38. The Physical Quantum Computer
38.1What a qubit physically is
Two states of some physical system: two energy levels of an atom, charge or flux states of a superconducting circuit, spin orientations of an electron, polarization of a photon. "Qubit" is the mathematics; the platform is the engineering that creates two levels, isolates them from everything else, drives transitions between them, and reads the result out. Real systems have more than two levels — a transmon is an oscillator with infinitely many — and the computational subspace is carved out of the ladder; falling out of it is leakage (29.6). Every platform in this part is a different answer to one question: which two levels of which system, at what cost in every other requirement?
38.2Requirements for a physical qubit
DiVincenzo's checklist (1997) is still the standard scorecard: (1) a scalable system of well-characterized qubits; (2) initialization to a simple fiducial state; (3) coherence times much longer than gate times; (4) a universal set of quantum gates; (5) qubit-specific measurement. For networking, add interconversion between stationary and flying qubits. Apply it to every platform in this part and you will find each one fails some criterion gracefully and one painfully: superconductors are fast but forgetful, ions are stable but slow, photons fly but barely interact. The checklist also explains the field's structure — each hardware subfield exists because one criterion is hard for the others.
38.3Coherence
Coherence times (29.2–29.3) set the clock. What matters is not raw T1 but the ratio to gate time: error per idle period ≈ duration / coherence time. Transmon: T1 ≈ 100 µs with gates ≈ 20 ns — a ratio near 5,000. Trapped ions: coherence in seconds to minutes with gates of 10–600 µs — a ratio near 10⁴–10⁵. Similar ratios from opposite extremes: fast-but-forgetful versus slow-but-stable. Note that coherence is engineered, not granted — transmon T1 rose from ~1 ns in 1999 to ~100 µs–1 ms today through materials, surface treatment, shielding, and filtering, and it is still improving. Demand this arithmetic from any platform claim: ratios, not adjectives.
38.4Controllability
You must drive arbitrary single-qubit rotations and at least one entangling gate, faster than decoherence, with errors ≲ 10⁻³. Control is analog: shaped microwave pulses, laser pulses, flux pulses — the "program" of a quantum computer is ultimately a waveform library. Stronger, faster drives buy speed but spill into higher levels (leakage, 29.6) and bleed into neighbors (crosstalk, 29.7); that triangle — speed, leakage, crosstalk — is the daily trade of control engineering. The classical control stack (FPGAs, arbitrary waveform generators, GHz front ends) is a major cost center of every machine and, increasingly, the most software-intensive subsystem — a realistic entry point for engineers who cannot touch the physics.
38.5Readout
Extract the state quickly, with high fidelity, without disturbing the neighbors: dispersive microwave readout for superconducting circuits (39.4), state-dependent fluorescence for ions and atoms (40.7), photon counting for photonics (42.6). Fidelities run 99–99.9%, durations from ~100 ns to milliseconds — and readout time counts twice: it bounds the machine's repetition rate and consumes coherence budget in error-correction cycles (Part X). Readout should be quantum non-demolition: measure without flipping. Errors here are the only noise you can patch purely in software (29.5). When benchmarking platforms, always ask for readout fidelity and readout time separately — they trade against each other.
38.6Scalability
From a 50-qubit laboratory demonstration to 10⁶ qubits, wiring is the recurring villain. Each superconducting qubit wants 2–3 coaxial lines through a refrigerator with microwatt cooling power — hence cryo-CMOS multiplexing as the proposed escape. Ion traps need laser delivery to every qubit; neutral atoms need optical control over thousand-site arrays; photonics needs on-chip sources and detectors at scale. Platform roadmaps diverge here more than anywhere else, and small-scale laboratory results say almost nothing about the scaling limit. The engineering rule: ask for the per-qubit wiring, cooling, and optics budget at 10⁶ qubits, and ask what component technology has to change to get there.
38.7Calibration
Every qubit needs its frequencies, pulse amplitudes, and readout discriminators calibrated — and recalibrated, because parameters drift (29.8). A large chip runs continuous automated calibration as a scheduling and optimization pipeline, consuming hours of machine time per day. This is also one of the most software-heavy parts of "hardware" companies: calibration is a classical control loop wrapped around an analog quantum system, and teams that build it look a lot like robotics teams. As a remote user you see the outputs — per-qubit, per-gate error rates published daily. Study those pages before running anything: they are the most honest performance data in the industry.