High-temperature MEMS Heater Platforms: Long-term Performance of Metal and Semiconductor Heater Materials

Micromachined thermal heater platforms offer low electrical power consumption and high modulation speed, i.e. properties which are advantageous for realizing non-dispersive infrared (NDIR) gas- and liquid monitoring systems. In this paper, we report on investigations on silicon-on-insulator (SOI) based infrared (IR) emitter devices heated by employing different kinds of metallic and semiconductor heater materials. Our results clearly reveal the superior high-temperature performance of semiconductor over metallic heater materials. Long-term stable emitter operation in the vicinity of 1300 K could be attained using heavily antimony-doped tin dioxide (SnO2:Sb) heater elements.


Introduction
Silicon micro-heaters have been a subject of intense research within the past decade. So far the interest in such devices was largely driven by the request for low-heating-power supports for gassensitive metal oxide materials. Moreover, the small thermal inertia of micromachined heater devices allows triggering rapid temperature changes which has opened up new perspectives for developing non-stationary methods of sensor operation [1][2][3][4].
As detailed in previous reviews of this subject [5][6][7] micro heater elements can be realised following two basic architectural designs: (i) closed dielectric membrane and (ii) hotplate devices. The architectural simplicity of the first kind of device is basically enabled by the low heat conductivity of SiO 2 and Si 3 N 4 , which form the structural membrane materials there. In the latter kind of devices semiconductor materials such as silicon (Si) or silicon carbide (SiC) are used as structural membrane materials. Due to the much higher thermal conductivities of these latter materials, closed-membrane architectures are no longer possible and hotplate designs need to be introduced to reduce the heat flow from the heated hotplate to the supporting silicon rim. Using this latter approach, relatively sophisticated devices have been realised in the meantime, co-integrating sensitive materials, bulk Si heaters and electronic read-out chips at the same time [8][9]. Recently, also gas-sensitive field-effect devices have been successfully integrated into such devices [10]. Meeting the relatively modest hightemperature requirements of such gas-sensing devices platinum (Pt) and polycrystalline silicon (poly-Si) have found wide-spread acceptance as electrically conductive heater materials [11][12].
Building on this work, attempts have been made to introduce novel materials into the micro heater design that allow attaining much higher temperatures in the range of 1300 K or above. Examples are SiC membranes and HfB 2 heaters [13]. Attaining temperatures in this range, novel applications of micro heater devices become feasible such as thermal IR emitters [14][15][16], flameless ionisation detectors [17] and surface ionisation devices [18]. In the present paper, we expand on this subject, reporting on the high-temperature-stability of silicon-on-insulator (SOI) based thermal IR emitters, which have been heated using different heater materials and heater designs. In particular, we compare the performance of metal (Pt, PtSi) and semiconductor heater materials (Si:B and SnO 2 :Sb) to point out that semiconductor heaters provide a significantly better performance at operation temperatures in the 1300 K range.

Emitter design
For reasons of attaining a low power consumption and short thermal response time, thermal IR emitter devices should be based on micro-machined membrane structures. For reasons of hightemperature compatibility the membrane should further consist of materials capable of operating at temperatures up to 1300 K for prolonged periods of time.
Both materials have a high thermal conductivity [23][24], which supports short thermal response times and thus a high modulation depth when operated in a discontinuous AC mode. High thermal conductivity, however, is not compatible with low electrical power consumption. For this latter reason, closed membrane designs cannot be used; hotplate structures rather need to be introduced to limit the cross sectional area of the heat conduction paths from the heated hotplate towards the supporting silicon substrate (Fig. 1).
1. 0 m m 1. 0 mm 1. 0 mm Figure 1. EM1 (left) mask layout of thermal IR emitter chips with on-chip temperature sensing elements for controlling the membrane temperature T mem and for assessing the long-term drift of the heater meanders [25]. EM2 (middle) is a design, which uses the thin, highly doped Si membrane in the middle as a bulk heater element with lateral Pt contact structures. EM3 (right) emitter design featuring an area heater based on SnO 2 :Sb contacted by lateral Pt metallisations and additional on-chip temperature sensing elements for controlling the membrane temperature T mem .
It needs to be kept in mind that not only the membrane itself has to withstand the high operation temperatures, but also the heater and temperature sensor metallisation on top of the membrane structure. Here, platinum and highly doped tin dioxide (SnO 2 :Sb) have been closely investigated. Platinum as a noble metal exhibits a high melting temperature and a low oxidation rate at elevated temperatures. At the same time, it possesses a low specific resistance and a positive temperature coefficient, which is attractive for temperature sensing and temperature control purposes. When directly deposited on Si the evaporated Pt layer is also able to form a platinum silicide (PtSi) layer by interdiffusion and reaction with the substrate, a property which is attractive for providing an improved film adherence. Tin dioxide (SnO 2 ), on the other hand, is a widely used metal oxide semiconductor, featuring a very high melting point and a negligibly small oxidation rate at elevated temperatures, as it is already in a fully oxidised state. Its specific resistance is intrinsically very high but can be reduced by adding substitutional dopants such as antimony (Sb). Its temperature coefficient of resistivity is in principle negative but may attain positive values in the case of heavy doping. In Table 1 the most relevant properties of the materials employed are listed.

Fabrication
In realizing the emitter designs shown in Fig. 1, silicon-on-insulator (SOI) wafers with an overall thickness of 356 µm and a top silicon layer thickness of 6 µm were used as substrate materials. In order to allow the top silicon layer to be used as an active heater material itself, this top layer was heavily boron doped (Si:B; C B ~ 10 20 cm -3 ; ρ ~ 5x10 -3 Ωcm). As previously reported [27], the main advantage of using SOI wafers is the simplicity of the device processing and the ease of obtaining monocrystalline silicon membranes with a reproducible thickness, which corresponds to the top silicon layer thickness of 6 µm [4]. The first processing step of the above devices consists in the growth of a 700 nm thick oxide on both surfaces of the SOI wafers. In this way a high-quality insulation is formed that separates the thinfilm metallisation on the front side from the top Si:B layer which will ultimately form the suspended membrane structure. In case the top Si:B layer is used as a bulk heater material itself, this surface oxide serves as a passivation against further rapid oxide growth. In a later process stage, the two surface oxides are additionally used as etch masks for the Si membrane formation (backside) and the hotplate definition (front side). In the second step, the heater and temperature sensor metallisations are carried out on the front surface and heater meanders, temperature sensing and reference resistors are structured. For this purpose, thin layers of metals or metal oxides have been deposited using standard vapour deposition techniques. The degenerately doped SnO 2 :Sb layers, for instance, were obtained by e-beam evaporation of mixtures of SnO 2 and Sb powders followed by a subsequent annealing step in air at 1320 K for 2 hours.
During this latter annealing step the deposited amorphous SnO 2 :Sb changes to a rutile structure which causes the incorporated Sb atoms to become incorporated as substitutional dopants [28][29]. The average thickness of the deposited layers of SnO 2 :Sb was about 900 nm. A maximum conductivity in fully annealed films of about 4500 µΩcm was obtained by employing Sb admixtures on the order of 5% by weight. The structuring of such layers was performed by lift-off prior to the SnO 2 :Sb annealing step using a pre-patterned mask of photo resist. This simple method of structuring SnO 2 :Sb layers is enabled by the fact that SnO 2 :Sb can be deposited at room temperature with the high-temperature dopant activation step following afterwards.
Pt layers were deposited by electron beam evaporation and structured using wet etching in aqua regia and photo resist masks. For improving the adhesive properties of the Pt films, additional annealing steps at temperatures of about 1000 K had been applied.
Following the completion of the metallisation steps, membranes were realized using an anisotropic wet etching process of silicon to open the membrane window on the backside of the SOI wafer. Subsequently, the buried silicon dioxide of the SOI wafers was removed in an additional wet etching process. Finally, the hotplate was realized by performing a dry etching process from the front side to create the suspended membrane structures displayed in Figs. 2 and 3. The geometrical parameters of the fabricated emitters are listed in Table 2.

Results and discussions
All investigations had been performed on the above-described emitter designs. Measurements were performed to find out the DC heating power consumption and the maximum modulation frequencies of the fabricated emitter designs. With regard to the employed heater materials, detailed measurements had been performed to the end of assessing their long-term performance at high operation temperatures. In these measurements novel heater materials such as platinum silicide (PtSi), degenerately borondoped silicon (Si:B) and heavily n-type doped SnO 2 :Sb were compared to the more commonly employed Pt heater metallisation.

Heating power consumption
Bearing IR emitter applications in mind, relatively large hotplates were designed with heated areas as large as 1500 x 1500 µm 2 . For this reason our hotplates do not conform to the low power consumption values of our previously designed hotplates where we aimed at gas sensing applications [4]. Instead of the 50 -100 mW power consumption at 700 K there, the power consumption of our present devices amounts to about 500 mW at this same temperature. Fig. 4 below shows that the power consumption increases roughly linearly from room temperature (T 0 ~ 290 K) up to about 800 K. Thereafter -up to the melting point of the Si support structure -the power consumption increases in a supra-linear fashion with (T -T 0 ) due to a dominating increase of IR radiation. As indicated in Fig. 4, a temperature calibration of the hotplate devices had been performed by a number of independent techniques. In the first set of measurements, the devices were passively heated on a thermo chuck and the built-in Pt thermometer was used to measure the membrane temperature (triangles); in the second kind of measurement, the devices were actively heated and the melting points of special temperatureindicator crayons (squares) were used to assess the membrane temperature; in the third set of measurement, IR emission spectroscopy (diamonds) was used to estimate the membrane temperature. Results of all tree measurement principles have been included to create the polynomial fit shown as solid line in Fig. 4. The plot exhibits that every measurement method reproduced comparable values of the membrane temperature, consistent with a thermal resistor modelling of the hotplate structures [25].   Fig. 3. Data points correspond to independent kinds of temperature measurements. The solid (red) and dashed (black) lines stand for a polynomial fit and for the simulation results obtained from a thermal resistor model, described in detail in [25].

Modulation frequency
The maximum useful operation frequency is determined by the value of the thermal time constant of the IR emitters. This time constant depends on the emitter geometry, on the one hand, and on the heat conductivity and the heat capacity of the hotplate structure, on the other hand.
As shown in Fig. 5, measurements of the cut-off frequency were performed by measuring the membrane temperature T mem as a function of the modulation frequency. Devices with two different values of the membrane thickness (6 µm, 13 µm) were evaluated. From these measurements a cut-off frequency of 12 Hz at 70% modulation depth was revealed.

Accelerated degradation tests
With all kinds of emitters and heater materials, accelerated degradation tests had been performed as illustrated in Fig. 6. Increased temperature (el. input power) leads to a rapidly increasing Pt degradation, whereas SnO 2 :Sb emitters can be operated up to temperatures in the 1300 K range demonstrating the superior performance of SnO 2 :Sb based heater elements. The seeming instability at lower temperatures is due to adsorption/desorption effects at the SnO 2 :Sb surface.
In these tests the heaters were operated at a constant heater power input, i.e. at a constant membrane temperature for a period of 30 min. Thereafter, the heating power input was increased in 50 mW steps until the heater finally failed. In the particular test shown in Fig. 6 two EM1 emitters, one metallised with Pt and the other with SnO 2 :Sb, were compared. The data in this figure clearly demonstrate that Pt heater elements rapidly fail at temperatures above 1100 K, whereas SnO 2 :Sb ones survive up to at least 1300 K.
In Fig. 7 the relative values of the resistance drift of the heater materials at each individual temperature T are plotted as a function of 1/T. From such Arrhenius plots an activation energy E a for the resistance degradation of Pt heaters of about 2.2 eV was obtained. In exactly the same way activation energies for the resistance degradation of PtSi (E a ~ 1 eV) and bulk Si:B heaters (E a ~ 4.3 eV) were obtained as well. Referring to the high-temperature data of the SnO 2 :Sb heater in Fig. 6 it is evident that these latter heaters in principle exhibit very large activation energies. Reliable values, however, could not be extracted from the two highest temperature steps in Fig. 6. The seeming degradation drift of SnO 2 :Sb heaters in the lower temperature range is an artefact produced by the residual gas sensitivity of SnO 2 :Sb layers. Most likely, these latter resistance changes are due to changes in the Oor OHadsorbate density and therefore, do not yield any information concerning the bulk degradation rate of the SnO 2 :Sb resistors. Keeping in mind that the SnO 2 :Sb heaters exhibit an even larger activation energy than Si:B ones, the data of Fig. 7 clearly reveal the superior performance of semiconductor heaters with respect to metallic ones.  Activation energy for heater degradation for different heater materials. Si:B heaters are seen to perform considerably better at high temperature than metallic ones. SnO 2 :Sb heaters exhibited even larger activation energies. Because of a residual gas sensing effect, reliable values of E a could not be obtained in this latter case.
With the activation energies E a obtained, the "low-temperature" lifetime of the different heater materials could be obtained by extrapolation. These latter results are presented in Fig. 8. In this comparison, SnO 2 :Sb heaters were not been taken into account, due to the mentioned uncertainties in determining an activation energy.  According to these data, attaining thermal lifetimes of the order of 10 years needs to limit the maximum operation temperature of PtSi emitters to 550 K, that of Pt emitters to 850 K and that of Si:B ones to 1050 K.

Direct measurement of the long-term stability of heater metallisations
Considering the difficulties of extrapolating the SnO 2 :Sb heater performance towards commonly employed operation temperatures, direct measurements of the long-term stability of all the above heater materials were initiated. In these latter tests, the emitter devices were heated at about 1100 K or 1300 K, respectively, for a prolonged period of time to directly assess their long-term stability under typical operation conditions. In this way, a straight-forward assessment of the long-term stability of SnO 2 :Sb heaters was made and the lifetime estimates for Pt, PtSi and Si:B heaters, as obtained from the accelerated degradation tests, could be validated.
Such direct tests again reveal the superior performance of semiconductor with regard to metallic heaters. In the particular experiment shown in Fig. 9 a SnO 2 :Sb heater and two Pt heaters were operated at about 1100 K with constant heating power or alternatively with constant supply voltage. Fig. 9 shows that after one week of testing at 1100 K the SnO 2 :Sb emitter exhibits the same resistivity as in the beginning of the measurement. As the emitter devices were operated in open air, changing air flows caused some noise to become superimposed on the otherwise straight SnO 2 :Sb resistivity line. In comparison, the Pt heaters exhibited right from the beginning a large and continuous drift of the heater resistance and finally, failed within 36 hours. Fig. 10 displays the results of another experiment in which an IR emitter with SnO 2 :Sb heater metallisation was mounted inside a sealed tube filled with ambient air. This kind of housing ensured that the emitter was protected against disturbing influences of changing humidity, air pressure and air flow. In this latter experiment, the emitter was permanently operated at a temperature of about 1100 K for four weeks. Thereafter, its operation temperature was increased to about 1200 K for another three weeks after which the experiment was interrupted. Fig. 10 vividly shows that no significant degradation of the SnO 2 :Sb heater metallisation can be detected during the whole operating period of about seven weeks.  In order to obtain more clues on the nature of the degradation phenomena, optical microscopy was performed on degraded heater meanders. Some of these results are shown in Fig. 11. The left-hand panel in this picture shows the state of high-temperature degradation of a Pt heater resistor after 31 hours of operation at a constant temperature of 1100 K. It can be seen that at the hottest points the Pt heater became increasingly porous until it finally failed. This local degeneration can be traced back to electro-migration and inter-diffusion effects both of which are accelerated by high temperature. In contrast, the SnO 2 :Sb based heater meander shown in the right panel does not exhibit any visible signs of degradation after operation at 1300 K for a period of 18 hours. The resistance drift visible in Fig. 10 therefore cannot be attributed to the same kind of degradation as in the Pt case. Very likely the resistance drift in this later case is due to a slow equilibration of the oxygen vacancy density in the semiconductor material [30].

SnO 2 :Sb based area heater elements
The above results have clearly shown the superior performance of semiconductor over metallic heater elements and in particular the excellent performance of heater elements based on oxide semiconductors such as SnO 2 :Sb. A drawback of this latter material is its relatively high resistivity of about 4500 µohmcm (see Table 1) which is about three orders of magnitude higher than the resistivitity of Pt. As a consequence, operation voltages of about 100 V were necessary to achieve operation temperatures in the vicinity of about 1300 K with the EM1 emitter design, which we commonly employed in the case of Pt heater elements.
From an application point of view such high operation voltages are undesirable. In order to attain operation voltages of less than 24 V, the EM3 emitter design was introduced (Fig. 1). In this latter case SnO 2 :Sb strips were deposited across the hotplates and low-resistivity Pt contacts on the much colder Si suspensions to provide a low-resistance access to the SnO 2 :Sb strips (Fig. 3d). With this design the heater resistance could be reduced to about 130 Ω, which compares very favourably with the 8 kΩ heater resistance of the EM1 design in Fig. 3a. At present, investigations are underway to validate the performance of the new emitter design [31].

Conclusions
High-temperature IR emitter operation could be obtained using SOI-based hotplates as thermally insulated base structures. With all heater materials investigated, high-temperature operation in the range of 1000 K and above could be obtained. Semiconductor heater materials proved to be clearly superior in performance with regard to metallic ones. Whereas metallic structures failed after several days at high-temperature (T~1000 K), excellent high-temperature stability was obtained with semiconductor heaters such as Si:B and SnO 2 :Sb. These superior properties are likely to be related to lower current densities in the heater material and lower self-diffusion constants in semiconductor as compared to metallic heater materials. The best results so far have been obtained with the SnO 2 :Sb heater material, which -as an oxide -cannot suffer from any ongoing oxidation during hightemperature operation. The obtained results show clearly that SnO 2 :Sb is a promising material for high temperature heater elements in microsystems. Future applications for such high-temperature MEMS heater platforms are, e.g. flameless ionisation detectors [17] and surface ionisation devices [18]. Both applications rely strongly on a stable, and very constant high temperature heat source with an excellent long-term stability.