Home› Transistor Size Evolution
77 Years of Miniaturization

From Finger-Sized
to Smaller Than DNA

By EcrioniX · Updated Jun 6, 2026

In 1947, a transistor was the size of your palm. Today, 100 billion of them fit in a chip smaller than your fingernail. This is the most extraordinary engineering achievement in human history — and it's not over yet.

1 cm
1947 First Transistor
2 nm
2024 TSMC N2
5,000,000×
Size Reduction
100B+
Per Chip Today
Scale Comparison
How Small Is 2 Nanometers?
A nanometer is one-billionth of a meter. To put that in perspective — if a marble were 1nm, the Earth would be 1 meter across. Here's how today's transistors compare to things you know.
LOGARITHMIC SIZE SCALE 100mm 1mm 10μm 1μm 100nm 10nm 1nm 0.1nm 💈 Human Hair ~70,000 nm 🩸 Red Blood Cell ~7,000 nm 🦠 Bacteria ~1,000 nm 🦠 Virus (COVID) ~100 nm 🧬 DNA Width ~2.5 nm ⚡ TSMC 2nm (2024) 2 nm ≈ Same as DNA width 💧 Water Molecule 0.28 nm ⚛️ Silicon Atom 0.22 nm Macroscopic Microscopic Biological Transistor Zone Atomic
💈
Human Hair
70,000 nm
35,000× wider than a 2nm transistor. You could stack 35,000 transistors across a single strand.
🩸
Red Blood Cell
7,000 nm
Your blood cells are 3,500× larger than a modern transistor gate length.
🦠
E. coli Bacteria
~1,000 nm
500× bigger than a 2nm transistor. Once, a single transistor was the size of a bacterium. Now 500 fit inside one.
🦠
COVID-19 Virus
~120 nm
60× bigger than today's transistors. The thing that shut down the world is 60× larger than TSMC's gate length.
🧬
DNA Double Helix
2.5 nm wide
TSMC's 2nm transistor is NOW the same width as a DNA strand. We are engineering at the scale of life itself.
💧
Water Molecule
0.28 nm
A water molecule is smaller than a transistor — for now. The next frontier pushes toward this boundary.
⚛️
Silicon Atom
0.22 nm
The absolute physical wall for silicon. You cannot make a silicon transistor smaller than a silicon atom. Period.
⚡
TSMC 2nm (2024)
2 nm
Apple M4. ~100 billion transistors per chip. The most complex object humans have ever manufactured.
The Journey
77 Years of Shrinking
Every decade, engineers told us we'd hit the wall. Every decade, they found a way through.
1947
1947 — Bell Labs
The First Transistor
~10 mm
Bardeen, Brattain, and Shockley built the first transistor using two gold contacts pressed against a germanium crystal. It sat on a base the size of your palm. Nobody knew this device would reshape civilisation. They won the Nobel Prize in 1956.
Germanium · Point-contact · Nobel Prize 1956
1971
1971 — Intel
Intel 4004 — First Microprocessor
10,000 nm
Intel's 4004 packed 2,300 transistors at 10 microns. It had the same computing power as the room-filling ENIAC computer of 1945 — but fit on a thumbnail. The computing era began. A calculator chip.
2,300 transistors · 108 KHz · $200 in 1971
1989
1989 — Intel
Intel 486 — The PC Revolution
1,000 nm
At 1 micron (1,000nm), the 486 had 1.2 million transistors — 500× more than the 4004. Home computing exploded. Transistors crossed the 1-micron barrier — smaller than the wavelength of visible light. A milestone.
1.2M transistors · 25–100 MHz · Windows 3.1 era
2000
2000 — Intel
Pentium 4 — The GHz Wars
180 nm
The Pentium 4 ran at 1.5 GHz and generated so much heat it needed a dedicated fan. The race for clock speed hit a thermal wall — you can't keep adding GHz when the chip melts. The industry pivoted to multicore. A turning point.
42M transistors · 1.5 GHz → 3.8 GHz · Thermal wall
2012
2012 — Intel
Ivy Bridge — The FinFET Revolution
22 nm
Intel's FinFET (Fin Field-Effect Transistor) was a 3D gate — the gate wrapped around the channel like a fin. This was the breakthrough that kept Moore's Law alive when flat planar transistors couldn't go smaller. A radical reinvention of the transistor shape.
1.4B transistors · 3D FinFET · Game-changer geometry
2020
2020 — TSMC / Apple
Apple M1 — The Silicon Revolution
5 nm
16 billion transistors. Apple ditched Intel and designed their own ARM-based chips. The M1 destroyed decade-old x86 chips in performance-per-watt. The world realised the PC era was over. At 5nm, transistors were 10× smaller than the COVID-19 virus.
16B transistors · ARM64 · Performance-per-watt dominance
2023
2023 — TSMC / Apple
Apple M3 — 3nm GAA Preview
3 nm
35 billion transistors on a single die. At 3nm, gate lengths approach the width of just 15 silicon atoms stacked. Samsung introduced GAA (Gate-All-Around) nanosheet transistors, where the gate wraps around all four sides — the next evolution after FinFET.
35B transistors · TSMC N3 · 15 silicon atoms gate
2024
2024–2025 — TSMC / Apple / Intel
The 2nm Era Begins
2 nm
TSMC N2 process. Apple M4. ~100 billion potential transistors per chip. The gate length is now the same as the width of a DNA double helix. You are no longer making silicon chips — you are sculpting matter at the scale of molecules. Intel's 18A and TSMC N2 are both racing to volume production.
100B transistors · DNA scale · GAA Nanosheets · TSMC N2
The Hard Limit
Why We Can't Shrink Silicon Forever
Physics is not negotiable. As transistors approach atomic scales, the rules of the universe itself push back.

⚡ Quantum Tunneling

Below ~3nm, electrons no longer stay where you put them. They quantum-tunnel through the gate oxide barrier — the transistor "leaks" and can't be turned fully off. It's as if your light switch randomly turns itself on. You cannot fix this with better engineering. It's fundamental quantum mechanics.

⚛️ The Silicon Lattice Wall

Silicon's crystal lattice spacing is 0.54nm. You cannot etch a pattern smaller than the atoms that form the material. The theoretical absolute minimum for a silicon transistor is somewhere around 0.5nm — roughly 2–3 silicon atoms. We are approaching this in 10 years.

🔥 Heat — The Invisible Enemy

100 billion transistors switching at 3 GHz in a fingernail-sized area generates enormous heat per square millimetre. Cooling at this density is becoming as hard as the transistor physics itself. Apple's chips are power-sipping ARM precisely because of this thermal reality.

💰 The Cost Cliff

A 2nm fab costs $20–30 billion to build. Only TSMC, Samsung, and Intel can afford it. The EUV lithography machine (from ASML) needed to print 2nm features costs $350 million — each. Economics may stop progress before physics does.

Moore's Law Is Dying — But It's Not Dead Yet

Gordon Moore predicted transistor counts would double every two years. From 1971 to 2015, this held with stunning accuracy. Since 2015, it has slowed — doubles now take 3–4 years and cost exponentially more. The era of "free" performance improvements every 2 years is over. But the industry is not surrendering. They're just changing the rules.

"The number of transistors on a chip will double approximately every two years."
— Gordon Moore, Intel Co-Founder, 1965. He was right for 50 years.
What Comes Next
Beyond Silicon — The Next Frontiers
When silicon hits its limit, engineers don't stop. They change the material, the physics, and sometimes the entire concept of what a "transistor" means.
Near Future · 2025–2028
1nm — Carbon Nanotube FETs
~1 nm gate
IBM demonstrated a carbon nanotube transistor with a 1nm gate in 2016. Carbon nanotubes are rolled sheets of graphene — single atom thick, conducting, and immune to the tunneling problems that plague silicon. Intel and IBM are both developing CNT chips for production. These aren't theory — they're in labs right now.
💡 A carbon nanotube is 50,000× thinner than a human hair.
Near Future · 2026–2030
2D Materials — MoS₂ & Graphene
0.7 nm thick
Molybdenum disulfide (MoS₂) is a 2D material — literally one atom thick. MIT researchers built a transistor with a 1nm channel from MoS₂ in 2016. Graphene transistors switch 1,000× faster than silicon. The challenge: manufacturing 2D transistors at billion-unit scale is still unsolved.
💡 MoS₂ is the same material as the lubricant in your car engine — just one atom thick instead of a powder.
Mid Future · 2030–2040
Photonic Computing — Computing with Light
Speed of light
Instead of electrons (which have mass and generate heat), photonic chips use photons — particles of light. Light doesn't heat up wires. It travels at 3×10⁸ m/s. Silicon photonics already exists in optical data links. Intel and TSMC are building photonic interconnects. A full photonic processor that replaces electrical transistors is the next paradigm.
💡 Google's data centres already use photonic interconnects. The transistors inside are next.
Mid Future · 2030+
Quantum Transistors
Single electron
A quantum transistor controls the flow of a single electron using quantum effects. Google, IBM, and Microsoft are building quantum computers that don't use traditional binary transistors at all — they use qubits that can be 0, 1, or both simultaneously (superposition). Not faster classical computing — a completely different kind of computing.
💡 A 1,000-qubit quantum chip could factor numbers that would take classical computers longer than the age of the universe.
Far Future · 2040+
Molecular & Atomic Switches
Single molecule
Scientists have already demonstrated transistors made from a single organic molecule that switches between conducting and non-conducting states. In 2012, IBM spelled "IBM" using individual xenon atoms. Controlling individual atoms as transistors is not impossible — it's just manufacturing at a scale we can barely imagine. DNA computing (using DNA strands as logic gates) is also being actively researched at MIT and Caltech.
💡 A single DNA strand can store 1 exabyte of data — that's 1 billion gigabytes — in the space of a sugar cube.
Speculative · Far Future
Neutrino-Level Computing?
< 0.000001 nm
A neutrino is a subatomic particle with near-zero mass and almost no interaction with normal matter. Trillions pass through your body every second without touching a single atom. Could we theoretically use neutrino beams to carry information? In theory — yes. In practice — almost certainly no. Their near-zero interaction is the exact property that makes them useless for switches. You can't build a gate from something that ignores everything. Neutrino detectors require entire mountains of water. The universe itself is the obstacle.
⚠️ This is the transistor equivalent of asking "can we build a bridge out of music?" Interesting to think about. Not buildable.
Numbers That Break Your Brain
The Scale of Modern Chip Manufacturing
100B+
Transistors on a chip
Apple M4 has ~100 billion. More than 12× the number of stars in the Milky Way.
3 GHz
Switching speed
Each transistor switches 3 billion times per second. 100B transistors × 3B switches = 3×10²⁰ operations/sec.
$350M
Cost of one EUV machine
ASML's EUV lithography machine — the only one that can print 2nm features. Only one company in the world makes it.
5,000,000×
Size reduction since 1947
From 1cm to 2nm. If cars improved at the same rate, a car would now travel at 1% the speed of light for free.
13.5 nm
EUV light wavelength
The light used to print 2nm features is itself 13.5nm. Engineering beyond the wavelength of your own light source.
$30B
Cost of a 2nm fab
TSMC's new Arizona fab. Only 3 companies on Earth can build one. It is now a matter of national security.

The Most Audacious Project in Human History

We started with a germanium crystal in a lab in 1947. We now manufacture structures smaller than DNA — billions of them — on a wafer the size of a pizza, at a defect rate of less than one per trillion. No other industry has improved its product by a factor of 5 million in 77 years. Not aviation. Not medicine. Not energy.

The transistor is not just a switch. It is the reason you can hold the sum of all human knowledge in your pocket. It is the reason we can model climate, decode DNA, and train AI. Every great technology of the 21st century — from smartphones to satellites to the AI reading your text — runs on transistors smaller than the cells in your body.

And the engineers are still not done.

⚡ Try the Transistor Lab → 🔌 Logic Gate Simulator → 🏭 VLSI Design Hub →