Hardware approaches: superconducting, ion, photonic, atomic, spin

A physical implementation of a qubit must satisfy a small list of operational requirements, traditionally summarized as the DiVincenzo criteria:

1. Well-defined qubits — a scalable physical system with two reliably distinguishable quantum levels.
2. Initialization — the ability to start the system in a known state (typically $|0\rangle$).
3. Long coherence — the qubit's quantum state must survive long enough to perform a useful number of operations.
4. Universal gate set — a controllable interaction that implements at least one non-Clifford gate plus single- and two-qubit gates.
5. Measurement — a way to read out the qubit's value in a chosen basis with high fidelity.
6. (For communication) — the ability to convert stationary qubits to flying qubits (photons) and to send them faithfully between sites.

Different physical platforms satisfy these criteria with different engineering trade-offs. The rest of this lesson surveys the major families with the numbers that distinguish them.

The leading platform by qubit count is the superconducting transmon: a nonlinear LC circuit using a Josephson junction, fabricated on a chip and cooled to ~15 millikelvin in a dilution refrigerator.

Characteristic engineering numbers:

• Coherence times ($T_1$, $T_2$): typically 50–200 microseconds at the leading edge.
• Single-qubit gate time: ~20 nanoseconds.
• Two-qubit gate time: ~30–200 nanoseconds.
• Single-qubit gate fidelity: ~99.95%.
• Two-qubit gate fidelity: ~99.5–99.8%.
• Readout fidelity: ~98–99%.

Processors are fabricated lithographically on silicon or sapphire substrates. The dilution refrigerator and the microwave control electronics dominate the system's footprint. Major implementations include Google, IBM, Rigetti, and several academic groups.

The main scaling challenges are wiring (each qubit needs control and readout lines from room temperature to the cold plate) and crosstalk (signals from one qubit's control line can affect its neighbors). Processors of several hundred to a thousand physical qubits have been demonstrated; getting from there to the million-qubit scales that fault-tolerant operation would require is an active engineering problem.

Trapped-ion systems use individual atomic ions (e.g., $^{171}\text{Yb}^+$, $^{40}\text{Ca}^+$) confined in a vacuum chamber by oscillating electric fields. Two internal energy levels of each ion encode a qubit. Gates are performed with focused lasers.

Characteristic numbers:

• Coherence times: seconds to minutes — orders of magnitude longer than superconducting.
• Single-qubit gate time: ~1–10 microseconds.
• Two-qubit gate time: ~100 microseconds.
• Single-qubit gate fidelity: ~99.999%.
• Two-qubit gate fidelity: ~99.9% at the leading edge.
• Readout fidelity: ~99.9%.

Major implementations include IonQ and Quantinuum.

The trade-offs against superconducting:

• Better: much higher gate fidelity, much longer coherence, all-to-all qubit connectivity within a trap (any ion can be entangled with any other).
• Worse: much slower gates (microseconds vs nanoseconds), more complex laser control infrastructure, harder to scale beyond ~100 ions per trap.

Larger systems use modular architectures with multiple traps connected by ion shuttling or photonic links between traps.

Photonic implementations use single photons traveling through optical circuits as qubits. Different physical degrees of freedom (path, polarization, time-bin) can encode the qubit value.

Characteristic numbers:

• Coherence times: effectively unlimited at the timescales of useful computation (photons do not decohere in transit).
• Gate operations: beam splitters, phase shifters, polarization rotators — simple linear optics.
• Two-qubit gates: require nonlinear interaction or measurement-induced effects (KLM scheme); deterministic two-photon gates are an open engineering challenge.
• Detection efficiency: ~80–95% for state-of-the-art superconducting nanowire detectors.

The distinctive trade-off: photonic systems can operate at room temperature for the photons themselves, scale via standard photonic integrated-circuit manufacturing, and naturally connect across distance (the qubits are photons). But deterministic two-qubit gates require either heralding (probabilistic success that is then corrected) or active error correction, which raises the resource overhead substantially.

Major implementations include PsiQuantum and Xanadu. The platform's strength is scalability via photonic-IC manufacturing; the open question is whether the gate-resource overhead from probabilistic operations is acceptable at the system scale needed for useful computation.

Neutral-atom systems trap individual atoms (e.g., rubidium, strontium) in arrays of optical tweezers. Qubit states are encoded in hyperfine or Rydberg levels; gates use laser-driven Rydberg interactions.

Characteristic numbers:

• Coherence times: seconds for hyperfine states.
• Single-qubit gate time: ~1 microsecond.
• Two-qubit gate time: ~100 nanoseconds for Rydberg-blockade gates.
• Two-qubit gate fidelity: ~99.5% at the leading edge.
• Array sizes demonstrated: 100–1000+ atoms.

Major implementations include QuEra, Pasqal, and Atom Computing.

The distinctive features:

• Reconfigurable connectivity — atoms can be moved between gate operations, allowing the physical layout of qubit-qubit interactions to change during a computation.
• Higher densities than superconducting because there are no wires per qubit.
• Room-temperature vacuum chamber operation (no millikelvin refrigeration needed for the qubits themselves; only the laser control needs the precision optics).

The trade-off relative to superconducting: gate times are intermediate (microsecond-scale), and the laser control infrastructure scales differently than microwave control. Reconfigurable connectivity gives more flexibility for circuits with non-local interactions.

Silicon spin qubits encode quantum information in the spin of a single electron (or nucleus) confined to a quantum dot fabricated in isotopically purified silicon. Gates are performed with electric or magnetic fields applied through electrodes patterned on the chip.

Characteristic numbers:

• Coherence times: ~1 millisecond in $^{28}\text{Si}$ (other silicon isotopes shorten coherence).
• Single-qubit gate time: ~10–100 nanoseconds.
• Two-qubit gate time: ~100 nanoseconds.
• Single-qubit gate fidelity: ~99.9%.
• Two-qubit gate fidelity: ~99% at the leading edge.

Major implementations include Intel, Diraq, and several research groups.

The distinctive feature is CMOS compatibility. Spin qubits can be fabricated using process steps that overlap with conventional silicon manufacturing, which holds out the prospect of leveraging the existing chip-manufacturing supply chain for control electronics and qubit fabrication itself. The trade-off: qubits are much smaller than superconducting ones, which makes individual qubit control harder; gate fidelities are still climbing toward the level of superconducting and trapped-ion. The platform is younger by demonstrated qubit count than superconducting or trapped-ion, but the scaling argument from CMOS infrastructure is structurally distinct.

Topological approaches encode qubits in the global properties of a many-body quantum state — particularly in non-Abelian anyons such as Majorana zero modes in certain superconductor-semiconductor heterostructures. The encoding is structurally protected against local noise because local perturbations cannot change global topology.

In principle, the platform's distinctive feature is error rates orders of magnitude lower than other platforms for the same physical hardware, because the protection is built in rather than imposed by software-level error correction.

In practice, the experimental status is different from the others. As of writing, the existence of stable topological qubits has been a subject of active experimental verification rather than demonstrated qubit counts and fidelities at the level of the other platforms. Microsoft has been the most public commercial pursuer. The platform's structural promise — exponentially suppressed error rates — is the appeal; the structural difficulty — fabricating and verifying the underlying non-Abelian anyon states — is what has kept it on a longer timeline than the other approaches.

Whether topological qubits become a practical platform depends on experimental confirmation of stable encoded qubits, gate operations, and measurement at fidelities competitive with the alternatives. This is a question the next decade will answer empirically.

No single platform dominates on every metric.

Which platform is 'best' depends on the workload. For gate-heavy circuits where many operations must complete before decoherence, fidelity-per-gate matters most. For large-circuit-depth algorithms requiring many error-corrected logical qubits, the physical-qubit-per-logical-qubit ratio and gate fidelity determine resource requirements. For specialized applications (e.g., distributed quantum networks), photonic interconnects matter more than gate count.

The next lesson moves from these per-physical-qubit numbers to the question of how many physical qubits it takes to make one error-corrected logical qubit — the bridge between current hardware and the regimes where useful algorithms run.