I decided to design and build a scanning electron microscope (SEM) in my home workshop to see if it was even possible. Spoiler alert: it is. I didn’t originally intend to create an SEM that could compare to a $75,000 entry-level commercial model, but the project turned out more successful than I expected. It produces clear, accurate images, and after some improvements that I’m currently working on, it may be practical for hobbyists to build an SEM that’s suitable for scientific research for under $2,000.
How SEMs Work
Ordinary optical microscopes shine visible light onto or through a specimen and use lenses to create a magnified image. This works well for many applications, but light can only resolve features greater than about 200 nanometers for visible light. This is small, but not small enough for looking at many interesting biological and material structures. You can use light with a shorter wavelength (i.e. ultraviolet light) to obtain slightly better resolution, but this adds a lot of expense and difficulty for only incremental improvement.
Electron microscopes, in contrast, offer a tremendous improvement in resolution. Like photons, electrons have both particle and wave-like properties, but the wavelength of a fast-moving electron is substantially shorter than that of visible light.
The SEM scans a tiny beam of electrons across a sample, following a raster pattern, and measures the amount of electrons that bounce off each point and onto a nearby detector. For example, if the beam hits a hole in the specimen, the electrons may become trapped and won’t reach the detector, but if the beam strikes a protrusion on the surface, many electrons will reach the detector since the protrusion provides more surface area than surrounding flat areas.
In this way, the SEM builds up its image pixel by pixel, and the device’s maximum resolution is determined by 2 attributes of the electron beam: its spot size and its scan rate. A smaller spot size will resolve greater detail, and slower scanning improves resolution by raising the signal-to-noise ratio at each spot. So that the electrons are not absorbed, the sample must be conductive or else coated with a thin layer of metal.
This method lets you image 3D objects over a wide range of magnifications without slicing them to bits, and the images it creates look like black and white photographs with a high depth of field. These attractive image qualities make SEMs very common for studying small 3D objects, and they influenced my choice to build this type of electron microscope.
Create a Vacuum
One challenge of SEMs is that the electron beam and the specimen must be manipulated within a vacuum. If the electron beam were fired through air, the electrons would strike gas molecules and scatter, blurring and destroying any image. For the electrons to travel unimpeded from source to sample and from sample to detector, you need a vacuum about a million times lower than atmospheric pressure, or 0.00076 torr, where a torr is the unit of pressure required to support a column of mercury 1mm high. Atmospheric pressure is about 760 torr at sea level.
You can get to these low pressures a few different ways, but my favorite (the least expensive) is by combining a mechanical rotary pump and a diffusion pump, plumbing them in series. The mechanical pump reduces the pressure by about 4 orders of decimal magnitude, and the diffusion pump takes it down another 2. For the rotary pump, I settled on a $150 air conditioner pump from Harbor Freight, and for the diffusion pump I bought an air-cooled 3″ Varian pump on eBay for about $200.
Diffusion pumps operate by creating high-speed jets of hot oil vapor that push air molecules out of the vacuum chamber. Inside the pump, an electric element heats silicone oil into vapor. After the droplets are done bumping air out, the pump’s cooled walls condense them back into liquid, which drips down to the bottom to be boiled again.
I connected the rotary pump to the diffusion pump with ¾”-ID wire-reinforced tubing from McMaster-Carr, where I got most of the hardware and raw material (Figures A and B). The wire reinforcement prevents the tube from collapsing under vacuum. I also included a tee fitting between the 2 pumps and added a digital vacuum gauge that I bought on eBay for about $100. The gauge reads from 0.001 to 12 torr, and was made for refrigeration technicians to use with a vacuum pump.
I didn’t have a commercial vacuum chamber, and I wanted the microscope to work inside a transparent enclosure, since its main purpose would be demonstration. So I used a glass bell jar that I found on eBay a while ago. The glass thickness indicated that the jar was built for vacuum use, rather than just for ornamentation or dust shielding. For a base, I used a 1″-thick aluminum plate. I cut a hole in the plate to fit the diffusion pump and machined a water-cooled baffle to go between the pump and plate (Figure C).
The baffle prevents diffusion pump oil from migrating into the bell jar. Boiling oil gets messy, and getting even small amounts of oil into the sensitive parts of a SEM would cause many problems. Air molecules can pass through the baffle’s tortuous pathway, but hot oil molecules condense on its water-cooled surfaces and drip back down.
I cut another hole in the aluminum base plate and added an additional vacuum monitor called a Penning gauge, also bought on eBay for about $250.This device measures vacuum from 0.001 down to 10–8 torr, and will indicate when the diffusion pump has taken the chamber pressure down to the range necessary for SEM operation.
The first time I pumped down the jar, I started the rotary pump, then exited the garage and shut the door behind me. If the jar imploded, I would be far enough away to escape the wreckage. But below a pressure of 0.01 torr, variations in pressure don’t much affect the strength needed for a vacuum chamber. This is a key point that often tricks people. Once you remove 99% of the air molecules, there are so few left that they exert almost no pressure on the inside wall. Removing more doesn’t change much. If a container can withstand 10–1 torr, it can probably safely hold 10–11.
Spark Plug Power
Ordinary automotive spark plugs are designed to supply insulated high voltages through metal walls and across pressure differentials, so I used them to bring power for the electron gun into the SEM chamber.
I drilled and tapped a series of holes in the base plate to hold the spark plugs, and added O-ring glands. I also made some low voltage pass-through connections for other circuitry using wide-head screws sealed to the plate with Buna-N (nitrile) washers. And to let users move a small stage to locate the specimen under the electron beam, I added spring-loaded teflon shaft seals that transfer rotary motion through the base plate while the chamber is under vacuum.
The Electron Gun
There are many ways of generating electrons for an electron microscope, but the easiest is to simply heat up a piece of wire. This goes by the exciting name of thermionic emission, and these filaments are used in vacuum tubes and cathode ray tubes; they make the orange glow inside the back of old TVs and radios. From eBay I bought a set of tungsten filaments with ceramic insulator holders that were originally made for use in commercial SEMs.
I connected the filament to a low-voltage power supply that I built from a Variac variable transformer, isolation transformer, bridge rectifier, and smoothing capacitors. I originally fed low-voltage AC to the filament, but that resulted in image quality issues, so I designed an unregulated but smoothed DC power supply.
Once the filament is glowing, it emits lots of electrons in all directions. To motivate them into a single direction, you need to apply high voltages across pieces of metal strategically arranged around the filament. The whole assembly is termed an electron gun, and when the applied voltage is 10kV, my gun shoots electrons out in a stream at about 2% of the speed of light (6,000,000 meters/second). To supply this voltage, I use a regulated high-voltage supply that I bought at a surplus sale, and I can adjust the voltage to fine-tune the electron velocity.
Focus the Beam
The beam from an electron gun is narrow, but not nearly fine enough for useful electron microscopy. To focus the beam, an SEM needs to run it through electron optics — controlled apertures and electric or magnetic fields that bend and shape the beam something like the way glass lenses bend the paths of photons.
Most commercial SEMs use magnetic fields to focus the beam because of their bending power and lower voltage requirements, but I used electric fields, because they don’t require custom-machined precision iron pole pieces. I used copper pipe and teflon insulators to construct 2 electrostatic lenses, which are nothing more than 3 lengths of conductive pipe, insulated from each other and arranged inline. As the electrons pass through the charged pipes, their trajectory is affected by the polarity and magnitude of the voltage applied to each. With the correct voltage and geometry, the beam of electrons will focus down to a tight spot on the specimen.
Scan the Sample
In the first SEMs, the process of scanning the sample and displaying the image were intertwined. The scan beam was steered over the sample in sync with a raster pattern that the CRT beam traced over the screen phosphors, and the microscope’s emission detector was used to drive the beam intensity in the CRT.
I took this same approach because of its simplicity; to capture images I currently aim my camera at the screen. But for the next version of my SEM, I am implementing a digital image storage system that will record the sample’s surface emission (image brightness) pixel by pixel.
To perform the synchronized scan and display, I bought 2 identical analog oscilloscopes (eBay again) and took one of them apart. Analog oscilloscopes use oppositely charged pairs of metal plates to deflect the electron beam in their CRTs, with plate size, spacing, and applied voltage determining the amount of deflection. So I removed the CRT from the disassembled scope and rerouted the wires driving its x-axis and y-axis deflection to smaller plates mounted in the SEM column.
To create the horizontal and vertical scan patterns, I built a simple raster generator similar to what’s inside a TV, but made from 555 timer chips. I fed its output into both the hacked scope, for steering the SEM beam, and the intact scope set to x-y mode, for driving the display.
Pick Up the Signal
To generate its signal, the SEM detects the quantity of electrons that are emitted from the specimen surface as the electron beam strikes it. But it’s a relatively small number within a small range, so it needs to be amplified.
To accomplish this, the electrons are attracted toward a phosphor screen which converts them into flashes of light. The flashes of light are then converted back into electrical signals and amplified by a photomultiplier tube, which consists of a photocathode that produces electrons when hit by photons, and a series of 12 dynodes that generates an avalanche of electrons about 106 greater in number than the first bunch. The detector is positioned to one side of the stage. It has a curved light guide so that the phosphor screen faces the sample and the photomultiplier runs up vertically.
The signal from the multiplier tube is then fed into the z-axis or blanking input on the intact oscilloscope. At fairly fast scan rates, the oscilloscope will then display an image from the SEM at live video rates.
So far, I’ve just used the SEM to image conductive objects (Figure F), since nonconductive objects must be coated with a vanishingly thin layer of metal before being imaged, done in a sputtering chamber. I may end up building one from a power supply and vacuum chamber.
Biological samples must be dried via special means so that the sample doesn’t lose its structure as the water evaporates. You can repeatedly soak the sample in alcohol until the alcohol replaces the sample’s internal water almost entirely. Then the sample is placed in a chamber and submerged in liquid CO2 at about 700psi. Finally, the CO2 is heated under pressure until it becomes supercritical, a liquid with no surface tension. I’ve built a supercritical drying chamber and used it to make homemade aerogel.
Meanwhile, I’m also developing a detector system that uses an electron multiplier instead of a photomultiplier, for greater simplicity, purity of the signal path, and to allow the SEM to operate without a light shield (I use heavy black plastic) covering the bell jar, thus improving the signal-noise ratio.