MODULE 4 · LESSON 1
Free — no login requiredSign in to track progress, save quiz attempts and enrol in the full course.
Sign in to track progress / enrolDecoherence and Noise
Module 2 established that a quantum computer works by arranging a precise pattern of cancellation. This lesson is about how easily that pattern is destroyed.
What decoherence is
A quantum computer must be isolated from its environment while it computes. Any uncontrolled interaction, a stray photon, a vibration, a fluctuating magnetic field, a passing cosmic ray, disturbs the phase relationships between amplitudes.
Once the phases are disturbed, the cancellations stop being exact. Wrong answers that should have vanished retain some amplitude. The interference pattern degrades into noise, and the machine returns a random result.
This is decoherence, and the essential point is that it is not a component failure. The hardware has not broken. The information has leaked into the environment, and the environment has effectively measured the system, which as Module 1 explained destroys the state.
Perfect isolation is impossible, so every qubit has a coherence time, a window during which it remains usable. All computation must finish inside it.
Depth matters as much as width
This gives you a second dimension for judging hardware, and it is the one usually omitted.
Width is how many qubits a machine has. It is the number in headlines.
Depth is how many operations can be performed in sequence before decoherence ruins the result.
A processor with many qubits and shallow depth is limited in a way the qubit count entirely conceals. It is like a large workshop where every tool breaks after a few minutes: the floor space is real and it does not let you build anything complicated.
Every gate also introduces a small error of its own. Those errors accumulate. If each two qubit gate has an error rate of about 0.3 percent, which is roughly what good current hardware achieves, then after a few hundred sequential gates the accumulated error makes the output meaningless. Shor's algorithm on a cryptographically relevant number would require billions of sequential operations.
That gap, between hundreds and billions, is the real distance to a cryptographically relevant machine. It is not a small engineering step, and error correction is the only known route across it.
Why the machines look the way they do
The elaborate apparatus follows directly.
Dilution refrigerators. Superconducting processors operate at around 10 to 15 millikelvin, colder than deep space. Not to keep components cool in the ordinary sense but because thermal energy disturbs phases. At room temperature the qubit would decohere immediately.
Magnetic and electromagnetic shielding. Stray fields shift qubit frequencies and corrupt phases.
Vacuum chambers. Trapped ion and neutral atom machines hold their qubits in ultra high vacuum, since collision with a stray gas molecule ends coherence.
Vibration isolation. Mechanical noise couples into the system through the apparatus.
The familiar photograph of a gold coloured chandelier is mostly the refrigerator and its wiring. The actual processor is a small chip at the very bottom, typically a centimetre or two across. Most of what you see is the effort required to keep the world away from it.
The NISQ era
The current period has a name: Noisy Intermediate-Scale Quantum, coined by John Preskill. Machines with tens to hundreds of qubits, too noisy for error correction to be applied at full scale, and too small for the famous algorithms.
Two consequences shape what is realistic today.
The famous algorithms cannot run at useful scale. Shor's algorithm has been demonstrated on tiny numbers, which proves the principle and nothing about capability. Scaling it requires error correction.
Work concentrates on hybrid algorithms. A classical computer runs an optimisation loop and calls a short quantum circuit repeatedly as a subroutine. Short circuits are chosen because they finish inside the coherence time. This is where most current commercial experimentation sits, and results are mixed: many hybrid approaches have been matched or beaten by purely classical methods once someone tried properly.
The honest position is that NISQ machines are extremely valuable as research instruments and have not yet demonstrated commercial advantage on a real problem. Both halves of that sentence matter, and most commentary drops one or the other.
Newcomers reasonably assume the goal is maximising coherence time, and that comparing machines means comparing that number. The reality is a set of tradeoffs, which is why different hardware approaches make opposite choices, as Module 5 describes.
What matters is not coherence time alone but how many operations fit inside it. That is coherence time divided by gate time, sometimes called the number of operations before decoherence.
This creates a genuine tension. Qubits that couple strongly to their control apparatus can be operated quickly, which is good, but strong coupling also means stronger coupling to noise, which shortens coherence. Qubits well isolated from the environment hold their state far longer and are correspondingly slower to manipulate.
Superconducting qubits sit at the fast and short end: gate times measured in tens of nanoseconds and coherence measured in tens to hundreds of microseconds. Trapped ions sit at the slow and long end: coherence times that can reach seconds, with gates thousands of times slower. The ratio, which is what actually matters, ends up broadly comparable, with each approach making different sacrifices.
Two further factors complicate any simple comparison.
Connectivity. Can any qubit interact directly with any other, or only with its neighbours? Limited connectivity means extra operations to shuffle information around, consuming depth you would rather spend on the algorithm. Trapped ions typically offer full connectivity; superconducting chips typically do not.
Gate fidelity. A fast gate that is slightly wrong may be worse than a slow accurate one, because errors accumulate through the circuit.
So a single number never characterises a quantum computer, and any comparison resting on qubit count alone is uninformative. Serious benchmarking attempts to combine these factors, though no single measure is universally accepted, which is itself a sign of how early the field remains.
A company announces a processor with 1,000 qubits. What additional information do you most need before judging its capability?
Decoherence
Click to flipEnvironmental interaction disturbing phase relationships so that cancellations fail and the interference pattern becomes noise. Not a component failure.
Click to flip backDecoherence is environmental interaction disturbing the phases on which cancellation depends, so the machine returns noise rather than an answer. It is not component failure but information leaking into the environment. This makes depth as important as width: gate errors around 0.3 percent limit useful circuits to a few hundred sequential operations, while cryptographically relevant Shor would need billions. The extreme refrigeration and shielding exist to keep the world from disturbing phases. We are in the NISQ era, where machines are valuable research instruments that have not yet shown commercial advantage.