alter its behavior. Again it depends from the selection of the Ta as adhesive layer for the Pt film and the thermal treatment at 600 °C, 200 °C above the working temperature, which avoids any resistor degradation. This confirms that this kind of devices is suitable to develop efficient and robust sensors.