Surface micromachining techniques are being used to create a miniature, low cost replacement for conventional optical spectrometers. Spectrometers are used in scientific instruments for many important measurements including chemical analysis by optical absorption and emission line characterization. A miniaturized instrument or microspectrometer will offer significant advantages over existing instruments including size reduction, low cost, high reliability and fast data acquisition. Because of these advantages, a broad application of optical measurement techniques may be anticipated in the future.
Previous authors have developed micro-spectrometers using bulk micromachining techniques[1]. These devices suffer from a number of limitations, including size, planarity control, and undesired absorption in the substrate. To overcome these problems, we are fabricating a surface micromachined device. There are two major advantages of surface micromachining in this application. The first is that there is no fundamental limit to how small the device can be. Devices occupying 30µm x 30µm with active areas as small as 5µm x 5µm are feasible. One could easily conceive of multiple spectrometers forming an array or adjacent spectrometers designed for different spectral regions. In fact, small spectrometers are easier to build than larger ones. The second is that our design separates the dielectric mirror and the flexure. This permits the variation of the spacing between mirrors while avoiding deformation of the moving mirror thus providing better resolution.
Figure 1 shows a simplified cross-section of the proposed visible microspectrometer and Figure 2 shows the top view. Two elements are required to make a spectrometer; a light detector and a wavelength selective element. A photodiode is fabricated in a silicon substrate and a wavelength selective element is micromachined above the diode. The choice of silicon as the substrate material allows the incorporation of a sense amplifier and drive electronics on the same chip.
The wavelength selective element is a Fabry-Perot interferometer. The center layer of the interferometer is an air gap created by fabricating a micromechanical bridge above the silicon photodiode. The two mirrors that are components of the interferometer are deposited both in a hole on the bridge and directly on the surface of the photodiode.
In the schematic diagram shown in Figure 1, an n-type emitter layer is diffused into a p- type silicon substrate to create a photodiode. The n+ layer itself becomes part of the lower interference mirror, which includes a quarter wave SiO2 layer and a quarter wave silicon layer. An air gap width of half the center wavelength is created. Above this a second interference mirror consisting of quarter wave silicon and silicon oxide layers is formed. The choice of silicon and silicon oxide is for convenience and not critical. Other material pairs can be used, where one film has a high index of refraction, such as silicon, and the other a low index material, such as silicon dioxide.
The number of layers in the mirrors determine their maximum reflectance. The greater the number of layers, the narrower the band width of the interference filter. In our work, we plan to use a seven layer interference filter.
The optical constants for silicon and silicon dioxide were entered into a computer model and the transmission of the filter was calculated as a function of the thickness of the air gap. The transmission represents the amount of light that enters the photodiode to be collected and converted to an electrical signal. A center wavelength of 5000 Å was chosen. Figure 3 shows the results of these calculations. Note first the curve representing the transmission when the gap is set to 2500 Å. In this case, the curve peaks at exactly 5000 Å or twice the gap spacing as expected. It should be noted that there are second order responses at large wavelengths and zeroth order peaks at short wavelengths. Four other curves show the results if the gap is set to 2000 Å, 3000 Å, 3500 Å and 4000 Å.
Based on these calculated results, the full width half maximum of the transmitted output is approximately 1/12th of the spacing between the peaks. This suggests a resolution limit for this spectrometer of approximately 160 Å. This result is not as good as that available from conventional spectrometers which would typically have a resolution exceeding 20 Å.
The resolving power of a Fabry-Perot spectrometer can be expressed as
where R is the reflectivity of the mirrors and N = 2nd/l with n the index of refraction and d the spacing between the mirrors. The software used in this analysis predicts a reflectivity for a seven layer mirror centered at 0.5 mm will have a reflectivity of approximately 99%. Use of the formula above would result in an estimate of the resolving power, RP = 310. By definition, RP = l/dl and the predicted resolution at 0.5 mm is 160 Å. By increasing d the resolving power can be increased significantly at the expense of free spectral range. Conventional Fabry-Perot spectrometers are capable of RP = 100,000.
The work conducted so far has been aimed at the development of a working microspectrometer. The most important aspect of this work is the establishment of a baseline process for fabricating the device. Developing a process involves, among other things, the verification that the materials and etchants chosen are compatible throughout the process.
In Figure 4, the process for the microspectrometer is outlined. In this process, the first mirror is fabricated on a glass substrate using vacuum evaporation. The silicon and the silicon dioxide layers are adjusted to be quarter-wave at the wavelength of interest. An optical thickness monitoring system similar to those used by the optical coatings industry insures an accurate deposition of the individual layers of the dielectric mirror. A bridge structure is plated above the sacrificial layer. A second mirror is plated above the bridge and finally, the sacrificial layer is removed to release the device.
Figure 2 shows a top view of the completed device. At the center of the bridge structure is a transparent circular area. Within this area, the materials which compose the upper mirror are located. Below the bridge, two electrodes are shown. When an electrostatic potential is applied between these electrodes and the bridge, a force is generated between the electrodes and the bridge which draws the bridge closer to the substrate. The bridge is patterned at both edges in order to reduce the associated stress in the filter area. The patterned area extends between the center portion of the bridge which supports the filter and the edge of the bridge which drops down and contacts the substrate. Most of the bending of the bridge structure is confined to the patterned area.
Nickel bridge structures were fabricated using the process described above with copper as the sacrificial layer and electroless plated nickel as the beam material. Figure 5 shows a top view of the mirror region of a device. A grid structure is used to improve the flatness of the mirror. Since the mirror is fabricated from multiple thin layers, built in stresses may cause the mirrors to warp. By reducing the size of the mirror under a critical dimension, mirror buckling can be avoided.
Strain measurements were performed on plated nickel test structures using strain diagnostic structures reported previously[2]. The built-in tensile strain was found to be less than 10-3. A strain gradient on the order of -0.2 mm-1 is observed. This negative gradient is significant and can cause thin film structures to curl up and away from the substrate.
A variety of spectrometer designs are included in our test chip. Their total mirror size varies from about 4 mm square for the largest down to 0.05 mm for the smallest. The built-in strain causes the larger structures to buckle or deform slightly. This deformation can be detrimental to the performance of the devices.
Preliminary work has been conducted to determine the suitability of silicon and silicon dioxide for the multilayer mirrors. Mirrors were deposited on glass slides. The reflectivity of the mirrors was measured and found to have a broad band over the visible spectrum (400 nm to 800 nm) with a maximum reflectivity of 80% and an average value of about 75%. The slides were clamped together and examined under an optical microscope. The photomicrograph in Figure 6 below shows the color fringes obtained. A 546 nm interference was placed in the optical light path of the microscope. The narrow fringes that appeared were measured and their half widths compared with the spacing between fringes to determine the resolution. From these measurements we determined that the resolving power was approximately 64 which compares well with the predicted value.
