Self-Cleaning Micro-Windows for In-Tailpipe Optical Exhaust Gas Measurements

Exhaust gas measurement in the harsh environment of the tailpipe of a combustion engine by optical techniques is a highly robust technique, provided that optical access is maintained in the presence of particulate matter (PM). The considerations are presented for the systematic design of membranes with integrated heaters in SiC-on-Si technology for generating a well-defined lateral temperature profile with peak temperatures above 600 °C. Periodically raising the temperature of the membranes to such a level is demonstrated to keep the surface transparent by oxidation of soot deposits. This paper is about continuous heating of the membrane to a temperature slightly higher than that of the exhaust gas. At such temperatures thermophoretic repulsion of PM allows allows long-term optical measurement in the exhaust without the thermo-mechanical loading by repetitive thermal cycling.


I. INTRODUCTION
Sensor systems for the measurement of gas composition and particulate matter (PM) in the exhaust emission of a combustion engine are essential for ensuring compliance with strict emissions requirements that are set by regulatory agencies and imposed on the automotive industry [1]. The exhaust system is a notoriously harsh environment, due to the presence of corrosive gases, oil residues, and soot particulate matter with temperatures up to 400 o C, and mechanical vibrations. These conditions come on top if specifications of the primary functionality of the sensor in terms of selectivity and detection Partly funded by: Strategic vehicle research and innovation programme (FFI), Vinnova. limit is demanding, which makes the design of such a system challenging. The commercial state-of-the-art exhaust sensors are often based on electrochemical principles for gas composition measurements, e.g. oxygen sensor (lambda sensor) [2] and electrical conductivity for particulate matter (PM) [3]. Compared to these, optical techniques provide a non-contact and robust measurement approach, as long as the optical access of light (electromagnetic radiation) into and out of the harsh environment of a tailpipe is ensured. Results in the literature are restricted to short-term measurements [4], which is due to the reduced optical transmission over time caused by soot deposition on the sensor.
Electrically conductive PM sensors are periodically regenerated by heating to a temperature in excess of 600 o C to burn away the deposited soot [5]. Similarly, heating of a transparent optical window is expected to effectively burn away any deposited soot and provide sustained optical access to the sampling environment. The thermophoretic force that results from the temperature gradient from wall to the inner exhaust pipe during regular engine operation is the primary deposition mechanism of PM on the cold wall [6]. Locally reversing the temperature gradient, i.e. increasing the wall temperature to above that of ambient, has been shown to create a soot-free volume above the surface and suppress the particle deposition [7]- [9].
Our initial work on this topic using SiC membranes fabricated in MEMS technology has resulted in devices in which the self-cleaning of hot spots on the window of soot deposits by pulsed heating was demonstrated [1]. Moreover, 978-1-7281-5635-4/20/$31.00 ©2020 IEEE the effectiveness of thermophoretic repulsion for keeping the membrane clean at PM exposure was successfully validated. However, maintaining a uniform temperature over a large-area membrane at a level required for thermal regeneration proved highly challenging due to limited understanding of properties of the used poly-crystalline SiC at high temperatures. Furthermore, the long-term impact of high-temperature cycling on the reliability of the window is still unexplored. The present work aims to study methods to obtain high temperatures and uniform distribution of temperature over the membrane. Furthermore, the use of thermophoresis for active repulsion of the soot is elaborated.

II. MEMS DESIGN
The structure of a self-cleaning MEMS window is essentially composed of a thermally and chemically inert membrane with integrated heaters that can be used to locally increase the temperature of the membrane. The design of the throughwafer openings defines the size of the MEMS windows. Fig. 1 schematically shows the design of the membrane and its heater. The membrane material must be transparent over the entire measurement spectrum enabling direct optical access to the harsh environment of the exhaust. Silicon-carbide (SiC) has high mechanical strength and rigidity, transparent over visible range, and chemically inert which makes it suitable for the membrane material. Furthermore, in situ electrically doped SiC layers can be deposited by low pressure chemical vapor deposition (LPCVD) and used for the resistive heater design that can reach temperatures of above 600 o C. In the actual implementation, the membrane structure is exposed to an exhaust flow at temperatures from 400 o C down to less than 100 o C depending on the placement of the sensor in the exhaust tailpipe. The window must be able to sustain pulsed heating up to temperatures above 700 o C for periodically burning any organic deposits. Such high temperatures are required for a short regeneration time, as the oxidation threshold of soot is at about 500 o C and the oxidation rate increases rapidly with temperature [10]. The layout design of the heaters is critical for the achievable temperature uniformity and peak temperature of the transparent window. The temperature distribution of the suspended membrane is the steady-state result of the competition between the generated resistive heat and the heat loss mechanisms involved during the operation. Although all three main heat loss mechanisms of heat (i.e. conduction through the substrate, convection through the fast-moving gas, and radiation from the heated membrane) are present in this device, numerical calculations shows that at the operating conditions involved here, the heat loss is dominated by the conduction to the substrate (about 3 orders of magnitude higher than the other loss mechanisms). Finite element analysis (FEA) was used to model the temperature distribution of various heater layouts at realistic conditions. Fig. 2 shows the temperature distribution over a 1000 µm diameter membrane for operation with soot containing gas at room temperature. The heater is a circularly shaped loop with a diameter of 700 µm and has a thickness of 700 nm. The resistivity of the doped SiC was measured to be 2.41× 10 3 Ω·cm and was used for finite element modeling. An engine operating at stationary conditions would result in exhaust gas flow in the exhaust system with the gas cooling downstream the system. For a typical sensor position the temperature could, depending on the specifics of the engine and exhaust system, be 400 o C. This is basically the ambient temperature of the membrane, which is heated to even higher temperature using the integrated heater. Fig. 3 shows the thermal distribution over the surface. An important observation is the reduced heating power required due to the elevated ambient temperature. As compared to heating from room temperature to 700 o C, about half of the input power is required. Obviously the distance between the heat source (heater) to the edge of the membrane decreases with diameter of the heater. Consequently, the conduction loss to the bulk wafer decreases with a smaller heater diameter, which results in a higher membrane temperature at a given power level. As shown in Fig. 4 for a constant power and membrane diameter, the temperature of the membrane increases almost linearly with decreasing heater diameter. However, this leads to an important design constraint. For maximizing the distance to the membrane circumference, the heater should be placed right at the centre of a square or circular membrane (hence a rectangular design with unequal side-lengths or an elliptic design result in a reduced performance). Furthermore, as a consequence it is by necessity within the light beam transmitted through the membrane. Therefore, the heaters should be fabricated in a transparent material to avoid partially blocking of light. Doped SiC layers are used here, which are transparent, but the optical constants (index of refraction and extinction coefficient) are slightly different from the undoped SiC of which the membrane is composed. As a consequence, the optical transmission is not constant over the membrane area. This effect can be minimized using a specially designed deposition process and membrane design.
The thermal conduction loss to the substrate is also reduced when designing for a larger window diameter. However, for any given yield stress, the membrane becomes more vulnerable to breakage due the stresses introduced at fabrication or at high-temperature thermal cycling. A 1000 µm diameter membrane proved sufficiently reliable, while also of sufficient diameter for passing light beam of practical dimensions of the aperture.
The heater width gives yet another parameter for the design of the optical windows. For a given sheet resistance of SiC, the width of the heater determines its resistance. At a constant temperature operation, thus for operating at a given driving power, the width is tailored for a reasonable current or voltage specification. For a thin-film heater with a rectangular crosssection, the input current increases with the square root of the width of the wire in a current source configuration (see Fig. 5). A wider heater increases the heat transfer from the membrane to the substrate, and as shown in Fig. 5, decreases the temperature of the membrane. An approach for relaxing the structural constrains is to make use of multi-loop heater designs, e.g. having several concentric heaters connected in series. The heaters thermally operate in cascade, with each heater generating a relatively small temperature step. Therefore, from the edge of the membrane toward the center the total temperature step is realized by the combined effect, which can be used to effectively decrease the overall conductive heat loss from the inner center of the membrane by design. Therefore, using a multi-loop design, a higher local temperature at the center of the heater is achieved at any given input power. Fig. 6 shows the modeling result of temperature distribution over a membrane with different heater designs, operating at room temperature. Accordingly, at a constant power the peak temperature increases linearly with the number of loops.

III. THERMOPHORETIC REPULSION OF AIRBORNE SOOT PARTICLES
Periodic oxidation of deposited soot at temperatures over 600 o C is a proven reliable mechanism for periodical regeneration of the conductometric soot sensor [1], [5]. However, cycling would have to be much more frequent in an optical window. Another concern is very thin films of remaining deposits, which may not be electrically relevant, however basically result in a multi-film optical filter that may have a significant impact on the optical transmission through the, supposedly cleaned, window.
The resulting frequent repetitive thermal cycling of the windows is expected to significantly impact reliability. Thermophoretic repulsion is investigated here to alleviate this constraint. The thermophoretic effect is the force acting on airborne particles moving in a temperature gradient field. The fact that the wall of the exhaust pipe is the part at the lowest local temperature is a primary cause for soot deposition on the wall of PM from the exhaust gas while passing through the interior of a cold tailpipe. The same effect can be used to prevent soot from deposition at the window surface when the temperature of the window is raised to a level slightly above that of the soot-containing exhaust gas. Such a reversed temperature gradient repels soot particles toward the center of the flow channel and reduces the number of the deposited soot particles.
As shown in Fig. 7, maintaining a slightly higher temperature over the window than the gas does effectively keep the surface free from soot deposits. As a consequence, the period between soot oxidation cleanings is strongly enlarged.

IV. FABRICATION AND MEASUREMENT RESULTS
The fabrication process is a combined surface and bulk micromachining process, where the surface micromachining defines the heater layout. The size of the transparent windows is also set by patterning the wafer backside and through-wafer DRIE etching. A SiO 2 thin film was used as the etch stop layer. MEMS Windows with diameters ranging from 100 µm up to 1000 µm with different heater designs were fabricated. The heaters were patterned in a doped SiC layer, while an undoped SiC layer acts as the membrane material. The details of the fabrication process are presented in [1]. The 10 × 10 mm 2 chips were then diced and packaged on a PCB attached to a 3D printed flow chamber, as shown in Fig. 8. A propane burning soot generator (Jing MiniCAST 5201C) connected to a diluter (Dekati FPS4000) generated a reproducible soot flow that was fed to the test packaging. The flow rate was controlled using a vacuum pump and a mass flow controller. A transparent glass was glued to the 3D packaging providing visual access to the surface of the membrane during the experiments. Fig. 9. Active repulsion of soot particles using the thermophoretic effect. The soot deposition on an unheated transparent membrane as after constant exposure to a flow of soot over a period of 94900 sec (off state). A similar, but heated window on the same chip remains clear from any soot deposition during the same experiment (on state).
The effect of the thermophoretic force on moving particles was examined by long exposures to soot containing flow. In this experiment two similar windows on the chip were selected where one is heated up to several degrees above gas temperature, while the other window remained unheated. In one experiment, the package was kept under a constant flow of soot over 26 hours, while one membrane was kept at 70 o C, mere 35 o C above the gas temperature. The surface of the chip after the experiment is shown in Fig. 9 and showing no visible soot deposition over the heated window. The unheated window, on the other hand, was subjected to a considerable soot deposition, which significantly reduced its transparency. Measuring the brightness in an area of about 600×600 µm 2 in the center of the window using image processing, shows that the overall brightness decreases by 0.20±0.05 on the unheated membrane, while the brightness of the heated window did not significantly change within a margin of error of 0.05.

V. CONCLUSIONS
The work presented focuses on the design of a window in SiC that can be operated in the harsh environment of the tailpipe and equipped with a mechanism for maintaining window transparency, despite significant exposure to PM. Thermal effects are explored for use in maintaining the transparency of a MEMS fabricated window by either regeneration of the deposits or active repulsion of particulates.
The regeneration of the window transparency requires a high temperature generation over the entire membrane, which is power consuming. Optimized multi-loop designs are presented. Moreover, the concept of thermophoretic repulsion is elaborated and a mode of operation is designed for enabling long-term operation.
In our earlier experiments the leakage current in the undoped SiC at high temperatures resulted in spot heating, instead of a well-controlled temperature profile over the membrane area, which results in self-cleaning at hot-spots only. Currently, the designs presented here are being fabricated using an improved fabrication process for avoiding hot spots. The thermophoretic repulsion was convincingly demonstrated and will be characterized in more detail in future experiments. Moreover, dynamometer tests in actual exhaust systems are scheduled.