MEMS for real-time infrared imaging

ABSTRACT

This project investigates an innovative approach to imaging with Micro Electro-Mechanical Systems (MEMS) based devices. By using a Linnik interferometer and advanced phase unwrapping algorithms for processing data, the feasibility of generating high-resolution grayscale images in real-time was proven with an array of individually addressable MEMS micro-mirrors. Further investigations on a thermal imaging detector consisting of an array of pixels defined by surface micromachined bi-material beam structures were carried out. A thermal loading fixture was manufactured and incorporated into the interferometer setup, which was also optimized to provide high measuring resolution. Interferometric images were collected at several temperatures in order to determine the beams’ response as a function of temperature, which successfully demonstrated the suitability of the detector to imaging with high-sensitivity and with a linear response. Experimental results were used with analytical and computational models to further predict the thermo-mechanical characteristics of the beams and to perform parametric investigations and optimization of their design. Further developments will consist of integrating the detector into a highly advanced, completely mechanical, imaging device having mK thermal resolution. The availability of such device will greatly improve current thermal imaging technology.

Keywords: interferometry, infrared imaging, MEMS, metrology, thermoelasticity, micromachining


Introduction

This project investigates an innovative approach to thermal imaging using MEMS devices with laser interferometry, a technique that allows for the generation of continuous grayscale variations that are proportional to temperature differences. The current state of the art thermal imaging devices use CMOS and CCD sensors – digital devices which are only able to identify quantized levels of infrared radiation. Since this radiation is emitted from any warm body, these devices are subject to high levels of noise; even from the devices themselves. As a result the device requires expensive cooling systems in order to maintain high thermal resolutions, thus limiting them to stationary use. These cameras have digital resolution on the order of 25-30mK, which is insufficient for some of today’s thermal imaging applications such as development of advanced circuitry and long range infrared detection devices for homeland security.

Higher resolutions can be obtained using MEMS devices that react to infrared radiation. These devices use micro scale, bimaterial cantilever beams, which react to changes in temperature by deflecting out of plane. Naturally, the scale of these variations is on the order of nanometers, so highly sensitive measurement techniques with high resolution must be utilized.

We have developed a Linnik interferometer for measuring changes in position with sub-nanometer resolution. Changes in optical path-length of the object beam shown in Fig. 1, create constructive and destructive proportional to the position of a sample.

The principal of operation of the setup is that thermally induced deflections of microcomponents can be quantified interferometrically as grayscale patterns that encode temperature information.

Linnik interferometer configuration used in our developments [1].

Fig. 1. Linnik interferometer configuration used in our developments [1].

Interferometric measurements

Principle of operation

A four phase stepped algorithm with a quarter wavelength step is used to extract optical phase information corresponding to the shape of the micro-components of interest. Therefore, the phase distribution across an object is determined with,

tmp16221_thumb

3 A with I1 to I4 corresponding to images from 0 to — step positions of the phase shifter and 1 is the wavelength of the laser diode 4 used for illumination. By using Eq. 1, it is possible to determine deformations, Lz, with

tmp16222_thumb

While this algorithm is useful for absolute measurements of shape and deflection of the surface of an object, the algorithm can be used in double-exposure mode to measure changes in displacement rather than absolute positions [ 1]. This allows for measurements of very small displacements relative to the initial shape of the surface of an object.

Calibration and accuracy of the system

A crucial part on the quality of the measurements is the positioning accuracy of the phase shifting reference mirror, which is driven by a piezo-based nanopositioner. Calibration of the nanopositioner is achieved by implementing a high-resolution Hariharan algorithm [2] that leads to positioning resolutions on the order oftmp16223_thumb

After calibration, the interferometric system was tested for accuracy and resolution with a standard calibration flat plate having a flatness in the A range. The demonstrated measured resolution achieved is on the order of 0.25nm. The setup was also tested for accuracy and mechanical stability and they were found suitable for our measurements.

Proof of concept system for real-time interferometric imaging

Initial experimentation consisted of a proof of concept to determine the feasibility of using an interferometer for real time grayscale pattern generation. To test this, a sample array of pixels was required that could predictably be moved in and out of plane. A commercially available micro-mirror array manufactured by Boston Micromachines Corporation [3] was acquired for testing. The chip consists of an array of 18*18 individually addressed 600*600^m2 micro mirrors; each having a maximum out-of-plane displacement of approximately 400nm. The actuation of the mirrors is completely analog and can therefore be used to demonstrate the interferometer’s ability to measure continuous displacements. Figure 2 shows an example of a pattern generated with our interferometer measuring unwrapped optical phase in double-exposure mode while 6 micro-mirrors are operated independently

(a) MEMS generation of the letter "J", viewed using interferometry; Top view (b) of the micro-mirror array.

Fig. 2. (a) MEMS generation of the letter "J", viewed using interferometry; Top view (b) of the micro-mirror array.

Realization of thermally induced pattern generation using MEMS bi-material beam arrays

Analytical and FEM characterization of a bi-material beam

Full characterization of the beams requires a combined analytical computational and experimental approach. However due to the complex shape of the bi-material beams used in the MEMS device [4, 5] no closed form analytical solution is available. To overcome that an indirect method was implemented by initially calibrating a FEM model of a simply supported bi-material beam against an analytical model [6, 7]. The computational model of the simple cantilever beam yielded a sensitivity of 24.1 nm/K while the analytical model predicts a sensitivity of 23.8 nm/K – less than one percent error between the solutions for 64^m long beam. Both models predict completely linear thermal sensitivity. The FEM model was then further modified to match the true geometry of the bi-material pixels in the MEMS array [8].

FEM model characterization of MEMS bi-material beam

Our computational model of a simple cantilever beam validated that the FEM model could be reliably expanded to predict the behavior of beams with more complex shapes, such as the ones in the MEMS bi-material beam array that was implemented in our system [9, 10, 11]. Shown in Fig. 3, note the presence of release holes, a trademark of the surface micromachining process. Figure 3b shows the release holes (and corner fillets) modeled, although the FEM model does not take these into consideration. These features affected the computation by only 1-3 pico-meters, thus these features were deemed negligible.

50x magnified images: (a) of a section of bi-material cantilever pixel array; Graphic representation (b) of bi-material cantilever pixel from FEM model.

Fig. 3. 50x magnified images: (a) of a section of bi-material cantilever pixel array; Graphic representation (b) of bi-material cantilever pixel from FEM model.

Experimental characterization of MEMS bi-material beams array

Experimental data was collected using the Linnik interferometer; the data were processed and analyzed using a collection of softwares for data capturing [12] and phase unwrapping [13]. Raw data was first masked using a MATLAB program to avoid errors when using the unwrapping software. The raw masked data was then phase-unwrapped using a fluid unwrapping algorithm. Finally, unwrapped data was exported to MATLAB for analysis; with this data, both absolute and relative position could be determined. By relating these displacements to the temperature of the beams at the time of the experiment and taking several readings at different temperatures, the sensitivity of the beams was calculated. Then, by combining the thermal sensitivity of the array with the resolution of the interferometer, the thermal resolution of the system was determined.

In order to conduct an experiment with such a thermally sensitive MEMS device a special thermal loading station was designed and built [14] to accurately determine the temperature of the experimental array of beams. Figure 4 shows a CAD model of this setup while of Fig. 5 shows its realization and use. Two thermoelectric cooling devices (TECs) were used to generate a temperature difference between the base plate and a heat sink fastened below it. There were two major considerations in the design of this setup: first, that the experimental array is to be kept stationary during the heating and cooling process, and second, that the two thermal bodies – the MEMS array and the support plate, be thermally isolated to ensure thermal stability.

CAD model of thermal loading station

Fig. 4. CAD model of thermal loading station

Design realization of thermal loading station

Fig. 5. Design realization of thermal loading station

Data sets were collected by first capturing a reference image at an initial temperature and then taking ten images at even intervals as the array was either heated or cooled. Data sets were taken at a variety of starting temperatures, both heating and cooling, and with the ten images captured in temperature ranges from 1°C to 5°C. Two high sensitivity thermistors inserted in the support plate were used to provide feedback for temperature of the beams.

After obtaining the raw experimental data it was noticed that directly applying unwrapping algorithm to it resulted in errors mainly because of data noise at the edges of the beams. The noise was due to irregular illumination of the edge areas of the beams which caused errors in readings of the surface at these points. This was accounted for by creating a MATLAB program than automatically masks the raw data so that only the surface of the beams is visible along with some of the surface of the substrate. The substrate areas were used as a reference for the unwrapping process so that the true deflection of the beams can be calculated. The mask greatly reduced the noise in the data and thus allowing for the unwrapping algorithm to produce much smoother surface representing the surface of the beam. Figure 6 shows the final result of the unwrapping algorithm of two pixels, while Fig. 7 shows the 3D representation of that data of one pixel.

Masked unwrapped phase data of two beams, which correspond to two pixels of the array tested

Fig. 6. Masked unwrapped phase data of two beams, which correspond to two pixels of the array tested

3D representation of the shape of a beam at a detected temperature of 17.49oC

Fig. 7. 3D representation of the shape of a beam at a detected temperature of 17.49oC

The unwrapping algorithm produces a continuous grayscale image so that the grayscale intensity of each pixel corresponds directly to the displacement of the beam at that point. This allows for full-field of view measurements of the thermally induced displacement of the bi-material pixels as well as their shape.

Based on this information, the curvature of the beam was extracted by tracing a line on the 3D surface of the beams. This technique was applied for the extraction of the curvature of the beam at several temperatures. By reading the position of the surface near the tip of the beam, a data for the thermal sensitivity of the beams can be extracted. While beams are 136.5^m long, information at 122 ^m from the base of the beams was sampled to minimize noise associated with inaccuracies of the readings at the tip of the beam. A graph of the experimentally measured beams’ displacement with respect to temperature along with FEM predictions are shown in Fig. 8.

Beam sensitivity comparison for FEM and experimental results

Fig. 8. Beam sensitivity comparison for FEM and experimental results

According to Fig. 8 it can be clearly observed that computationally predicted thermal response of the beams is within the error margin of the experimentally predicted response. The FEM model yields a sensitivity of 87.4nm/K and the experimental characterization shows a sensitivity of 85.5nm/K – a difference of only 2.2%.

This gives confidence that the experimental procedures were done with high degree of accuracy. Having the sensitivity of the MEMS bi-material beams array and the sensitivity of the interferometric measurement system, the resulting overall thermal resolution of the system is determined to be 3mK. By comparing this value to the resolution of existing high-end infrared imaging systems – 25mK [15], the proposed system offers more than 8 times increase in thermal resolution.

Conclusions and future work

In our project we successfully demonstrated that interferometry can be used for real time gray-scale imaging proportional to temperature. We also managed to characterize MEMS thermally actuated bi-material beam array. With our analytical and FEM model of the thermal sensor we allow for future design optimization of the beams array for specific applications. Overall, our system demonstrated a thermal resolution not achievable by any currently available conventional commercial technologies. Future work should be focused on miniaturization and full system integration of the system. Further investigations on achieving full system mobility should be conducted.

Next post:

Previous post: