Model NITI Aayog, 24-27 Dec, online. Write an original white paper.

How Is a Microchip Made? | How It Works

12 min read

On this page

What is the short answer?

A microchip is a small piece of silicon carrying a whole electronic circuit, made mostly of transistors: tiny switches that let a current through or block it.

Factories called fabs make thin discs called wafers from silicon purified out of sand, build circuits on them layer by layer, using light to print each layer's pattern, then test, cut and package the chips.

Definition: An integrated circuit (IC), or microchip, is a circuit whose parts, mainly transistors, and their connecting wires are built into one piece of semiconductor, usually silicon. Each chip cut from a wafer is a die.

The Computer History Museum dates the Intel 4004 microprocessor to 1971: 2,300 transistors on a die of 12 square millimetres. In 1965, Gordon Moore projected that the components on a chip would double every year; in 1975 he slowed this to every two years, an observation known as Moore's law.

How does sand become a silicon wafer?

Sand and quartz are mostly silicon dioxide. A textbook by Bill Wilson of Rice University sets out the route:

  • Reduction: quartzite and carbon react to give metallurgical grade silicon, about 98% pure.
  • Purification: this silicon is made into trichlorosilane, which is distilled and reacted with hydrogen. The resulting electronic grade silicon must be better than 99.999999% pure.
  • Crystal growth: in the Czochralski method, a seed crystal on a rod is dipped into molten silicon and slowly pulled up, growing one continuous crystal, the boule. Rod and crucible turn in opposite directions, inside argon that keeps out oxygen.
  • Slicing: the boule is ground to a standard diameter, sliced into wafers, then etched and polished.

An industry executive quoted by NIST, the US standards agency, in August 2018 said each wafer 300 millimetres (mm) across can cost up to US$400.

How are the layers of a chip built up?

The circuit is built at the wafer's surface in repeated rounds:

  • Oxidation: heating the wafer in dry oxygen or steam grows a thin insulating film of silicon dioxide.
  • Deposition: other thin films, such as silicon nitride or polysilicon, are laid on top.
  • Photolithography: light prints a pattern that marks where material should stay.
  • Etching: material is removed where the pattern leaves it bare, in one of Wilson's examples by hydrofluoric acid.
  • Ion implantation: to dope silicon, an implanter ionises a dopant gas and accelerates the ions through tens to hundreds of kilovolts into the wafer. Heating, called annealing, then repairs the crystal.
  • Metal wiring: in sputtering, argon ions knock metal atoms off a target onto the wafer; lithography and etching then leave only the wires joining the transistors.

NIST reported in August 2018 that some chips need as many as 3,000 process steps. An industry executive it quoted put a wafer's time in the fab at 1 or 2 months.

How does light print the patterns, and what do node names mean?

In photolithography, a photomask, or reticle, holds one layer's pattern. The wafer is coated with light-sensitive photoresist, and a lens projects the mask's image onto it.

Light weakens a positive resist where it lands, and a solvent washes that part away. This development leaves a stencil for etching or implanting.

Shorter wavelengths print finer detail. IEEE Spectrum, an engineering magazine, explained in March 2025 that lasers brought deep ultraviolet (DUV) light of 248 and 193 nanometres (nm).

Almost all lens materials absorb light below about 150 nm, so extreme ultraviolet (EUV) lithography uses mirrors and light of about 13.5 nm, made by firing laser pulses at about 50,000 tin droplets a second. Only ASML, a company based in the Netherlands, makes EUV systems, the magazine reported in January 2025.

A process node name such as "3 nm" labels a generation, not a size. IEEE Spectrum reported in July 2020 that node names began drifting from real features in the mid-1990s, and that most critical features of a "7-nm" transistor are considerably larger than 7 nm.

Why are chips made in a cleanroom?

At this scale, one speck of dust can ruin a circuit, and Wilson's textbook says every wafer ends up with some "killer" defects. Fabs therefore work in cleanrooms, where filtered air removes particles.

Cleanliness is graded under the ISO 14644-1 standard. The University of Kentucky's nanoscale centre explains that the class number is the power of ten of particles 0.1 micrometre (µm) or larger allowed per cubic metre of air. An ISO class 5 room allows at most 10⁵, or 100,000.

Lower classes are cleaner. City air, with up to 35,000,000 particles of 0.5 µm or larger per cubic metre, matches ISO class 9. An older US scale counted per cubic foot; its "class 100" roughly matches ISO class 5.

How does a finished wafer become packaged chips?

The Computer History Museum splits chip making into a front end, which processes wafers, and a back end of packaging, assembly and test:

  • Wafer test: probe machines test each die while still on the wafer.
  • Dicing: the wafer is cut into dies. The Semiconductor Industry Association (SIA), a US industry body, says faulty dies are discarded.
  • Packaging: each working die is sealed in a package that protects it and connects it to a circuit board.
  • Final test: packaged chips are tested again before shipping.

Yield is the share of dies that work. In an illustrative textbook example, Wilson finds 15 good chips out of 21, about 71%. Shrinking the circuit by 30% gives 40 good out of 46, or 87%, as each smaller die is less likely to hold a killer defect.

Where are the steps carried out, and which public programmes support them?

Several governments fund chip making at home:

  • United States. The CHIPS and Science Act became law on 9 August 2022. NIST says it provided $52.7 billion, of which its CHIPS Program Office administers $39 billion in incentives.
  • European Union. The European Chips Act, Regulation (EU) 2023/1781, entered into force on 21 September 2023.
  • The Netherlands. On 28 March 2024, the Dutch government announced a €2.51 billion package, shared with regional authorities, for education, knowledge and infrastructure to strengthen the chip industry around Brainport Eindhoven.
  • South Korea. Korea.net, the government's official website, reported on 16 January 2024 a plan for a chip cluster in Gyeonggi-do Province by 2047, adding 16 fabs to the 21 already there.
  • Japan. Rapidus, a company aiming to make 2 nm logic chips at full scale by 2027, said on 5 June 2026 that the Information-Technology Promotion Agency, under the Ministry of Economy, Trade and Industry, had invested a further 150 billion yen.
  • India. The India Semiconductor Mission's Semicon India Programme, launched in 2021 with ₹76,000 crore, supports fabs, display fabs and units that assemble, test and package chips; a second phase of ₹1,27,500 crore was approved in July 2026.

What do supporters and critics argue?

  • Public money for fabs at home. Supporters include the US Department of Commerce, a government department: NIST's CHIPS page lists aims such as a resilient supply chain, national security, jobs and a skilled workforce. The European Commission, the EU's executive, says its Chips Act has helped mobilise over €80 billion of announced or planned investment. Critics include Derek Scissors of the American Enterprise Institute, a think tank, who argued in August 2025 that such policy makes the state decide what deserves support, and that waste does more harm when a government borrows heavily.
  • Water and energy. Those who raise concerns cite Eco-Business, a news site, which reported in February 2026 that one fab can use around 14 billion litres of ultrapure water a year. It cited a case study by the Taskforce on Nature-related Financial Disclosures (TNFD), an international reporting initiative: 40% of existing fabs, and over 40% of new fabs announced since 2021, are in basins projected to face high or extremely high water stress by 2030. Greenpeace East Asia, an environmental campaign group, projected in 2023 that chip making could use 237 terawatt-hours (TWh) of electricity in 2030, about Australia's use in 2021. Chip makers point to recycling: TSMC, one of them, says about 65% of the water at its Arizona site will come from its own recycling at startup, aiming for 90% once a reclamation plant due in 2028 is built.
  • The end of Moore's law? Those who expect shrinking to stall include researchers at MIT, a university, who said transistors are nearing their size limit, so speed-ups must come from better software, new algorithms and leaner hardware, MIT News reported on 5 June 2020. Those who expect it to go on include imec, a chip research centre in Belgium, whose February 2023 roadmap takes node names from 7 nm to 0.2 nm by 2036.

How does this work with real numbers?

Both examples use illustrative inputs with sourced facts: the 300 mm wafer, the 4004's count and Moore's 1975 rate.

How many chips fit on one 300 mm wafer?

  • Wafer area: π × (300 ÷ 2)² = π × 150² ≈ 70,686 mm².
  • Simple division: for an illustrative die of 100 mm², 70,686 ÷ 100 ≈ 707.
  • Edge loss: square dies cannot fill the round rim, so a common estimate subtracts π × d ÷ √(2A), with diameter d and die area A: π × 300 ÷ √200 ≈ 942.5 ÷ 14.14 ≈ 67.
  • Dies per wafer: 707 − 67 ≈ 640. This estimate ignores the scribe lines between dies.
  • Good dies: at an illustrative yield of 80%, 640 × 0.8 = 512.

How fast does doubling every two years add up?

Start from the 4004's 2,300 transistors and assume, as an illustrative projection, a doubling every two years:

  • After 10 years: 5 doublings, and 2⁵ = 32, so 2,300 × 32 = 73,600 transistors.
  • After 20 years: 10 doublings, and 2¹⁰ = 1,024, so 2,300 × 1,024 = 2,355,200, about 2.4 million.

Where can the student go deeper in World 101?

This page follows silicon from sand to a sealed and tested chip. The World 101 subject Semiconductors & the Chip War takes in the wider world of chips: who designs and makes them, where the factories are, and why so many industries and governments depend on them.

Three related subjects take these ideas further, each from a different side. Materials Engineering explains how materials such as silicon, metals and insulators are chosen, processed and tested.

Electrical & Electronics Engineering shows how transistors are combined into circuits that compute, store and amplify signals. Critical Minerals and Supply Chains asks where the raw materials for modern technology come from and how they reach the factories that use them.

Sources

  • Computer History Museum: Intel 4004 (1971); Moore's law (1965, revised 1975); front and back ends; read 9 November 2026.
  • B. Wilson, Rice University, Introduction to Physical Electronics (LibreTexts), read 9 November 2026.
  • NIST news, 22 August 2018.
  • IEEE Spectrum: S. K. Moore, 21 July 2020; J. Boyd, 2 January 2025; J. Stewart and others, 5 March 2025.
  • University of Kentucky nanoscale centre, cleanroom page, read 9 November 2026.
  • SIA primer, 25 February 2015; SIA back-end manufacturing page, read 9 November 2026.
  • NIST: Public Law 117-167, 9 August 2022; CHIPS page, updated 29 October 2026.
  • European Commission, Chips Act page (Regulation (EU) 2023/1781), updated 6 November 2026.
  • Government of the Netherlands, 28 March 2024.
  • Korea.net, 16 January 2024 (plan to 2047).
  • Rapidus, 5 June 2026 (2027 target).
  • India Semiconductor Mission (2021 programme; second phase, July 2026), read 9 November 2026.
  • D. Scissors, American Enterprise Institute, 25 August 2025.
  • Eco-Business, 10 February 2026, on a TNFD case study (fabs since 2021; stress by 2030).
  • Greenpeace East Asia, 20 April 2023, corrected 18 May 2023 (2030 projection; Australia, 2021).
  • TSMC, Arizona page (plant expected 2028), read 9 November 2026.
  • MIT News, 5 June 2020; imec, 2 February 2023 (0.2 nm by 2036).

Common errors and misconceptions

  • Misconception: Chips are made almost straight from sand. Correct: The silicon is purified to better than 99.999999% and grown as one crystal before any circuit is built.
  • Misconception: A "3 nm" chip has transistors 3 nm across. Correct: Node names label a generation; critical features of a "7-nm" transistor are considerably larger than 7 nm.
  • Misconception: A chip is carved out in one step. Correct: It is built in many rounds of oxidation, deposition, lithography, etching and implantation, up to 3,000 steps for some chips.
  • Misconception: Every die on a wafer works. Correct: Killer defects ruin some dies; testing removes them, and yield is the share that works.
  • Misconception: Moore's law is a law of physics. Correct: It was Gordon Moore's projection of 1965, slowed in 1975 to a doubling every two years.

Key takeaways

  • A microchip is an integrated circuit: transistors and their wiring built into one piece of silicon, cut from a wafer as a die.
  • Sand is reduced to silicon, purified to better than 99.999999%, grown into one crystal by the Czochralski method, then sliced into wafers.
  • Chips are built in layers through oxidation, deposition, photolithography, etching, ion implantation and metal wiring, sometimes over thousands of steps.
  • Shorter wavelengths print finer patterns: deep ultraviolet uses 248 and 193 nm; extreme ultraviolet uses about 13.5 nm and mirrors.
  • Cleanrooms are graded under ISO 14644-1; finished wafers are tested, diced and packaged, and yield is the share of dies that work.
  • Several governments fund chip making, while subsidies, water and energy use and the limits of shrinking are debated.

Test yourself

What does the Czochralski method produce?

A single silicon crystal, the boule, grown by slowly pulling a turning seed crystal out of molten silicon.

Why does EUV lithography use mirrors instead of lenses?

Almost all lens materials absorb light below about 150 nm, and EUV's wavelength is about 13.5 nm.

What does ISO class 5 mean?

Each cubic metre of air may hold at most 10⁵, or 100,000, particles of 0.1 µm or larger.

In the wafer example, how was 640 found?

70,686 mm² ÷ 100 mm² gives about 707; subtracting an edge loss of π × 300 ÷ √200 ≈ 67 leaves about 640.

If transistor counts double every two years, how much larger are they after 20 years?

Ten doublings give 2¹⁰ = 1,024 times as many; from 2,300, that is about 2.4 million, as an illustrative projection.

Why does a smaller die usually give a higher yield?

Killer defects are scattered across the wafer, so a smaller die is less likely to contain one.

Go deeper in World 101: Semiconductors & the Chip War

Write about it

Turn a question from this chapter into a paper.

The Writing Studio has a structure guide, a rubric and citation tools for 15 formats, from essays to research papers. A free account keeps up to five drafts. Learnacy Essential adds feedback from an AI writing coach and submission to The OYI Review.

Publication depends on review and is not guaranteed.