During the chip shortage that followed the pandemic, car factories around the world stopped production lines full of nearly finished vehicles. The missing parts were often not cutting-edge processors but cheap, unglamorous microcontrollers – chips that cost a few dollars and sit inside window motors and engine controls. It was a vivid lesson in something the industry had known for decades: how semiconductors are made is one of the longest, most specialised and most geographically concentrated manufacturing processes on Earth, and when one link slips, everything downstream feels it.
This article walks through the chip manufacturing process step by step, then explains why the semiconductor supply chain has become a matter of national strategy rather than just procurement.
From sand to wafer
Chips start as silicon, the second most abundant element in the Earth's crust, but the silicon has to be extraordinarily pure – far purer than solar-grade material. Quartz is reduced to metallurgical silicon, then chemically refined into polysilicon.
That polysilicon is melted and a single seed crystal is slowly pulled from the melt while rotating, a technique called the Czochralski process. The result is a large cylindrical ingot that is one continuous crystal. The ingot is sliced into thin discs – wafers – which are then polished to be almost perfectly flat. Leading-edge fabs use wafers 300 mm in diameter; a single wafer will eventually hold hundreds of chips, or thousands of small ones.
The chip manufacturing process: building layers with light
A modern processor is built up in dozens of layers, and each layer goes through a repeated cycle of steps. Done hundreds of times, the whole sequence can take months from bare wafer to finished chip.
- Deposition – a thin film of material (an insulator, a metal, or more silicon) is laid down on the wafer.
- Photoresist coating – a light-sensitive chemical is spread across the surface.
- Lithography – light is shone through a patterned mask (the photomask), shrunk through precision optics, and projected onto the resist. Exposed areas change chemically.
- Development and etching – the changed resist is washed away, and the exposed material underneath is etched with plasma or chemicals, carving the pattern into the layer.
- Doping – ions are fired into the silicon (ion implantation) to change its electrical properties, creating the regions that make transistors work.
- Planarisation – the surface is polished flat (chemical mechanical planarisation) so the next layer starts level.
The lower layers form billions of transistors, the tiny switches at the heart of all digital logic. The upper layers are a dense web of copper wiring connecting them.
All of this happens in cleanrooms where the air contains vastly fewer particles than a hospital operating theatre, because a single speck of dust can ruin a chip. Fab workers wear full-body suits, and much of the wafer handling is automated.
EUV lithography: the bottleneck in the bottleneck
The smaller the features you want to print, the shorter the wavelength of light you need. For years the industry used deep ultraviolet light at 193 nanometres, with clever tricks such as immersion lithography and multiple patterning to print features far smaller than the wavelength.
For the most advanced chips, fabs now use extreme ultraviolet (EUV) lithography at 13.5 nanometres. Producing that light is astonishing engineering: tiny droplets of molten tin are hit by a powerful laser to create a plasma that emits EUV, which is then bounced off ultra-precise mirrors in a vacuum, because EUV is absorbed by air and by ordinary lenses.
A single company, the Dutch firm ASML, is the only supplier of EUV lithography machines, which themselves depend on a deep network of specialist suppliers, such as the German optics company Zeiss for the mirrors. That one fact tells you a lot about the semiconductor supply chain: some steps have exactly one source.
A note on names: when you read about "3 nm" or "2 nm" chips, those labels are marketing generations, not the literal size of any feature on the chip. They are useful for comparing a manufacturer's own roadmap, less so for comparing companies.
Testing, packaging and the "back end"
When the wafer is finished, each chip on it is electrically tested. The wafer is cut into individual dies, good dies are mounted in packages that connect them to circuit boards, and the packaged chips are tested again. This assembly, test and packaging stage historically clustered in other countries from the wafer fabs, notably across Southeast Asia.
Packaging used to be the less glamorous half of the industry. Not any more. Advanced packaging – stacking dies, placing several "chiplets" side by side on a shared base, connecting processors to high-bandwidth memory – is now essential for the most powerful AI accelerators, and capacity for it has itself become a constraint.
Who does what in the semiconductor supply chain
The industry is split into specialists:
| Role | What they do | Examples |
|---|---|---|
| Design software (EDA) | Tools used to design chips | Synopsys, Cadence |
| IP licensing | Reusable building blocks and instruction sets | Arm |
| Fabless designers | Design chips, outsource manufacturing | Nvidia, AMD, Qualcomm, Apple |
| Foundries | Manufacture chips for others | TSMC, Samsung Foundry, GlobalFoundries |
| Integrated manufacturers | Design and make their own | Intel, Samsung (memory), Texas Instruments |
| Equipment makers | Build the fab tools | ASML, Applied Materials, Lam Research, Tokyo Electron |
| Materials | Wafers, gases, chemicals | Shin-Etsu, SUMCO and many others |
The foundry model, pioneered by TSMC in the late 1980s, is what let companies like Nvidia exist without owning factories. It also concentrated leading-edge manufacturing: Taiwan produces the large majority of the world's most advanced logic chips, and South Korea dominates large parts of memory production.
Why chip supply chains matter
Several structural features make the semiconductor supply chain unusually fragile:
- Extreme capital intensity. A leading-edge fab costs many billions of dollars and takes years to build and ramp up. Supply cannot respond quickly to demand spikes.
- Single points of failure. Some tools, materials and processes have one or two qualified suppliers worldwide.
- Long lead times. A chip can cross borders many times between raw material and finished device, and the process takes months.
- Geographic concentration. Earthquakes, droughts (fabs use large amounts of ultra-pure water), power outages or geopolitical tension in a single region can affect global supply.
- Mature chips matter too. Cars, medical devices and industrial equipment depend on older-node chips with thin margins, where few companies are eager to add capacity.
Governments have responded. The United States passed the CHIPS and Science Act in 2022, and the European Union adopted its own Chips Act in 2023, both offering subsidies for domestic manufacturing; Japan, South Korea, India and China have all pursued their own programmes. Export controls on advanced chips and chip-making equipment have become a central tool of technology policy. Whether these efforts meaningfully diversify production will take years to judge, because fabs are slow to build and the expertise to run them is scarce.
What this means for businesses that buy hardware
You do not need to make chips to be exposed to the supply chain. Practical takeaways:
- Know which of your products depend on specific chips, especially microcontrollers and power components with a single source.
- Design for alternatives where possible – boards that can accept parts from more than one vendor.
- Hold strategic buffer stock of critical, low-cost components rather than running purely just-in-time.
- Watch lead times as a signal, not just prices.
The bottom line
Understanding how semiconductors are made explains why chips are both everywhere and hard to produce: hundreds of precise steps, cleanrooms, machines made by a single supplier, and a supply chain that circles the globe. That combination is why a shortage of cheap parts can stop a car plant, and why governments now treat chip manufacturing as critical infrastructure.
