Quantum Hardware

Qubit Implementations

Qubit Technology Types Comparison

A physical qubit requires a quantum two-level system with: (1) long coherence times, (2) high-fidelity initialization, (3) universal gate operations, (4) qubit-specific readout, and (5) scalability (DiVincenzo criteria).

Superconducting Qubits

The most mature platform, used by IBM, Google, and many others.

Physical basis: Nonlinear LC circuits using Josephson junctions (superconductor-insulator-superconductor). The junction's nonlinear inductance creates anharmonic energy levels, allowing the lowest two to serve as |0> and |1>.

Qubit types:

Transmon: Charge qubit operated at large E_J/E_C ratio, reducing charge noise sensitivity. Dominant design (IBM Eagle/Heron, Google Sycamore). Coherence times: T1, T2 ~ 100-500 microseconds.
Fluxonium: Uses a superinductance for larger anharmonicity. Longer T1 (>1 ms demonstrated) but harder to couple.
Xmon: Transmon variant with cross-shaped capacitor for improved coupling.

Gates: Microwave pulses for single-qubit rotations (~20 ns, fidelity >99.9%). Two-qubit gates via capacitive coupling: cross-resonance (CR) for fixed-frequency transmons, tunable coupler for flux-tunable designs. Two-qubit gate fidelities: 99-99.9%.

Readout: Dispersive readout through coupled superconducting resonators. Qubit state shifts resonator frequency; measured via microwave reflection/transmission.

Challenges: Requires millikelvin temperatures (dilution refrigerators, ~15 mK), limited connectivity (typically nearest-neighbor on 2D grid), frequency crowding, TLS defects, cosmic ray impacts.

Trapped Ions

Used by IonQ, Quantinuum (Honeywell), and academic groups.

Physical basis: Individual atomic ions (e.g., Yb-171, Ba-137, Ca-43) confined in electromagnetic traps (Paul traps/linear RF traps). Qubit encoded in hyperfine ground states or optical transitions.

Gates:

Single-qubit: Microwave or Raman laser pulses. Fidelities >99.99%.
Two-qubit: Molmer-Sorensen or light-shift gates mediated by collective motional modes of the ion chain. Fidelities: 99.5-99.9%.
All-to-all connectivity within a single trap zone (any ion pair can interact).

Readout: State-dependent fluorescence. Illuminate with resonant light; one state fluoresces brightly, the other is dark. Fidelities >99.9%.

Advantages: Identical qubits (atoms are fundamental), long coherence times (seconds to minutes), all-to-all connectivity, highest gate fidelities achieved.

Challenges: Slow gate speeds (~100 microseconds for two-qubit gates), scaling beyond ~30-50 ions per trap (motional mode crowding), QCCD architecture requires ion shuttling between trap zones.

Photonic Qubits

Used by PsiQuantum, Xanadu, and others.

Encodings:

Dual-rail: Qubit as single photon in one of two modes (path, polarization)
Continuous-variable: Squeezed states and homodyne detection (Xanadu's GBS)
Time-bin: Early vs late arrival time

Gates: Linear optical elements (beam splitters, phase shifters) perform single-qubit gates deterministically. Two-qubit gates require nonlinear interaction or measurement-based approaches (KLM scheme: nondeterministic CNOT using ancilla photons and postselection).

Advantages: Room temperature operation, natural for communication/networking, photons are ideal flying qubits, low decoherence.

Challenges: Photon loss is the dominant error, deterministic two-qubit gates are extremely difficult, single-photon sources and detectors have limited efficiency, massive resource overhead for fault tolerance via fusion-based approaches.

Neutral Atoms

Used by Atom Computing, QuEra, Pasqal.

Physical basis: Individual neutral atoms (Rb-87, Cs-133, Sr-88) trapped in optical tweezers -- focused laser beams creating microscopic traps arranged in programmable 2D/3D arrays.

Gates:

Single-qubit: Microwave or optical pulses on hyperfine/clock transitions. Fidelity >99.5%.
Two-qubit: Rydberg blockade -- exciting atoms to high-lying Rydberg states creates strong dipole-dipole interactions that entangle neighboring atoms. CZ gate fidelities: 99.5%.
Programmable connectivity via atom rearrangement (physical movement of traps).

Advantages: Scalability (arrays of >1000 atoms demonstrated), programmable connectivity, identical qubits, mid-circuit measurement and rearrangement, natural for analog quantum simulation.

Challenges: Atom loss during computation, Rydberg gate fidelity limitations, relatively slow cycle times, laser intensity noise.

Topological Qubits

Pursued by Microsoft (Station Q).

Physical basis: Non-abelian anyons (e.g., Majorana zero modes in topological superconductors). Qubit information stored in the global topological state, inherently protected from local perturbations.

Gates: Braiding anyons (physically exchanging their positions) implements topological gates. The gate depends only on the topology of the braid, not the geometric details -- built-in fault tolerance.

Status: Majorana zero modes have been experimentally challenging to confirm unambiguously. Microsoft reported evidence in InAs-Al nanowire devices (2025). If realized, topological qubits could dramatically reduce error correction overhead.

Challenges: Experimental realization remains in early stages, not all gates are topologically protected (T gate requires magic state distillation even with topological qubits), braiding must be supplemented by non-topological operations for universality.

Decoherence

Decoherence describes the loss of quantum information through interaction with the environment.

T1 (relaxation time): Time for energy decay |1> -> |0>. Caused by coupling to environmental degrees of freedom (phonons, photons, TLS defects). Limits computation depth.

T2 (dephasing time): Time for phase coherence loss, governed by T2 <= 2T1. Pure dephasing (T_phi) from low-frequency noise causes |0> + |1> to lose its relative phase. T2 (Ramsey decay, including inhomogeneous broadening) <= T2 (echo decay).

Common decoherence sources:

Charge noise (1/f noise from TLS defects)
Flux noise (magnetic field fluctuations)
Photon shot noise (residual thermal photons)
Quasiparticle poisoning (broken Cooper pairs)
Cosmic ray impacts (correlated errors across chip)

Mitigation: Dynamical decoupling (echo sequences: Hahn echo, CPMG, XY-4), decoherence-free subspaces, noise-aware gate design, material improvements.

Quantum Control

Precise manipulation of qubit states requires:

Pulse engineering: Optimal control theory (GRAPE, Krotov) designs microwave/laser pulses that implement desired unitaries while minimizing leakage to non-computational levels
Calibration: Regular recalibration of gate parameters (frequency, amplitude, duration) to track drifting device characteristics
Crosstalk mitigation: Unwanted interactions between qubits during gate operations, addressed by simultaneous gate optimization and frequency allocation
Randomized benchmarking: Protocol to estimate average gate fidelity independent of state preparation and measurement (SPAM) errors. Variants: interleaved RB (individual gate fidelity), character RB, cycle benchmarking

Current Processors

IBM

Architecture: Fixed-frequency transmon qubits with cross-resonance gates
Processors: Eagle (127 qubits, 2021), Osprey (433 qubits, 2022), Condor (1121 qubits, 2023), Heron (133 qubits with tunable couplers, improved fidelity, 2024)
Roadmap: Modular architecture connecting multiple chips, targeting >100,000 qubits by late 2020s
Software: Qiskit ecosystem, IBM Quantum cloud access

Google

Architecture: Flux-tunable transmon qubits with tunable couplers
Processors: Sycamore (53 qubits, quantum supremacy 2019), Willow (105 qubits, below-threshold surface code error correction, 2024)
Key results: First quantum supremacy demonstration, first demonstration of error rate decreasing with increased code distance
Software: Cirq framework

IonQ

Architecture: Trapped Yb-171 ions in linear Paul traps
Processors: 32+ algorithmic qubits, all-to-all connectivity
Approach: Focus on high gate fidelity and algorithmic qubit count (number usable for computation, not just physical count)

Quantinuum (Honeywell)

Architecture: QCCD (Quantum Charge-Coupled Device) -- trapped ions shuttled between zones
Processors: H-series (H1: 20 qubits, H2: 56 qubits), highest demonstrated two-qubit gate fidelities (>99.8%)
Key results: First real-time quantum error correction demonstrations, best quantum volume scores

Benchmarking Metrics

Quantum Volume (QV): IBM-proposed holistic metric combining qubit count, connectivity, and gate fidelity. QV = 2^n where n is the largest n-qubit random circuit executable with >2/3 success probability. Current records: QV > 2^20.
CLOPS: Circuit Layer Operations Per Second, measuring throughput
Algorithmic qubits: IonQ metric emphasizing usable qubits
Logical error rate per round: Most relevant metric for fault-tolerant QC, measuring residual error after error correction