What Is a QPU? Quantum Processors Explained

Qubits, superposition and interference

A classical bit is either 0 or 1. A qubit is a two-level quantum system, such as the energy state of a tiny circuit or the spin of an electron. Its state is described by two numbers called amplitudes.

This is what people mean by superposition. The qubit is described by a combination of 0 and 1 until it is measured. When you measure it, you get a single result, 0 or 1, with probabilities set by the amplitudes. You do not get both answers.

The common claim that a quantum computer tries every answer at once is misleading. Quantum algorithms work by interference. They arrange the amplitudes so that paths leading to wrong answers cancel out and paths leading to the right answer reinforce. Only a few problems are known to allow this.

Entanglement is the third ingredient. It links the states of several qubits so that they cannot be described one at a time. Together with superposition and interference, it lets a QPU represent and manipulate certain structures that a classical computer would need exponentially more memory to store.

How a QPU runs a program

A quantum program is a circuit, a sequence of operations called gates applied to qubits, followed by a measurement. Because measurement gives a random-looking result, the circuit is usually run many times and the results are tallied.

A QPU never works alone. A classical computer compiles the circuit, sends control pulses, reads the results and often adjusts the next run. Control electronics, and for some designs refrigerators or lasers, make up most of the physical machine.

Not every quantum machine is gate-based. Quantum annealers, such as those from D-Wave, are built for optimisation problems and work differently.

The main kinds of QPU

Superconducting qubits are tiny electrical circuits made from superconducting metal. They are cooled to roughly ten to twenty millikelvin, close to absolute zero, in a dilution refrigerator. Gates are fast, usually tens to hundreds of nanoseconds, and the chips can be made with semiconductor-style methods. Each qubit connects only to a few nearby qubits. IBM, Google, Rigetti and IQM use this approach.

Trapped-ion qubits are individual charged atoms held in place by electric fields in a vacuum. Their internal energy levels act as the qubit. Gates use lasers or microwaves. Trapped ions have some of the lowest error rates reported, and in small systems any qubit can interact with any other. Gates are slower than in superconducting chips, and scaling to many ions is an engineering challenge. IonQ and Quantinuum are leading examples.

Neutral-atom qubits are uncharged atoms held by tightly focused laser beams called optical tweezers. Arrays of hundreds to several thousand atoms have been demonstrated, though far fewer are used in computations, and the atoms can be rearranged during a run. Entangling gates use highly excited Rydberg states. QuEra, Pasqal and Atom Computing build these machines.

Photonic qubits use particles of light. Photons travel well and do not need extreme cooling themselves, although the best single-photon detectors usually do. The main difficulty is loss, because a lost photon is a lost qubit. PsiQuantum and Xanadu pursue this route.

Silicon spin qubits store information in the spin of an electron or atomic nucleus in silicon. They are very small and could in principle be made in existing chip factories. They are earlier in development than the others in terms of qubit numbers. Silicon Quantum Computing is one company working on them.

Why qubit counts mislead

Headline qubit counts are easy to compare, and they are a poor measure of usefulness. A machine with many noisy qubits may run fewer reliable operations than a smaller machine with low error rates.

Better questions are these. How often does a two-qubit gate fail? How long do qubits hold their state? How many qubits can interact directly? How fast is a gate? How many operations can run before the result is lost in noise?

Error rates have improved, with the best devices now reporting roughly one error in a thousand two-qubit gates or better. Long, useful algorithms need far lower error rates, roughly one in a billion or better for each logical operation. Error correction is how machines are meant to close that gap.

Error correction is the central problem

Quantum error correction spreads one logical qubit across many physical qubits. The logical qubit is the reliable one that an algorithm sees. The physical qubits are the noisy hardware. The system constantly checks for errors and fixes them without reading out the stored value.

Error correction only helps if physical error rates are below a threshold. Below it, adding more physical qubits makes the logical qubit better. In 2024 Google reported an experiment in which the logical error rate fell as the code was made larger, which is the behaviour the theory requires. It was a single logical qubit held in memory, not a full computation.

The overhead is large. Estimates for running algorithms that threaten today's public-key encryption range from under a million to tens of millions of physical qubits, depending on the assumptions. Current devices have far fewer.

What QPUs are not faster at

Quantum computers are not faster at everything. They are not a drop-in replacement for CPUs or GPUs. Browsing, databases, video and most machine learning gain nothing from a QPU.

The best-known quantum speedups are for simulating quantum systems such as molecules, for factoring integers using Shor's algorithm, and for unstructured search using Grover's algorithm, which gives only a quadratic gain. Claims of quantum advantage on other tasks have often been narrowed when classical researchers found better algorithms.

Treat any announcement with three checks. Was a useful problem solved, or a benchmark built for the machine? Was the result compared with the best classical method? Did the work use error-corrected qubits, or raw noisy ones?