Measurement-based (one-way) quantum computing

In short: Measurement-based (one-way) quantum computing replaces a sequence of unitary gates with a large pre-built entangled cluster state that you consume qubit-by-qubit through single-qubit measurements, where each measurement basis is adaptively chosen from earlier outcomes to steer a logical computation to completion.

If you learned quantum computing from the circuit model, you picture a program as a line of gates applied to qubits that live from start to finish. Measurement-based quantum computing (MBQC), also called the one-way quantum computer, turns that picture inside out. You first create a fixed, highly entangled resource called a cluster state, then you run the algorithm by destroying it: measuring qubits one at a time. The irreversibility of measurement is why Raussendorf and Briegel named it "one-way." Nothing is ever un-measured.

Why bother measuring instead of gating?

The appeal is architectural. In the circuit model, every two-qubit gate is a delicate, dynamic interaction you must apply on demand. In MBQC, all the entanglement is created once, up front, using a single type of two-qubit operation (controlled-Z) on a regular lattice. After that, the only operations left are single-qubit measurements and classical feedforward. This matters enormously for platforms where entangling operations are expensive or probabilistic but measurement is cheap and fast. Linear-optical quantum computing is the textbook example: deterministic two-photon gates are nearly impossible, but you can fuse small entangled photon clusters probabilistically offline and then measure deterministically.

Building a cluster state

The recipe is mechanical. Initialise every physical qubit in the |+⟩ = (|0⟩+|1⟩)/√2 state, then apply a controlled-Z (CZ) between every pair of nearest neighbours on your chosen graph. CZ commutes with itself, so the order does not matter and the whole lattice can be entangled in one global step. A 2D cluster state on a square lattice is a universal resource: by choosing which qubits to measure and in what basis, you can emulate any quantum circuit. A 1D chain, by contrast, only gives you single-qubit operations plus teleportation along the wire.

A square-lattice 2D cluster state is universal for quantum computation. Horizontal direction simulates time (gate sequence); vertical direction provides the qubit register. One logical qubit typically needs a "wire" several physical qubits wide.

Adaptive measurements: the actual computation

Here is the heart of MBQC. You measure each qubit in a basis defined by an angle θ in the X-Y plane of the Bloch sphere, i.e. the observable cosθ X + sinθ Y. Measuring at θ=0 (the X basis) propagates a Hadamard-like operation down the wire; a nonzero θ injects a rotation R_z(θ). The catch is randomness: each measurement yields 0 or 1 with equal probability, and a "1" outcome applies an unwanted Pauli byproduct (an X or Z error) to the still-entangled downstream qubits.

Because the byproduct operators do not simply vanish, you compensate by adapting the next measurement angle. If an earlier outcome demands an X byproduct, you flip the sign of a later θ to −θ. This classical feedforward — compute the correction from past bits, then choose the future basis — is what makes the computation deterministic despite random individual outcomes. The pattern of angles and the dependency rules together are called a measurement pattern, and formalising which patterns run deterministically is the job of "flow" and "gflow" in the measurement calculus.

Circuit model vs. one-way model at a glance

The two models are formally equivalent — you can compile any circuit into a measurement pattern and vice versa — but the equivalence is a compilation step, not a free lunch. A single logical CNOT, for instance, maps onto a 15-qubit cluster fragment in the canonical Raussendorf-Briegel construction, so naive translation inflates qubit count. This is why vendor toolchains matter: whether you design in a gate-level framework and let the stack handle lattice layout, or model the photonic hardware directly, the back-end compiler decides your real resource cost.

Fault tolerance folds in naturally: the 3D Raussendorf-Harrington-Goyal topological cluster state encodes a surface code in the lattice geometry itself, with error thresholds around 0.6–1% per operation — comparable to the circuit-based surface code.

Where it shows up in real hardware and tools

PsiQuantum and Xanadu build their photonic roadmaps around fusion-based and continuous-variable cluster states precisely because measurement is the cheap resource in optics. Neutral-atom platforms generate cluster states across hundreds of atoms using Rydberg CZ gates. On the design side, engineers reach for simulation and verification flows much as they would in semiconductor work: just as a VLSI team leans on Synopsys, Cadence, or Siemens EDA to verify a netlist before tape-out, quantum architects use measurement-calculus tooling (such as graphix or PyZX) to verify that a measurement pattern has valid gflow before committing it to expensive hardware time. If you want to go deeper on the design-verification mindset that carries across both worlds, our colleagues at Avecas VLSI design services apply the same verify-before-commit discipline to silicon.

A worked intuition

To teleport-and-rotate a single qubit: take a 2-qubit chain, prepare both in |+⟩, bond them with CZ. Measure the first qubit at angle θ. The second qubit now holds your input state with an R_z(θ) rotation applied, up to a Pauli X correction that you apply classically if the outcome was 1. Chain several of these and you have assembled an arbitrary single-qubit rotation out of nothing but measurements — the clearest demonstration that measurement, not unitary evolution, can drive computation.

For readers coming from the circuit world, revisit how projective readout collapses a state in our primer on quantum measurement, and see how the same entanglement resource powers quantum teleportation — MBQC is, in a precise sense, teleportation used as a computational gate.

Ready to build cluster states and measurement patterns hands-on? Explore the Quantum Computing track in the CourseTron course catalogue and practise compiling circuits into one-way patterns.

Frequently asked questions

Why is it called "one-way"?

Because the computation proceeds only by measurement, and measurement is irreversible. Once a qubit is measured the entanglement that held it is destroyed, so the resource can be consumed in only one direction — unlike unitary circuits, which are reversible.

Is MBQC more or less powerful than the circuit model?

Exactly as powerful. Any quantum circuit can be compiled into an equivalent measurement pattern on a 2D cluster state and vice versa. They differ in hardware convenience and resource accounting, not in computational class.

Why do measurement angles have to adapt?

Each single-qubit measurement gives a random 0/1 outcome, and a "1" injects an unwanted Pauli byproduct on downstream qubits. Choosing later measurement bases based on earlier outcomes (feedforward) cancels these byproducts and makes the overall logical result deterministic.

Which hardware suits MBQC best?

Platforms where entangling operations are costly or probabilistic but measurement is fast and reliable — principally photonic systems (PsiQuantum, Xanadu) and large neutral-atom arrays. Superconducting and trapped-ion machines usually favour the circuit model.