Quantum Annealing Quantum Computers

Quantum annealing is an analog computational paradigm designed specifically to find low-energy states of Ising spin models — which in turn can represent a wide class of combinatorial optimization problems. Rather than implementing discrete quantum gate operations, an annealer prepares the system in the known ground state of a simple transverse field Hamiltonian (all qubits in superposition) and then slowly evolves the system by reducing the transverse field while simultaneously increasing the strength of the problem Hamiltonian. If this evolution is slow enough (the adiabatic theorem), the system remains in the ground state throughout and ends in the ground state of the problem Hamiltonian, which encodes the solution.

D-Wave's annealers use superconducting flux qubits as the physical qubits. A flux qubit encodes |0⟩ and |1⟩ as two directions of persistent current flowing around a superconducting loop. Programmable Josephson junction couplers between qubits allow the coupling strengths J_ij and local biases h_i to be set, encoding any problem that can be written as a quadratic unconstrained binary optimization (QUBO) or equivalently the Ising model E = Σ h_i σ_i + Σ J_ij σ_i σ_j.

In practice, D-Wave systems do not run full adiabatic anneals. The anneal time is tunable (1–2000 µs), and at finite anneal speed the system does not always find the true ground state — it finds low-energy states that are typically good but not provably optimal solutions. Hybrid classical-quantum solvers (via Ocean SDK) decompose large problems, run subproblems on the QPU, and recombine results, significantly extending the class of tractable problems.

D-Wave has scaled from 128 qubits (2011) to 5,000+ qubits (Advantage, 2020) to 7,000+ qubits (Advantage2, targeting Zephyr topology). The Zephyr graph provides ~20-way connectivity, reducing embedding overhead. Future scaling will be limited by the same physical constraints as superconducting: wiring density, cooling capacity, and crosstalk. D-Wave has also expanded to hybrid computing with their Leap platform, which can handle problems with millions of variables by decomposing them into QPU-sized chunks. The long-term question is whether quantum annealing ever provides a practical speedup over classical heuristics like simulated annealing, branch-and-bound, or tensor network methods on problems that matter to enterprise customers.

Available QPUs

Workload Suitability