77. Beyond Noisy Intermediate-Scale Quantum Computing
77.1NISQ
Preskill's 2018 name for the era of 50–1,000 physical qubits, no error correction, algorithms limited to circuits shallow enough to finish before decoherence wins. NISQ's 2026 status: it delivered physics (quantum supremacy-class sampling experiments, useful calibration and error-mitigation science) but not products — the variational application program (Part XIII's 49–50) largely underdelivered against its 2018–21 billing, as Parts XIII and XV honestly recorded. The NISQ era's real legacy is the engineering base for what replaces it: calibration automation, control electronics, benchmarking culture, and cloud access — plus the pedagogical truth this book is built on: NISQ-era tools made the entire field laptop-accessible. The era is ending not with failure but with graduation.
77.2Fault-Tolerant Quantum Computing
The successor: machines where errors are corrected faster than they accumulate — logical operations with error rates far below any physical component's. The physics that makes it possible: the threshold theorem (Part X) — below a critical physical error rate, adding code distance suppresses logical error exponentially. The 2024 proof point: Google's Willow demonstrated below-threshold operation — logical error halving as distance grew (d=3→5→7), plus a hours-long logical-memory survival far beyond any physical qubit's coherence. The engineering that remains: scale (one logical qubit ≈ 10²–10³ physical, so useful machines need 10⁵–10⁷ physical qubits), real-time decoding (Ch. 37's µs constraint), and the entire logical-gate architecture (magic states, lattice surgery) at industrial scale. FTQ is no longer a physics question; it is the largest systems-engineering program in computing.
77.3Logical Qubits
The abstraction the whole era turns on: a logical qubit is an encoded, actively-corrected quantum register whose effective error rate and coherence beat its physical constituents. 2025–26 state of the art: 10²-scale physical:logical ratios demonstrated in research settings; groups (Harvard/QuEra neutral atoms, Google, IBM, Quantinuum trapped ions) showing 10s of logical qubits with fidelity exceeding physical-qubit operations on some gates. What makes a logical qubit good: error suppression factor Λ per distance step, logical clock speed (cycles/second — the underrated metric; a perfect logical qubit at 1 kHz runs nothing), and gate universality (Cliffords are cheap; T gates need magic-state distillation — Ch. 79). Watch this space the way a 1970s engineer watched transistor density: it is the field's Moore's-law-shaped curve, and everything (algorithms, applications, careers) hangs on its slope.
77.4Scalable Architectures
The four horses, 2026 edition: superconducting (IBM's Starling roadmap: ~200 logical qubits by 2029 from ~10⁵ physical; Google's Willow lineage scaling lattices; strengths — speed and fabrication ecosystem; weaknesses — cryogenic I/O and 2D wiring congestion); trapped ions (Quantinuum: highest physical fidelities, all-to-all connectivity, slow gates; the fidelity-first bet); neutral atoms (QuEra/Harvard lineage: 1,000+ atom arrays demonstrated, reconfigurable connectivity, long coherence; the newest serious entrant); photonics (PsiQuantum: skipping NISQ entirely toward FTQ in silicon-photonic foundries; the highest-risk/highest-leverage bet). Plus the fusion/sqc and cat-code outliers. The engineering truth: no architecture has won; each hits a different wall (I/O, speed, fidelity, scale); and the 2030s will select on systems economics, not physics elegance — which is why Ch. 44–45's compiler-and-control layer may decide the race as much as qubit physics does.
77.5Quantum Memory
The unglamorous, potentially decisive component: storing quantum states longer than computation needs them. Two meanings, both live: logical memory (surface-code-stabilized registers — Willow's hours-long survival was this; the metric for memory benches is rounds-per-failure and Λ) and physical quantum memories (for networking: ensembles, cavities, defects — Ch. 78's repeater bottleneck). Why memory may decide applications: many FTQ algorithm costs (Ch. 79) are dominated by data intake and intermediate storage, not logic — a machine with cheap memory changes which algorithms are practical. The research texture: memory benchmarks are where "logical qubit quality" claims go to be tested (anyone can run a deep circuit once; holding 100 qubits for a million rounds is the exam). Watch for memory-bandwidth metrics appearing in vendor roadmaps — their arrival will signal the FTQ era's application-planning phase.
77.6Quantum Networking
The connective frontier: linking processors with quantum channels — entanglement distribution, teleportation-based gates between modules, and eventually the quantum internet (Ch. 78 maps it fully). Its place in this chapter: networking is a scaling strategy — modular architectures (networked small machines) may beat monolithic ones (one giant cryostat) on engineering grounds, the way classical computing chose clusters over single exaflop CPUs. 2026 status: metropolitan quantum networks deployed (QKD-flavored), entanglement between nodes demonstrated across room-distance and satellite-distance, but quantum repeaters — the enabling technology — remain laboratory-scale (the no-amplification theorem forces memory-based repeating; Ch. 78.3). The crossover to watch: when networked modules' interconnect fidelity beats local scaling's wiring penalty — a systems-economics event, and one your compiler-era skills (Ch. 46's routing, generalized to network topologies) would directly serve.