The NISQ Era: Where We Are
Noisy Intermediate-Scale Quantum — where the field stands today and the gap to fault tolerance.
Fault-tolerant quantum computers that run Shor's algorithm on RSA-2048 (~10⁶ physical qubits, error rate <10⁻³) and simulate industrial molecules (FeMoco, nitrogenase) with ~100 logical qubits at error rates below 10⁻⁴.
Noisy Intermediate-Scale Quantum devices — tens to hundreds of physical qubits, no error correction, circuit depths limited by decoherence. Google's 53-qubit supremacy (2019) and photonic Gaussian boson sampling confirmed quantum advantage on specific sampling tasks, but no practically useful algorithms yet. — Preskill (2018)
Current best two-qubit error rates are 10⁻³ to 10⁻². Surface codes need <10⁻³ per gate. At ~441 physical qubits per logical qubit (distance-21 surface code), RSA-2048 breaking requires ~2×10⁶ physical qubits — 3–4 orders of magnitude beyond today.
The DiVincenzo Criteria
Five necessary (and simultaneously required) conditions for a viable qubit platform.
Well-defined |0⟩,|1⟩ two-level subspace, replicable in large numbers without performance degradation.
Reliable preparation of the fiducial |000…0⟩ state before each computation. Requires optical pumping + ground-state cooling.
T₂ ≫ t_gate. Decoherence must be slow vs. gate time. For neutral atoms: T₂ > 1 s is achievable; tweezer photon scattering erodes it.
High-fidelity single-qubit rotations + at least one entangling two-qubit gate (e.g. Rydberg blockade for neutral atoms).
Site-specific measurement in the computational basis with near-unity fidelity, without disturbing neighbours.
Analog vs. Digital Quantum Computing
Two fundamentally different approaches — and why neutral atoms can do both.
🔢 Digital (Gate-Based)
Programs decomposed into discrete unitary gates from a universal gate set, applied to qubits, then measured.
Mathematically clean and compatible with quantum error correction.
Cost: Error correction needs 100s–1000s of physical qubits per logical qubit.
Every gate must exceed the fault-tolerance threshold (~10⁻³ for surface codes).
Who: IBM, Google, IonQ, Quantinuum, QuEra (Rydberg gates)
🌊 Analog (Simulation / Annealing)
Engineer a Hamiltonian whose ground state encodes the answer, then adiabatically evolve from an easy initial state.
Special-purpose, not universally programmable, but far larger system sizes are accessible in the NISQ era.
Examples: D-Wave's 5000+ superconducting flux qubit annealers;
neutral-atom arrays studying quantum magnetism on 2D lattices of hundreds of atoms.
Neutral atoms are unique: the same array can run Rydberg gates (digital) OR evolve
a programmable spin Hamiltonian (analog) — just by changing the control protocol.
Pasqal & QuEra exploit this dual capability as a near-term strategy.
Platform Comparison
Scores normalised 0–10 across six dimensions. Hover for details.
Current Qubit Counts & Gate Fidelities (≈ 2024–25)
Note: Atom Computing Phoenix (1,180 atoms) demonstrated array loading; not all atoms used as qubits simultaneously. PsiQuantum uses photonic fusion gates — qubit count metric is not directly comparable.
Best published values. Superconducting: IBM Heron/Google Willow. Trapped ions: Quantinuum H-series. Neutral atoms: individual atom pair demos (Evered et al. 2023, Harvard). Silicon spins: isotopically-purified ²⁸Si.
Platform Deep Dives
How each technology works, key metrics, bottlenecks, leading companies, and key papers.
Superconducting Qubits
How it works
Transmon qubits — Josephson junction circuits cooled to ~10 mK in dilution refrigerators. Anharmonic oscillators whose two lowest energy levels form the qubit. Microwave pulses drive single- and two-qubit gates. The Josephson junction provides the nonlinearity that makes the energy ladder anharmonic, isolating the |0⟩↔|1⟩ transition.
Key Metrics (≈ 2024)
Gate time: 10–200 ns 1Q fidelity: >99.9% 2Q fidelity: ~99.5% T₁, T₂: 50–500 µs Ops/T₂: ~10³ Qubit count: 100–1000+
Principal Bottleneck
Classical wiring infrastructure inside the dilution refrigerator. Each qubit needs multiple coaxial lines for control and readout. Heat load + cable density become prohibitive beyond ~few thousand qubits. Solutions: cryo-CMOS multiplexing inside the fridge, microwave-to-optical transduction for remote coupling.
Companies
IBM Quantum ↗
Eagle (127Q), Osprey (433Q), Condor (1121Q), Heron r2 (156Q, 2024). Quantum volume roadmap. Cloud access via IBM Quantum Platform.
Google Quantum AI ↗
Sycamore (53Q, 2019 supremacy), Willow (105Q, 2024 — below threshold error correction demonstration).
Rigetti Computing ↗
Aspen/Ankaa series. Publicly traded, cloud hybrid classical–quantum.
IQM Quantum Computers ↗
Finnish startup. Resonance (20Q), modular architecture. Focus on HPC integration.
Alice & Bob ↗
Cat qubit approach — biased noise qubits to reduce overhead for error correction.
D-Wave ↗
Quantum annealing (5000+ flux qubits). Advantage2 processor. Not gate-based.
Trapped Ion Qubits
How it works
Hyperfine "clock" states of ¹⁷¹Yb⁺ or ⁴³Ca⁺ ions confined in radiofrequency Paul traps. Ions are laser-cooled to the motional ground state; shared vibrational modes of the ion chain act as a quantum bus. The Mølmer–Sørensen gate drives an entangling operation via collective phonon modes.
Key Metrics (≈ 2024)
Coherence T₂: 1 s – minutes 2Q fidelity: ~99.9% Gate time: 20–200 µs 1Q fidelity: <10⁻⁴ error Qubit count: 30–56 Ops/T₂: >10⁴
Principal Bottleneck
Motional-mode crowding: adding ions to a linear chain increases shared vibrational modes, making gates slower and harder to address selectively. Scaling path: the quantum charge-coupled device (QCCD) architecture — ions shuttled between storage and gate zones in segmented traps. Photonic interconnects link separate traps into a modular network.
Companies
IonQ ↗
Forte (36 qubits). NASDAQ-listed (IONQ). #AQ metric for algorithmic performance. Partners with AWS, Azure.
Quantinuum ↗
H2 (56 trapped Yb⁺ qubits). Joint venture Honeywell + Cambridge Quantum. Highest algorithmic fidelity demonstrated. H-series roadmap.
Oxford Ionics ↗
Uses electronic microwave signals (no lasers). Built on existing semiconductor fab lines. Quieter than laser-driven gates.
AQT (Alpine Quantum Technologies) ↗
Ca⁺ ion trap systems. European focus. Partners with CERN for HPC integration.
Neutral Atom Qubits (Optical Tweezers) ⭐
How it works
Individual neutral atoms (Rb, Cs, Yb, Sr) trapped in tightly focused laser beams (optical tweezers) — intensity gradient provides a restoring force. Hyperfine ground states form the qubit. Two-qubit gates use the Rydberg blockade: transient excitation to highly-excited Rydberg states (n ~ 60–100) whose strong dipole-dipole interaction (V_dd ~ n⁷) prevents simultaneous excitation of two nearby atoms.
Key Metrics (≈ 2024)
Qubit count: 50–1000+ Coherence T₂: 1–10 s 2Q fidelity: 97–99.5% Gate time: 0.2–5 µs Identical qubits: atoms are perfect copies Reconfigurable geometry: real-time rearrangement
Three Bottlenecks — from this thesis
① Imaging fidelity: Must scatter photons for readout while atom survives.
Prior work on ⁶Li: 90–97% per-image survival. Chapter 3 achieves 99.950(2)% survival over 2000 consecutive images via Λ-enhanced gray molasses.
② Qubit temperature: Rydberg gate fidelity ∝ 1 − α⟨n⟩. At ω_trap ~ 2π×100 kHz,
going from 20 µK → 5 µK cuts motional error by 4×. Chapter 4 demonstrates ⟨n⟩ ≈ 0.01 for Cs via narrow-line
sideband cooling on the 685 nm quadrupole transition.
③ Precision benchmarking: τ(5D₅/₂) for Cs measured to 1% accuracy (Chapter 5),
pinning the saturation intensity, scattering rate, and magic-trap condition to the level needed for systematic gate optimisation.
Companies
QuEra Computing ↗
Aquila (256 atoms). Spin-out from Harvard/MIT (Lukin/Greiner groups). Analog + digital modes. AWS partnership. Largest publicly accessible neutral-atom QPU.
Pasqal ↗
Fresnel (100 Rb atoms). French startup (Browaeys/Lahaye group). Hybrid analog-digital. HPC integration via EDF, BASF partnerships.
Atom Computing ↗
Phoenix (1180 Sr atoms demonstrated). Focus on ⁸⁷Sr nuclear spin qubits (longer T₂). Modular architecture.
Infleqtion ↗
Formerly ColdQuanta. Rb & Cs platforms. Acquired SuperTech. Cloud access & quantum networking.
Key Papers
- Tweezer arrays review (Kaufman & Ni, Nat. Phys. 2021)
- Rydberg blockade gate 99.5% fidelity (Evered et al., Harvard Nature 2023)
- QuEra 256-qubit analog (Ebadi et al., Nature 2021)
- ⁶Li imaging 2000× (Blodgett, Phatak et al., PRL 2023)
- Cs quadrupole cooling (Blodgett, Phatak et al., PRA 2025)
- Generalized cooling theory (Phatak et al., PRA 2024)
Silicon Spin Qubits
How it works
Electron or hole spins confined in silicon/silicon-germanium quantum dots. An electrostatic gate defines a potential well holding a single electron; its spin-up/down states form the qubit. Single-qubit gates: microwave-driven electron spin resonance. Two-qubit gates: exchange interaction tuned by gate voltage. In isotopically purified ²⁸Si (nuclear-spin-free), the dominant dephasing source (background ²⁹Si nuclear spins) is eliminated.
Key Metrics (≈ 2024)
Gate time: 20–50 ns 2Q fidelity: 99.8% (²⁸Si) T₂: >1 ms in ²⁸Si Ops/T₂: ~10⁴–10⁵ Qubit count: 6–12 (2024) Variability: no two dots are identical
Why it's compelling long-term
Integration density: millions of quantum dots can in principle be fabricated using existing CMOS processes (same fabs that make Intel chips). This is the most credible path to millions of physical qubits if variability can be solved. Automated tuning protocols — machine learning for quantum dot tuning — are the active research frontier.
Companies
Intel ↗
Tunnel Falls (12Q, 2023). Fab-compatible process. Cryo-CMOS Horse Ridge controller chip. Integration with Intel's existing 300mm fabs.
Quantum Motion ↗
UK startup. CMOS-compatible Si/SiO₂ platform. Raises Series B 2023. Focus on scalable qubit arrays.
Silicon Quantum Computing ↗
Australian national initiative (UNSW). Precision atom placement in Si:P. Sub-nm gate control.
Equal1 ↗
Irish startup. Full stack on a single CMOS chip — qubits + classical control at 4K.
Photonic Qubits
How it works
Qubits encoded in polarisation, path, or time-bin modes of single photons propagating through integrated waveguide chips. Decoherence is negligible in transmission; photonic chips can integrate millions of modes on a wafer. Single-qubit gates: beam splitters and phase shifters. The fundamental difficulty: photons don't interact — entangling gates must be probabilistic (linear optical quantum computing, KLM protocol) or mediated through engineered nonlinearities.
Fusion-Based QC (PsiQuantum approach)
Rather than deterministic gates, resource states (small entangled photon clusters) are fused by Bell measurements. Failures are heralded and corrected by the architecture itself. Key advantage: photons are naturally flying qubits — perfect for quantum networking and modular QC with optical interconnects. Error rates per fusion: ~1–10%.
Near-term Value
Quantum communication / QKD; optical interconnects between otherwise isolated qubit modules; Gaussian boson sampling demonstrations (Xanadu Borealis, ~216 squeezed modes, 2022 — beyond-classical throughput on specific sampling problems).
Companies
PsiQuantum ↗
Largest-funded photonic QC startup (~$700M). Fusion-based QC on GlobalFoundries silicon photonics. Target: 1M+ photonic qubits.
Xanadu ↗
Borealis (216 squeezed modes). PennyLane open-source framework. Gaussian boson sampling. Amazon Braket partner.
QuiX Quantum ↗
Boson sampling processors. Si₃N₄ waveguide chips. European quantum flagship.
Quandela ↗
Semiconductor quantum dot single-photon sources (InGaAs). Muse cloud platform.
Solid-State Defects & Topological Qubits
NV Centres & Diamond Defects
Nitrogen-vacancy centres in diamond — a nitrogen atom adjacent to a lattice vacancy — have an electron spin with millisecond-to-hour coherence at room temperature. Unique among qubit platforms. Optically addressable; near-room-temperature operation. Principal near-term role: quantum repeater nodes in distributed networks and precision sensing (magnetometry at nanoscale), rather than dense on-chip processors. Related: SiV centres in diamond (Lukin group, Harvard), defects in SiC.
Topological Qubits (Majorana)
Theoretically compelling — Majorana zero modes at the ends of 1D topological superconductors encode quantum
information non-locally, giving intrinsic protection from local perturbations. Error rates theoretically far
below 10⁻⁶. Huge overhead reduction for fault-tolerance.
Status: Unambiguous Majorana detection remains
controversial; no braiding operation has been demonstrated. Physics frontier, not yet an engineering option.
Microsoft announced topological qubit progress (2023–2025) but peer-reviewed braiding demonstrations
are awaited.
Companies & Groups
Microsoft Azure Quantum ↗
Topological qubit programme (Station Q). InAs/Al heterostructures. Announced topological qubit chip (2025). Also partners with IonQ, Quantinuum on Azure.
Quantum Brilliance ↗
Room-temperature NV-centre quantum accelerators. Diamond-based QPU. Partnered with Oak Ridge National Lab.
Q-Next / Argonne ↗
US DOE quantum center. NV centres for quantum repeaters. Quantum network testbed Chicago–Argonne.
Road to Fault-Tolerant Quantum Computing
Milestone timeline across the three leading platforms (2019–2033).
The leading error correction scheme. Encodes 1 logical qubit in d² physical qubits (d = code distance). Can correct any error affecting fewer than d/2 qubits per syndrome round. At physical error rate ~10⁻³, a distance-21 surface code gives one logical qubit at 10⁻¹⁰ logical error rate — at the cost of 441 physical qubits and continuous syndrome measurement. Running Shor's algorithm on RSA-2048 requires ~4000 logical qubits → ~2×10⁶ physical qubits.
Open circles = projected milestones. Projected dates are speculative estimates based on current roadmaps (2025). ⭐ = platform studied in this thesis.
Platform Snapshot Table (≈ 2024)
Reproduced and updated from Table 8.1 in Phatak (2025). Gate times and fidelities are for two-qubit operations unless noted. T₂ is the dephasing time; ops/T₂ is two-qubit gates per coherence time.
| Platform | Qubits (2024) | Gate time (2Q) | 2Q Fidelity | T₂ | Ops / T₂ | Key Bottleneck |
|---|---|---|---|---|---|---|
| Superconducting | 100–1000+ | 50–200 ns | 99.5% | 50 µs – 0.5 ms | ~10³ | Classical wiring at mK |
| Trapped Ions | 30–56 | 20–200 µs | 99.9% | 1 s – minutes | >10⁴ | Mode crowding & gate speed |
| ⭐ Neutral Atoms | 50–1180 | 0.2–5 µs | 97–99.5% | 1–10 s | 10³–10⁴ | Gate fidelity, mid-circuit readout |
| Silicon Spins | 6–12 | 20–50 ns | 99.8% | >1 ms (²⁸Si) | ~10⁴–10⁵ | Variability & tuning |
| Photonic | >100 modes | probabilistic | 90–97% | N/A | limited | No photon-photon interaction |
| NV / Topological | 1–few | µs–ms | ~99% (NV) | ms – hours | varies | Scale / Majorana undemonstrated |
⭐ = Neutral atoms: the platform studied in this thesis (Phatak 2025, Purdue Hood Lab).
Neutral Atom Roadmap: What Needs to Happen
Five engineering milestones on the path from today's demonstrations to fault-tolerant neutral-atom QC.
Optimal-control pulse shaping + ground-state-cooled atoms (⟨n⟩ < 0.1). Chapters 3 & 4 address this directly for Li and Cs.
Multiplexed AOD arrays + parallel imaging. Atom Computing Phoenix demonstrated 1180 atoms loaded (2023).
Feed-forward operations demonstrated at few-qubit level. Essential for active error correction.
Modular scaling beyond a single array. Photonic links between tweezer modules. Research stage.
First demonstrations expected following similar achievements on superconducting (Google 2023) and trapped-ion systems.