How quantum gates work, in plain English
On this page
The algorithm article showed what quantum programs look like: a sequence of steps called gates. This piece looks at the gates themselves. What does each one actually do, why are they so different from the logic gates in an ordinary chip, and what physically happens inside the machine when one runs? As before, no maths is needed beyond a few small numbers.
Ordinary gates throw information away
Every ordinary computer is built from logic gates: tiny circuits that take one or two bits in and give one bit out. Three do most of the work:
| Gate | What it does | Can you work out the inputs from the output? |
|---|---|---|
| NOT | Turns 0 into 1 and 1 into 0 | Yes: just flip it back |
| AND | Gives 1 only if both inputs are 1 | No: an output of 0 could come from 00, 01 or 10 |
| OR | Gives 1 if either input is 1 | No: an output of 1 could come from 01, 10 or 11 |
That one-way quality is normal and useful in an ordinary chip. Quantum gates can't work that way. A qubit's amplitudes have to be carried through every step intact, because the whole point is to let them interfere later. So every quantum gate is reversible: for each gate there's another that undoes it exactly, and no information is ever thrown away. The only one-way step in a quantum program is the measurement at the end, which is why it isn't counted as a gate.
One-qubit gates are turns of the globe
In Part 2 we pictured a qubit as a point on a globe: 0 at the North Pole, 1 at the South Pole, even blends around the equator, and phase as longitude. Every one-qubit gate is simply a turn of that globe. Turning a globe never loses anything, which is exactly why the gates are reversible: turn it back and you're where you started.
- X, the flip. Turns the globe so the poles swap places: 0 becomes 1 and 1 becomes 0. It's the quantum version of NOT. Applied to a blend, it swaps the two amplitudes.
- Z, the sign flip. Turns the globe half way around its north–south axis. The poles don't move, so a plain 0 or 1 is unchanged and the odds of any state stay the same. What changes is the phase: the amplitude on 1 becomes negative. It looks like it does nothing, but that sign is what lets amplitudes cancel later.
- H, the blend. Moves a pole down to the equator, turning a plain 0 or 1 into an even 50/50 blend. Applied again, it moves the blend back to the pole. That's the "apply it twice and you're back to 0" trick from Part 2's worked example.
- Smaller turns. Real machines can also turn the globe by any angle, not just half turns. Those rotations let an algorithm set any weighting it needs, like the slider in the home-page demo.
Two-qubit gates: where qubits interact
One-qubit gates can only ever change qubits one at a time. To do anything a classical computer couldn't easily copy, qubits have to act on each other. The workhorse is the controlled flip, often called CNOT. It has a control qubit and a target qubit, and flips the target only where the control is 1:
| Before (control, target) | After | What happened |
|---|---|---|
| 0, 0 | 0, 0 | Control is 0: nothing |
| 0, 1 | 0, 1 | Control is 0: nothing |
| 1, 0 | 1, 1 | Control is 1: target flipped |
| 1, 1 | 1, 0 | Control is 1: target flipped |
On plain bits that's ordinary logic. The interesting case is when the control is a blend. Then the gate does both rows at once, each with its own amplitude, and the two qubits end up tied together. That's how entanglement is made, step by step, in Part 3.
A close cousin, the controlled sign flip (CZ), makes an amplitude negative only where both qubits are 1. It's how Grover's search marks the right answer in How a quantum algorithm works.
A small set builds everything
There are endless possible gates, but you don't need them all. A handful is enough to build any quantum calculation, to whatever accuracy is needed, in the same way a small box of Lego bricks can build almost anything. One well-known set is just H, the controlled flip and a small turn called T (an eighth of a full turn around the north–south axis).
That has a practical consequence. Each quantum machine offers its own small set of native gates, the ones its hardware does directly. Software called a compiler translates a program written in friendly gates into the machine's native ones, much as ordinary compilers turn code into a processor's instructions. It's one reason the same program can run on very different hardware.
What a gate physically is
An ordinary logic gate is a fixed piece of circuitry that signals flow through. A quantum gate isn't a part at all. It's an action: a precisely shaped pulse of energy aimed at the qubits.
- Superconducting qubits, used by IBM and Google, are turned by microwave pulses sent down wires into the refrigerator. A one-qubit gate takes a few tens of billionths of a second.
- Trapped ions are turned by laser pulses, or in some designs by microwaves. Their gates are slower, typically microseconds or more, but the qubits hold their state much longer.
- Two-qubit gates need the qubits to interact for a short, carefully controlled moment, through a shared circuit element for superconducting chips or the ions' shared motion in a trap.
The size, shape and timing of each pulse decide how far the globe turns. A pulse that's slightly too long or too strong over- or under-turns the qubit, and that's where errors come from.
Why gates go wrong
Real gates are never perfect. A pulse that's a little off, a stray signal or a qubit losing its blend part-way through all leave the state slightly wrong. One-qubit gates on the best hardware are now right well over 99.9% of the time. Two-qubit gates are harder, because two delicate qubits have to interact precisely while staying isolated from everything else; the best reach around 99.9%, and many machines are below that.
That's why the two-qubit error rate is usually the number to look for in a company's announcement, and why long algorithms need the error correction described in Why qubits are so fragile. A useful run of Shor's algorithm needs billions of gates; at one error per thousand, the answer would be lost long before the end.
Common questions
Is a quantum gate like a transistor? No. A transistor is a physical switch built into the chip. A quantum gate is an operation, a pulse applied to qubits that stay where they are. The same qubit can receive thousands of different gates during one program.
Why not just copy the qubit before a risky gate, as a backup? Quantum physics forbids making an exact copy of an unknown qubit state, the no-cloning rule. That's why error correction has to work differently, by spreading information across many qubits.
Is measurement a gate? Usually not. Every gate can be undone; measurement can't, because it ends the blend and gives one answer. Programs treat it as a separate, final step.
- Quantum gates are reversible turns of the qubit globe; nothing is thrown away until you measure.
- Two-qubit gates such as the controlled flip are where qubits interact and entanglement is made.
- A gate is a precisely timed pulse, and small errors in each one are why error correction matters.
Sources and further reading
Tags show what kind of source each one is. A standard or government guidance is an official document; a peer-reviewed paper has been checked by other experts; a preprint has not been peer-reviewed yet; an experiment reports a real-world demonstration; a company announcement is the company's own account. Dates and figures were checked against these sources on 14 October 2026. Spotted an error? Email hello@plainquantum.com and it will be corrected, with a note.
- BookQuantum Computation and Quantum Information (10th anniversary edition)Michael Nielsen and Isaac Chuang, Cambridge University Press, 2010
- Peer-reviewed paperElementary gates for quantum computationBarenco et al., Physical Review A, 1995
- ExperimentQuantum error correction below the surface code thresholdGoogle Quantum AI and collaborators, Nature, 2024
- Company announcementQuantinuum announces commercial launch of Helios (48 error-corrected logical qubits; 94-qubit entangled state)Quantinuum press release, 2025