Sandwiched between transparent glass and PDMS the channels were ghostly white until liquids were pumped through them, at which point the networks they composed stood out brightly in 3D space.
Growing up I would spend hours playing with a hose in a sandbox, making rivers and dams and watching them get destroyed by the chaotic forms that flowing water wants to take, and will take, regardless of the constraints imposed on it by its environment. The surprise and delight I found in those experiences resurfaced later on when studying modern efforts to yoke the dynamism of flowing fluids to techniques for manipulating the small-scale machinery of life (and other similar systems), an engineering domain known as "microfluidics". To me microfluidic technology is just another sandbox in which to delight in the struggle to understand the chaos of the natural world, and a lens through which to understand its beauty.
Microfluidic devices manipulate tiny amounts of liquid (femtoliters to microliters) and are utilized to help carry out tasks ranging from gene sequencing to inkjet printing. It's possible to make complex functional networks in microfluidics that are analogous to complex electrical circuits, but while electrical circuits are opaque to direct human observation, microfluidic plumbing is often literally transparent with inner workings that can be followed with the naked eye or a simple microscope.
During undergrad at MIT I took a course that was called Nanomaker. Nowadays the course materials are up on MIT OpenCourseWare free for anyone to access. The premise of the course was to replicate several different types of nanotechnology with cheap and simple tools and materials that anyone could get access to. Besides simple microfluidics devices we made electroluminescent panels, OLEDs, solar panels, and several other interesting devices. It was my all-time favorite course in undergrad because it was very light on lectures and heavy on hands-on tinkering with a wide variety of interesting technologies.
Making Microfluidic Channels Cheaply
(Read the lab overview lecture slides from the course for more details on the following)
To make microfluidic devices you need to be able to make tiny pipes that are connected in a way that is dictated by your design. There are many ways to go about this, from directly cutting tiny channels with lasers to milling them at a large scale and then shrinking them down with heat. In our lab work we used a technique that used casting to create the channels from a mold instead of by cutting them directly out of a material. The key was a material called PDMS, which is a widely used silicone. PDMS has a useful property in that when it is cast it can capture features of the mold on the scale of nanometers. It's food-safe, easy to work with, and nonreactive so it's basically the perfect material for making microfluidics devices.
With PDMS as the bulk material for our devices the hard part of the process becomes patterning the mold with the channels that we want to cast for our device. In industry this is often accomplished with standard silicon photolithography techniques. But that's hard to pull off for someone without access to the millions of dollars of equipment and clean room necessary for that work. So instead we made our mold using a thermal-transfer wax printer and standard overhead transparencies. By creating an image with our channels in black and then printing it in wax on a transparency we had our mold with the channels about 10 to 50 microns "tall".
Once the positive mold for the channels was realized in wax on the transparency the rest of the process was simply to cast the PDMS on top and then bond it to glass to close the channels. The casting part was very standard - we removed bubbles from our PDMS with a vacuum chamber and then poured carefully onto our mold. We used a hot plate to accelerate the curing process. Then we just peeled off the transparency and cleaned off any residual wax.
Bonding the PDMS to a glass microscope slide was a little more involved. The transparencies weren't necessarily pristine so there were oils and other trace contaminants on the surface of the PDMS after peeling away the transparency, and the same was true for the glass slides. If we tried to bond the PDMS to the glass without cleaning all those contaminants away then some parts might not stick and then the channels would leak. So first we cleaned both carefully. Then we needed a way to get the PDMS to stick to the glass without using an adhesive, because that would clog up the channels.
It turns out that PDMS and glass are chemically very similar, and if you expose PDMS to a plasma containing oxygen then its surface basically turns into glass. To get plasma we used a corona discharge wand, which is essentially a wire that is brought up to a very high voltage so that the air around it is ionized. After exposing the PDMS to the corona we could simply press it down on the glass microscope slide and the two would bond together on a chemical level.
After that we had our channels sealed between PDMS and glass. All we had to do to use them was poke holes through the soft PDMS to inject fluids. For experimentation we just used syringes but the microfluidic device could be made part of an automated system by hooking it up to digitally controlled pumps instead, as is often done in commercial microfluidics systems.
Making a Microfluidic Display
I was captivated by the way that the microfluidic devices looked. Sandwiched between transparent glass and PDMS the channels were ghostly white until liquids were pumped through them, at which point the networks they composed stood out brightly in 3D space. Seeing that visual effect made me want to create a microfluidic device that I could manipulate visually. I decided to attempt to make a simple display that could show arbitrary images by arranging colored droplets in a 2D field.
Plumbing a Plane
The display needed to fill a 2D field with channels and I needed to be able to control what droplet ended up at each coordinate on it. In another course I had homework where we used turtles graphics to draw art and for my submission I had the turtle draw an arbitrary image by following a Hilbert curve. Such patterns are known as space-filling curves because they make it possible to reach any point in a >1D space by sliding along the curve an appropriate distance. This is neat because it effectively maps a 1D space to a 2D or higher dimensional space, which is exactly what we need for our microfluidic display. At the input of the display channel we control the color of droplets we are feeding in, and by knowing the geometry of the channel we can arrange the 1D sequence of droplet colors so that the image we want is formed in the 2D space that the channel fills.
I wanted to play with this idea before building the actual microfluidic device so I made a quick simulation in Processing.
If you think for a second about what kinds of channels might work it becomes apparent that the complication from the Hilbert curve isn't really necessary if we just want to get a droplet to any spot on our 2D display. There are a lot of other simpler ways to fill that space, like a zig-zag line that goes straight from one side to the other and back, filling the space row by row. But consider the engineering limitations that may be encountered with the real display. For this concept to work we need to have very precise control over the volume of each droplet, because that determines the amount of space that the "pixel" takes up in the channel. If we have error in the volume of each droplet it will accumulate over the entire image because a droplet that is too big or too small will push all the droplets ahead of it too far or not far enough down the channel so that they don't end up in the right final positions. Here is what varying amounts of droplet size error looks like for a version of this display that arranges the display channel in a simple zig-zag:
As the error accumulates it becomes much harder to understand the image because the droplets can move much further away from each other than in the display with the Hilbert curve. The Hilbert curve version of the display has the nice property that if you shift the droplets around they will maintain their 2D neighborhood better than any other space-filling curve. Also it just looks a lot cooler.
Building the Prototype
Here's the pattern that I ended up creating for my display prototype:
On the left you can see the ports for injecting liquids into the display. The two ports connected to the channel start were meant to each be used with a single color. I didn't know whether the differently colored droplets mixing or interacting would be an issue so I just did the simple thing first.
After the fabrication process described above I had two copies of my design on a glass slide. Unfortunately the quality of the print was poor because the wax sagged a bit and spread out. As a result there wasn't enough space between the channels so some of the channel edges had too little bonded area. During testing that caused the walls of the channels to lift up and leak.
Sadly I don't have any good footage of the tests that were run with the prototype. But from the following GIF you should be able to see the first small region of the display channel fill correctly with food coloring.
The two prototype devices failed in the same way. The channel was partially blocked in some places so the pressure built up as fluid was injected and this eventually caused portions of the channel to lift off of the glass. As soon as this happened there was no way to recover because it would no longer be possible to have precise control over the positioning of the droplets.
After that first test I stopped working on the project. But it has been kicking around in my head since I was working on it in 2014. I still find the visuals of microfluidics very compelling, and I think I may find a home for a new attempt in a future art project.
I was recently excited to see this project from researchers in the Tangible Media Group in the MIT Media Lab showing a lot of similar devices with the same spirit that inspired my project - there's beauty in using microfluidics devices for visual displays. Interestingly one of their demos uses the same idea of laying a channel out on a Hilbert curve (but executed much better). I highly recommend checking out their work: