SEM and TEM Explained: How Electron Microscopes Work
An electron microscope makes things look bigger by shining a stream of electrons on them instead of light. Electrons behave as "waves" with a much shorter wavelength than light. So it can show shapes a few nm across, and even rows of atoms, that an ordinary light microscope cannot tell apart. SEM looks at the shape of a surface. TEM looks inside a thinly sliced sample.
What you'll learn
- Why a light microscope cannot resolve shapes finer than about 200 nm
- Why electrons let you see much finer detail
- How SEM (for surface shape) and TEM (for inner structure) work, and how they differ
- Why samples must go into a vacuum and be sliced thin
- How chip factories use them (measuring line widths, checking defects, failure analysis)
Prerequisites: none. The second half is easier if you know that chips are made by adding films, printing patterns and etching, over and over. "How Are Chips Made?" explains this in detail.
In everyday life: black-and-white 3D-looking photos
Have you seen black-and-white photos of pollen or an insect's eye in a science textbook? They look three-dimensional. Every single hair and every bump on the surface is sharp. Most of these photos were taken with an electron microscope.
Electron microscope images also show up often in chip news. Teardown reports on a new smartphone chip include cross-section photos of the chip, cut vertically. Only an electron microscope can capture the fine wires and tiny switches (transistors) stacked in many layers.
A human hair is about 80,000 to 100,000 nm wide, according to the US government's nanotechnology site. One nm (nanometer) is one-millionth of a millimeter. The patterns inside a chip are a few thousand times finer than a hair. This article explains how people "see" such a tiny world.
The limit of light microscopes: anything finer than the wave blurs
Light behaves like a wave. The distance from one wave crest to the next is called the wavelength. The wavelength of visible light is about 400 to 700 nm.
Waves spread out as they pass small gaps or the edges of objects. So an image made with light is always a little blurred, however good the lens is. The blur is about as large as the wavelength of the light. If two points are too close, their blurs overlap, and they look like one point.
The German scientist Ernst Abbe described this limit in 1873. Even under good conditions, two points can be told apart only if they are about half a wavelength apart. For a microscope using visible light, about 200 nm is the rough limit. This "smallest detail you can tell apart" is called the resolution.
There are ways to beat this limit with glowing markers (fluorescence), and they won the 2014 Nobel Prize in Chemistry. They are special methods for looking inside cells, so this article does not cover them.
So why not use light with a shorter wavelength? Ultraviolet light does show slightly finer detail. But light with an even shorter wavelength is absorbed by glass lenses and cannot pass through. "What Is EUV Lithography?" also explains this.
Electrons also behave like waves
Electrons are very small particles found inside every atom. They carry negative charge, and they are what flows as electric current in a wire. Yet an electron, while being a particle, also has wave properties, just like light.
The French physicist Louis de Broglie proposed this idea in 1924. Later, experiments shot electrons at crystals, where atoms line up in a regular pattern. The electron waves added up to form patterns, just as light waves do. These two results won the Nobel Prize in Physics in 1929 and 1937. "Quantum Tunneling in Transistors and Flash Memory" also covers electrons behaving as waves.
The faster an electron moves, the shorter its wavelength. Electrons are sped up by pulling them with a high voltage. This voltage is called the accelerating voltage, written in kV (kilovolts; 1 kV is 1,000 volts). An electron accelerated at 100 kV has a wavelength of about 0.0037 nm. That is about one 150,000th of visible green light (about 550 nm).
Electrons carry negative charge, so a magnetic force can bend their path. A coil of wire carrying current can focus electrons, much like a lens. Using this idea, Ernst Ruska of Germany built the first electron microscope with Max Knoll in 1931. By 1933 it clearly outperformed light microscopes, and Ruska won the Nobel Prize in Physics in 1986.
However, electron lenses cannot focus electrons as perfectly as glass lenses focus light. This kind of image distortion is called aberration. So the finest detail you can actually resolve is much coarser than the electron wavelength. Even so, the best electron microscopes see detail more than 1,000 times finer than light microscopes.
The inside of an electron microscope is a vacuum, with the air pumped out. If electrons hit air molecules, they change direction or stop. The part that emits the electrons (the electron gun) would also be damaged by air.
SEM: tracing the surface with a dot of electrons to see its shape
Electron microscopes come in two main types. The first narrows a thin stream (beam) of electrons even further and traces it across the sample's surface to build an image. This is called an SEM (scanning electron microscope). "Scanning" means moving step by step, as if tracing. The first commercial SEM went on sale in the UK in 1965.
An SEM focuses the electron beam into a tiny dot a few nm wide. It moves this dot a little at a time, left to right and top to bottom, like reading a page of a book. Where the beam hits, it knocks electrons out of the sample. These are called secondary electrons.
A part that catches these secondary electrons (the detector) counts them at each point. More electrons make the point brighter, fewer make it darker, and each point becomes one dot (pixel) of the image. Putting all the dots together gives one black-and-white image. So an SEM image is not formed by a lens. It is a grid of measurements, one per point.
Secondary electrons are weak, so only those born very close to the surface can escape. Slopes and corners have more material near the surface, so more electrons fly out there. That is why edges and slopes glow bright, and the image looks like a shaded 3D photo. High-end instruments can resolve detail finer than 1 nm.
Samples that do not conduct electricity need care. The electrons that hit them have nowhere to go, so charge builds up on the surface and distorts the image. To prevent this, the surface gets a very thin coat of metal or carbon, or the accelerating voltage is lowered.
TEM: passing electrons through a thin sample to see inside
The second type sends electrons through the sample and builds an image from the electrons that come out the other side. This is called a TEM (transmission electron microscope). "Transmission" means passing through. The first electron microscope Ruska built was this type.
A TEM usually accelerates electrons with a high voltage of 80 to 300 kV. Even so, electrons can pass only through very thin samples. So the sample is thinned to about 100 nm or less. That is about one-thousandth of the width of a hair.
Spots in the sample that are thick or rich in heavy atoms scatter electrons strongly. Scattered electrons rarely reach the image, so those spots look dark. The electrons that pass through are magnified by several stages of lenses and recorded by a camera. Here we only explain the most basic way an image forms.
The strength of TEM is seeing inner structure at the level of atoms. In a silicon crystal, the centers of neighboring atoms are about 0.235 nm apart. The best aberration-corrected TEM has been shown to resolve 0.05 nm. Lawrence Berkeley National Laboratory in the US announced this in 2008. Rows of atoms show up as lines of dots.
A sample hit by electrons also gives off X-rays. Each element emits X-rays with its own set energies. Studying these X-rays tells you which elements are at that spot. This analysis is called EDX, and it is attached to an SEM or TEM.
"SEM vs TEM" compares the two in a table, covering resolution, preparation work, speed and cost. In short, SEM is fast and easy, while TEM takes more work but shows finer detail, even inside.
Preparing samples: into a vacuum, and sliced thin
Samples for an electron microscope go into a vacuum. So things that contain water, such as living things, cannot be viewed as they are. They need preparation, such as drying or hardening. Silicon chips are dry solids, so in this respect they are easy samples.
Preparing for TEM is even harder. To look at a single defect on a chip, you must cut a thin slice exactly at that spot. This uses a machine that cuts the sample with a beam of electrically charged atoms (ions). It is called an FIB (focused ion beam).
An FIB focuses a narrow beam of ions of a metal called gallium. The ions knock atoms away bit by bit to carve the sample. First, a protective metal film is laid over the spot of interest. Next, deep trenches are dug on both sides, and the thin wall left between them is cut free and lifted out. Finally, it is thinned to about 100 nm or less and placed in the TEM.
Many FIB machines also have a built-in SEM, so you can watch the cutting as you work. Preparation often takes several hours. A chip that has been cut cannot be restored, so only some products are pulled out and examined.
How chip factories use them
Measuring line widths: CD-SEM
Chips are built on a thin, round plate of silicon called a wafer. After printing patterns with light, or after etching, the factory checks whether the line widths match the design. The special SEM used for this measurement is called a CD-SEM. CD stands for critical dimension, the key size to be measured. Hitachi (now Hitachi High-Tech) says it launched the first CD-SEM product in 1984.
A CD-SEM scans the pattern from above the wafer. It finds the line edges, where many secondary electrons come out, and calculates the width from them. The coating used to print patterns (photoresist) is easily damaged by electrons. So the measurement is done gently, at an accelerating voltage of about 1 kV or lower. The results are used to tune the pattern-printing step (lithography).
Checking defects: review SEM
Finding particles or missing bits of pattern across a whole wafer is the job of optical defect inspection systems. Light can check a wide area fast, but it cannot show the shape of the defects it finds well. So an SEM zooms in on each spot found and takes images to identify the type of defect. An SEM used this way is called a review SEM.
Recent machines sort defects into types automatically from the images they take. Knowing which defects increased at which step helps find the machine or setting at fault quickly. This is important work for raising the share of good chips (the yield).
Finding the cause of failures: cross sections with FIB and TEM
When a finished chip does not work, finding where it is broken is called failure analysis. Electrical tests narrow down the suspect area. Then an FIB exposes a cross section there, and an SEM or TEM looks at it. Engineers find things like gaps in wiring, peeling films and foreign particles.
Some films inside a transistor are only a few nm thick. One example is the very thin insulating film that separates the part that switches current on and off. TEM is essential for checking the thickness of such films and the boundaries between layers. Adding EDX also shows which elements are in which layer.
Common myths
Myth: Higher magnification lets you see any detail you want
In fact, what decides how fine you can see is resolution, not magnification. Enlarging a blurred image only makes the blur bigger. Even SEM magnification is just the ratio between the screen size and the size of the scanned area. Raising only the magnification of a light microscope will not resolve anything finer than about 200 nm. The myth is common because microscope ads tend to highlight magnification numbers.
Myth: Optical inspection cannot find defects smaller than 200 nm
In fact, "telling a shape apart" and "noticing something is there" are different things. Even a very tiny particle scatters a little light when light hits it. By catching that scattered light, you can tell something is there, even without knowing its shape. Wafer inspection machines use this to find defects much smaller than 200 nm. Checking the shape is the later job of the review SEM.
Myth: An electron microscope can see inside a chip without damaging it
In fact, an SEM can look at a wafer from above without damaging it. That is what a CD-SEM is for. But to see inner structure with a TEM, an FIB must cut out a thin slice, and that chip can no longer be used. Cross-section photos are easy to find, so people tend to assume every view is non-destructive.
By the numbers
| Number | What it is |
|---|---|
| About 80,000–100,000 nm | Width of a human hair (US National Nanotechnology Initiative) |
| 400–700 nm | Rough range of visible light wavelengths |
| About 200 nm | Rough limit of detail a light microscope can resolve (about half the wavelength of light) |
| About 0.0037 nm | Wavelength of an electron accelerated at 100 kV (about one 150,000th of green light) |
| 0.4 nm | Example spec of a high-end SEM (Hitachi High-Tech SU9000II, 30 kV; as of 2026) |
| 0.235 nm | Distance between the centers of neighboring atoms in a silicon crystal |
| 0.05 nm | Resolution confirmed with an aberration-corrected TEM (announced in 2008) |
| About 100 nm or less | Typical thickness of a TEM sample |
| 1931, 1965, 1984 | First electron microscope, first commercial SEM, first CD-SEM product (according to Hitachi) |
The limit of a light microscope is a few hundred times smaller than a hair's width. Electron microscopes show detail hundreds to thousands of times finer still. Making and checking leading-edge chips depends on this power to see.
Learn more
- See the differences in a table: "SEM vs TEM" (resolution, preparation, speed and uses) and "EDX vs SIMS" (two ways to find out which elements are present)
- Term pages: SEM, TEM, FIB, CD-SEM, and the AFM (atomic force microscope), which traces a surface with a needle
- Try wavelength and resolution with numbers: the lithography resolution calculator calculates the limit of printing patterns with light. As with microscopes, you can check that a shorter wavelength gives finer detail.
- Related articles: "Semiconductor Manufacturing Equipment" (where measurement and inspection tools fit), "What Is EUV Lithography?" (printing patterns with short-wavelength light), "Quantum Tunneling in Transistors and Flash Memory" (the wave nature of electrons) and "What Does 2nm or 3nm Mean?"