A single microchip in your smartphone contains over 16 billion transistors. Each one is smaller than a virus, yet together they perform trillions of operations per second. The journey from raw quartz sand to a functioning processor involves over 1000 individual steps, takes three months to complete, and requires environments 10,000 times cleaner than a hospital operating room.

The process begins with one of Earth’s most abundant elements: silicon. But the silicon in your processor bears little resemblance to beach sand. Semiconductor-grade silicon must reach purity levels of 99.9999999% (nine nines purity) – meaning impurities are measured in parts per billion. To achieve this, manufacturers subject raw silicon to chemical purification processes that transform it into electronic-grade polysilicon. This ultra-pure material is then melted and crystallized using the Czochralski method: a seed crystal is dipped into molten silicon and slowly withdrawn while rotating, pulling a single crystal ingot that can weigh over 100 kilograms and extend nearly two meters.

A silicon ingot and wafers showing the crystalline structure
A silicon ingot and wafers showing the crystalline structure

Image source: ASML

This crystal is sliced into wafers approximately 0.7 millimeters thick and 300 millimeters in diameter – roughly the size of a compact disc. Each wafer undergoes precision polishing until its surface varies by less than a nanometer across its entire area. At this point, the wafer is ready to become the foundation for billions of electronic switches.

The Architecture of Light: Photolithography

The core challenge of chipmaking is transferring circuit patterns onto silicon. This process, called photolithography, works similarly to photographic printing but at nanometer scales.

A light-sensitive polymer called photoresist is coated onto the wafer. When exposed to specific wavelengths of light through a patterned mask (reticle), the resist undergoes chemical changes. Areas exposed to light become either more soluble (positive resist) or less soluble (negative resist) in developer solution. The pattern is then revealed when the wafer is washed, leaving behind a protective stencil for subsequent processing.

The wavelength of light determines how small these features can be. Traditional deep ultraviolet (DUV) systems use 193-nanometer light from argon-fluoride lasers. But as transistors shrank below 7 nanometers, even these wavelengths became inadequate due to fundamental physics: you cannot resolve features smaller than roughly half the wavelength of light used.

The EUV Revolution

The semiconductor industry’s answer was extreme ultraviolet (EUV) lithography – a technology so difficult to develop that it took over 20 years and billions of dollars to achieve commercial viability. EUV uses light at 13.5 nanometers, approaching X-ray territory.

Generating this light requires an almost absurd process. A high-power carbon dioxide laser fires at tiny droplets of molten tin, moving at speeds over 100 meters per second. The laser hits each droplet twice: first to flatten it into a pancake shape, then to vaporize it into plasma that emits EUV photons. This happens 50,000 times per second, creating a continuous stream of 13.5-nanometer light.

EUV light cannot travel through air (it is absorbed by oxygen) and cannot be focused with glass lenses (glass absorbs it). The entire optical path must be in vacuum, and all focusing must use bragg reflectors – mirrors with alternating layers of silicon and molybdenum, each just a few nanometers thick. A single EUV machine contains over 100,000 components and costs more than $150 million.

The latest generation, High-NA EUV, increases the numerical aperture from 0.33 to 0.55, enabling even finer resolution for sub-2-nanometer nodes. These machines use mirrors the size of coffee tables polished to atomic-level smoothness.

Etching: Carving at Atomic Scales

Once a pattern is transferred to photoresist, the underlying material must be selectively removed. This is etching – the silicon equivalent of sculpting.

Wet etching uses chemical baths that dissolve exposed materials. It is simple and inexpensive but has a critical limitation: chemicals etch in all directions equally (isotropic etching), making it impossible to create truly vertical sidewalls at nanometer scales.

Dry etching uses plasma – ionized gas that bombards the surface with chemically reactive species. By carefully controlling gas chemistry, pressure, and ion energy, manufacturers can achieve anisotropic etching where material is removed primarily in one direction, creating the sharp vertical features essential for modern transistors.

Computer visualization of nanostructures on a computer chip
Computer visualization of nanostructures on a computer chip

Image source: ASML

Advanced memory chips like 3D NAND stack over 200 layers of memory cells vertically. Each layer must be etched with perfect uniformity, requiring precise control of plasma chemistry and etch rates across the entire wafer surface.

Doping: Creating Electronic Switches

Pure silicon is a semiconductor – not quite conductor, not quite insulator. By introducing specific impurities (dopants), manufacturers can create regions with excess electrons (n-type) or excess holes (p-type). Where these regions meet, a PN junction forms – the fundamental building block of transistors.

Ion implantation accelerates dopant atoms to high velocities and shoots them into the silicon crystal. Boron creates p-type regions, while phosphorus or arsenic create n-type. The depth and concentration of dopants are precisely controlled by adjusting acceleration voltage and implantation dose.

After implantation, thermal annealing at temperatures exceeding 1000°C repairs crystal damage and activates the dopants by incorporating them into the silicon lattice. Modern processes use rapid thermal annealing – heating the wafer to extreme temperatures for just seconds to minimize unwanted diffusion.

Deposition: Building Layer by Layer

Between lithography steps, thin films must be deposited to create insulators, conductors, and semiconductors. Three primary techniques dominate:

Chemical vapor deposition (CVD) introduces reactive gases that decompose on the heated wafer surface, leaving behind solid films. Variants include low-pressure CVD, plasma-enhanced CVD, and high-aspect-ratio CVD for filling deep trenches.

Physical vapor deposition (PVD) physically ejects material from a solid target using plasma sputtering or evaporation. This is commonly used for metal interconnects.

Atomic layer deposition (ALD) builds films one atomic layer at a time through alternating self-limiting surface reactions. This provides exceptional conformality – the ability to coat complex 3D structures uniformly – making it essential for modern gate dielectrics and barrier layers.

Planarization: Maintaining Flatness

After each processing cycle, the wafer surface becomes increasingly irregular. Chemical mechanical planarization (CMP) uses a combination of chemical slurry and mechanical polishing to flatten the surface. The wafer is pressed against a rotating pad while slurry flows between them, removing material at rates measured in nanometers per minute.

CMP enables true 3D integration by ensuring each new layer starts from a flat surface. Without it, depth-of-field limitations in lithography would make patterning impossible on non-planar surfaces.

The Transistor Evolution

Transistor architecture has fundamentally changed as dimensions shrank. Planar transistors, where current flows horizontally along the surface, dominated until around 22 nanometers. Below that, short-channel effects caused excessive leakage.

FinFETs solved this by wrapping the gate around a thin silicon “fin” that rises vertically from the wafer. This three-sided gate provides better electrostatic control, reducing leakage while maintaining current drive. FinFETs have enabled nodes from 22nm down to 3nm.

At 3nm and below, FinFETs are giving way to gate-all-around (GAA) transistors, where the channel is a horizontal nanowire or nanosheet completely surrounded by the gate material. This provides even better control and allows continued scaling. GAA architecture offers approximately 30% improvement in power efficiency compared to FinFETs at equivalent nodes.

Packaging: The Final Connection

After months of processing, completed wafers are tested and diced into individual chips. Each die is then packaged – connected to external electrical contacts and protected from environmental damage.

Traditional wire bonding uses fine gold or copper wires to connect the die to package leads. But as speeds increased and pin counts grew into thousands, this approach became limiting.

Flip-chip bonding places solder bumps directly on the die’s contact pads. The die is then flipped and bonded face-down to the substrate, creating shorter interconnects with better electrical characteristics. This is standard for high-performance processors.

Advanced packaging now enables 3D integration using through-silicon vias (TSVs) – vertical electrical connections that pass entirely through silicon dies. Multiple chips can be stacked and interconnected, dramatically reducing the distance signals must travel and enabling heterogeneous integration of different technologies (logic, memory, sensors) in single packages.

A chip on a motherboard showing circuits and connections
A chip on a motherboard showing circuits and connections

Image source: ASML

The Yield Challenge

Despite extraordinary efforts to control contamination, defects are inevitable. A single particle landing during lithography can destroy an entire chip. Modern fabs operate at ISO Class 1 or better – meaning fewer than 10 particles of 0.1 micrometers or larger per cubic meter of air. For comparison, typical outdoor air contains millions of particles per cubic meter.

Yield – the percentage of functional chips per wafer – determines economic viability. At advanced nodes, yields below 80% are often unprofitable. Achieving high yields requires not only clean facilities but also statistical process control across every step, inline inspection and metrology, and continuous feedback loops between design and manufacturing.

The cost is staggering. A new leading-edge fab costs over $20 billion. EUV machines cost $150 million each. A single processed wafer at the most advanced nodes costs thousands of dollars before it is even diced into chips.

Beyond Moore’s Law

The observation that transistor density doubles approximately every two years has driven the industry for six decades. But fundamental limits are approaching. Transistor gates below one nanometer approach atomic dimensions – a silicon atom is about 0.2 nanometers. Quantum tunneling causes electrons to leak through barriers that should block them. Heat dissipation becomes critical as billions of transistors switch billions of times per second in a chip the size of a fingernail.

The industry response is multidimensional. 3D stacking grows upward rather than outward. Chiplets allow different functions to be manufactured on optimal process nodes and assembled together. New materials like silicon-germanium and compound semiconductors offer superior properties for specific applications. And fundamentally different computing architectures – neuromorphic, quantum, optical – may eventually supplement or replace traditional approaches.

Every microchip represents an unprecedented convergence of physics, chemistry, materials science, and precision engineering. The transistors inside it operate on quantum mechanical principles their designers never explicitly learned – they simply engineered structures small enough for quantum effects to dominate. This inadvertent mastery of the quantum realm may be humanity’s most impressive technological achievement, reproduced billions of times each year in facilities that push the boundaries of what is physically possible.


References

  1. ASML. (2023). “6 crucial steps in semiconductor manufacturing.” ASML News. https://www.asml.com/news/stories/2021/semiconductor-manufacturing-process-steps

  2. Wikipedia. “Semiconductor device fabrication.” https://en.wikipedia.org/wiki/Semiconductor_device_fabrication

  3. ASML. “EUV lithography systems.” https://www.asml.com/products/euv-lithography-systems

  4. Synopsys. (2024). “What are Gate-All-Around (GAA) Transistors?” Synopsys Blog. https://www.synopsys.com/blogs/chip-design/what-are-gate-all-around-gaa-transistors.html

  5. Lam Research. (2020). “FinFETs Give Way to Gate-All-Around.” Lam Research Newsroom. https://newsroom.lamresearch.com/FinFETs-Give-Way-to-Gate-All-Around

  6. Wikipedia. “Extreme ultraviolet lithography.” https://en.wikipedia.org/wiki/Extreme_ultraviolet_lithography

  7. HORIBA. “Chemical Mechanical Planarization (CMP).” https://www.horiba.com/int/semiconductor/process/chemical-mechanical-planarization-cmp/

  8. Wikipedia. “Ion implantation.” https://en.wikipedia.org/wiki/Ion_implantation

  9. Wikipedia. “History of the transistor.” https://en.wikipedia.org/wiki/History_of_the_transistor

  10. IEEE Spectrum. “The Status of Moore’s Law: It’s Complicated.” https://spectrum.ieee.org/the-status-of-moores-law-its-complicated