MEMS device for optical cell analysis

The goal of my project is the recognition and selection of cells, using a optical MEMS device. The idea of Prof. Ho was the following :

This was the original idea of my work. I had to conceive a MEMS structure and make modifications depending what was physically possible to make.

At the same time, I had to work on the optical detection part. I didn’t know what were the different absorption spectra of cells? Is there any difference? Can we detect a variation of absorption between cells through such a short transmission? Do we need something else to increase the signal? Is the fluorescence a good way? I had to think about this, find some way, and experiment while working on the device itself.

MEMS was for me a totally new field and I had a lot of things to learn before building my device. I read some reference books, some lectures, and a lot of article concerning MEMS fabrication. I asked a lot of information from different persons. It was a time for researching ideas and collecting information. The Internet was very useful too, especially the forums. For each idea, I had to see if it was possible to realize or not. This part is difficult and in a constant evolution because it depends of many technical issues.

After I knew it was practically possible to realize it, I had to learn how to build it. Working in the Nanolab requires a lot of skill and knowledge, and there are many things to understand before manipulating. The Nanolab is not a free-access room, each machine is very specific and bad manipulations can be hazardous. I took a lot of training, in order to be able to use equipment important for my process.

The detection of the cells will be based on different absorption spectrum. So, we needed a spectrometer to detect the spectra. We choose an Ocean Optics device, which is small, relatively cheap, with different ranges of spectra (depend on the grating of course) and a correct sensitivity. This device is built on a 2048 pixels CCD array. The one we choose is optimized for the 200-850 nm band, with an optical resolution of 1.3 nm (25 m m slit). For more technical details, see appendix 1.

Since we wanted to use UV-Vis, we needed special fibers for this range. Ocean Optics can provide this kind of fiber (the transmission graphs are in appendix 2) with SMA connectors for light source and spectrometer. For my application, I needed to implant a bare fiber in my MEMS.

The light source is a problem. It is difficult to find a compact UV-Vis light source. Ocean Optics provides us a Xenon pulsed Lamp. This source can emit a full range from 250 to 800 nm (see appendix 3 for emission spectrum), with pulses of 5 ms. the rate varies with the acquisition board. This lamp is not adapted for fluorescence, because the near UV intensity is very low, and it is a pulsed lamp. We tried another pure UV light source and Argon laser for fluorescence detection.

Because we shared this equipment with another lab, which wanted small equipment, we choose a PCMCIA acquisition board. The main advantage is that it can be used from an electronic notebook. The sample rate is slower (100 kHz) than a classic PCI board, but speed is not a critical issue for us.

The spectrometer gives us a signal, via the acquisition board. The software, given with the spectrometer, was just able to give us the spectrum, without interactivity. That is why we bought the Lab View drivers, for more control and extension possibilities.

Fig 4: screen shot of the Lab View interface

The first idea was to send a white light (full spectrum) through the cell and to compare the signal from the transmission with the original spectrum. The software can calculate the absorption with S the spectrum of the signal, D the dark current, R the reference spectra. I tried several sample preparations of cells, in a 1-centimeter width cuvette holder. I never saw any interesting signal. My guess is the absorption of light is very weak, so the difference with the reference is tiny. Moreover, the light source is not very powerful, especially in the UV, so (S-D) and (R-D) are small, and the detector noise can totally perturb the logarithmic division, giving incoherent spectra. There were thousands of cells in each sample, so, if I couldn’t detect any signal here, it would not be better with one only. So, I tried to use a different way to characterize materials: fluorescence.

Another possibility to detect cells is to use fluorescent effect. The natural fluorescence of a cell is undetectable, but can be increased by using a fluorescent dye, which can bind to the cell. The problems with the fluorescence technique is there is an additional cost in dye and preparation, and it require a continuous light The signal is still very weak and need a sensitive device for measurement. For this optical detection part I worked with Jeff Wang.

Since we want to detect cells one by one, it was indispensable to work on a small scale. Jeff built some microchannel from 80 to 120 nm wide, in a Borofloat glass, and we used glass micro-capillary too. We used the same detector, a spectrometer, as before. It is not totally adapted to the fluorescence (a photomultiplicator with some filters is certainly more sensitive), but we wanted to try different dye and different light source, the easiest and cheapest way was to re-use the spectrometer. In the future, a more sensitive system can be envisaged. The main issue was to detect the signal from the emission, always much weaker than the excitation one.

Sample preparation: We had to fill the channel or capillary with a fluorescent liquid. We first tried with some ink from a marker pen, and it worked well. To be more realistic, we used different dyes, which can bind to DNA helix. Each dye bound to a specific DNA sequence, and thousands of dye particles can attach to a single DNA. The final sample has a high cost (more than 200 $ for 200 m l of YOYO1 dye), but DNA detection is very interesting itself, and this can be use for cells too.

Fig 5 : first configuration

We first tried to use a plate UV light source (300-420 nm), which provided a good intensity of low-wavelength light, suitable for fluorescence. The microchannel in its wafer was put down on the light source and a fiber linked to the spectrometer was adjusted by a 3D positioner above the channel, where the emission is maximum. We saw some signal with the DAPI dye, but only with concentrated solution (see fig 6)

The last attempt was to use an Argon laser, with a special dye optimized for the 488 nm absorption (see absorption/emission curves of the YOYO1 dye in appendix 4). We used a interferential filter to select only the 488 nm ray (hide the 515 nm one).

With the laser excitation, we detected the fluorescent signal from a 100 m m diameter capillary filled with diluted solution of YOYO1 dye + DNA. The set-up was not very accurate, and it was certainly possible to to better, but the signal was much stronger than when using the UV light source (see results fig 7)

On the fig 8, we can clearly see that the shape of emission is always the same. It is exactly the shape indicated by Molecular Probes for this dye (see appendix 4). This measurement were taken with the fiber probe at 90 degree from the direct laser beam, to avoid a too strong intensity, which saturate the detector. The capillary stayed vertical, so this can not be used for the microchannel in the wafer.

We imagined a horizontal configuration, with several adjustment possibilities, for more accurate measurement (fig 9). The laser arrive in the microchannel, induce fluorescence, which is caught by an optic fiber, close to the channel. For improving the amount of fluorescent light collected, a layer of metal will be deposed in the channel, to minimize the losses by the bottom. A lens can be insert between the fiber and the channel.

When I’m writing this report, we only had a few results with this configuration. Jeff will receive some equipment to set-up a new experiment, more accurate (an optical bench) with which we should be able to see fluorescence from a microchannel, using the argon Laser.

The micromachine techniques for fabricating micron-size parts are derived from integrate circuit (IC) manufacturing processes. The micromachines are made by using three basic technologies, which are bulk micromachining, surface micromachining and the LIGA process.

Bulk machining uses acid to remove the portion of the substrate which is not covered by the patterned photoresist. Many simple mechanical configurations can be produced by this method. Surface micromachining technique is built upon a simple and repeating process of depositing structure and sacrificial materials on top of each other and then removing the sacrificial materials by etching processes. The main advantage of the surface micromachining is that sophisticated spatial structures can be made. LIGA process uses synchrotron-generated x-ray to expose a patterned thick layer photoresist and to form the mold. By electroplating, high aspect ratio metal parts can be formed inside the mold.

In the Nanolab, also known as the cleanroom, there are no LIGA processes and this is not likely a problem for the device I wanted to do. There are a lot of different machines in the nanolab, and my device needs only few of them.

The first step of the elaboration of a MEMS is the conception of the structure, which relies on the techniques and idea available. All of the following steps are linked together, and a problem in any given step will affect the others.

The MEMS structure must contain a microchannel with transparent walls, two places for optic fibers and one place for the capillary. I designed several configurations for each part of the structure, changing with different technology I wanted to use.

After thinking of a specific shape, I had to transfer my idea for the fabrication technique to a more concrete support.

I used the software L-Edit Pro V8.0 from Tanner edition for the mask design. This software allows to work on several layers and provides a high resolution output file (postscript files).

I needed two mask layers, one for the SU-8 and one for another photoresist that provides protection during etching.

After some problems due to poor quality of printing (see section 4.5, ‘Mask Printing’), I designed a second generation of masks, with no complex shapes and more straight lines.

The reader certainly knows that MEMS are fabricated in a similar way to Integrated Circuits. It requires masks, lithography, photoresists, depositions, and a lot of elaborate equipment. The fabrication requires skills that take a substantial time to learn. Since the cleanroom is a potentially hazardous place, I took several training classes to properly gain access to the cleanroom, and to use some of the equipment. Since the cleanroom is a clean place, there are a lot of rules to follow for respecting the cleanliness. Nearly the entire body must be covered, requiring those who enter to wear a white suit, gloves, facial mask, protective goggles, hair cover and boots. After all my training, I began to work in the cleanroom, where I spent several weeks.

Equipment I am qualified to use: Spin coating, Karl Suss mask aligner, PECVD, Alphastep

Equipment I have been trained on: Spin coating, Karl Suss mask aligner, PECVD, Alphastep, DRIE, CVC sputtering, Ebeam Evaporator, furnace oxidation.

The most common way to get a mask for the lab is to send a postscript file to a printing company. The company will print the mask on a transparent film with a high-resolution printer. These masks cost approximately $20 - $30, quick to obtain, but the resolution is limited (more than 20 m m). There is a second solution, a real hard mask, consisting of metal mask on glass with a resolution of a few microns, but it takes more time and is much more expensive (several hundreds of dollars). For test or usual size devices, it is sufficient to use the less expensive masks.

On my device, the only critical parts are the channel and its walls. The other parts are big and do not need a high level of precision. We found a company, which claims a 5080 dpi resolution (5 m m). I sent the printing job (see appendix 6) and the result was quite unexpected. I designed several widths of channels (10, 20, 30, 40 and 50 m m) on the mask All the 10 m m and most of the 20 m m disappeared. Even the 30 m m and 40 m m were far from being perfect (see fig 12). Moreover, the walls were not straight and the small details were skipped due to the resolution of the printing process.

The only way to use this kind of quality of mask is to design something simpler, without such particular shapes. It is recommended to always prepare a selection of different dimension for the channel or other small details to ensure that at least one will be correctly printed.

All of these steps can only be realized in the cleanroom.

It is important to have a perfectly clean wafer, free of any dust or particle on the surface. That’s why the first step is always a cleaning step. Piranha is a very strong cleaner, mix of H₂SO₄ and H₂O₂ with a mixing ratio range from 50:1 to 3:1 (usually 5:1). 5 minutes are more than enough for a quite clean wafer.

After is the nitride deposition step. I used the PECVD to depose a 2500 å nitride layer. It takes 25 min, at a temperature 275^oC.

The third step, SU-8 deposition, is a very delicate step, which require several accurate manipulations. I followed this recipe:

• Dehydration bake 12 min at 150C
• 2 min HMDS (for better adhesion)
• Spin coating : 3 steps program :

1. 500 rpm - 10 sec - ramp 1000 rpm/s
2. 2500 rpm – 12 sec – ramp 200 rpm/s
3. 0 rpm – 0 sec – ramp 200 rpm/s

• Soft bake 75 to 95^oC for 30 min (2 min ramp)

The Karl Suss Mask Aligner makes patterning (step 3) with exposure with a strong UV light. The parameters are: 70 sec exposure (5.9 mA), 35 m m, softcontact. A good PEB (post exposure bake) is required : 40 min at 92C in a oven.

Step 4 is the developing. Ten min in the developer was a bit short. After 1 or 2 minutes more, the result can be better. Here I stopped my experimental process, the result was bad because of the mask problems (described in the previous section). It was not useful to continue the process with a bad base. In picture 14, we can see one of the best channels on the wafer, but far from the original mask design: no vertical sidewall, no straight line on the waveguide part, irregularities…

During step 5, another photoresist is added to delimitate the place of the etching. There is nothing particular here, the precision is not critical.

Step 6: the Fluorine RIE is the only way to drill the nitride layer, clearing the parts we want to be etched.

Step 7: Anisotropic etching with K-OH. Since the etching rate is very slow (around 1 m m/min), this process will take several hours to etch the deep V-grooves. One unknown: is how will the SU-8 react? It is supposed to be very strong once exposed, and the nitride layer should prevent the cantilever from bending, but the result is not certain.

So, because of mask problems, I had to design something a bit different, I tried to conceive another process too, quicker and simpler. As I said in the section "structure", the DRIE has the possibility of etching silicon vertically, so the cantilever can disappear. Moreover, it is possible to use the SU-8 as a mask for final etching too, so we don’t need another photoresist. Last thing, since the cantilever doesn’t exist anymore, the nitride step and Fluorine step can maybe be deleted.

I didn’t have the time to try this process, because of different problems with the DRIE machine in August. But I think it is a good way to explore in the future.