Quantum Technologies: A Practical Reference

A transmon is a superconducting circuit that behaves as an artificial atom. A Josephson junction supplies nonlinear inductance, a large shunt capacitor suppresses sensitivity to charge noise, and the resulting anharmonic oscillator has two lowest levels that serve as the qubit. It is controlled with microwave pulses at 4–8 GHz and read out through a coupled resonator. Everything lives on a chip at the bottom of a dilution refrigerator at 10–20 mK, because the qubit energy has to be far above the thermal noise.

Gates run in tens of nanoseconds, which is thousands of times faster than trapped ions, and the devices are made with lithography, so scaling is a fabrication problem rather than an assembly problem. The whole semiconductor toolkit applies. The weaknesses are coherence and uniformity. Coherence times of 100–300 microseconds are short in absolute terms, no two fabricated qubits are quite identical so each needs individual calibration, and connectivity is nearest-neighbor on a planar chip. Every qubit also needs its own control lines into the fridge, which is the wiring problem that caps how large one refrigerator can go.

This is the platform with the most industrial investment and the clearest path to large physical qubit counts, so it is the default assumption in most roadmaps. Bet on it if you believe error correction overhead can be paid down by fabrication scale. Be skeptical of qubit-count announcements without two-qubit fidelity and connectivity alongside, since a chip with more qubits and worse errors is worse. If your interest is algorithms that need all-to-all connectivity or very deep circuits, trapped ions currently deliver better fidelity per gate.

Operates at 10–20 mK in a dilution refrigerator · qubit frequencies 4–8 GHz · coherence times typically 100–300 microseconds · two-qubit gates in 20–100 ns with fidelities around 99.5–99.9% · chips of 100–1,000+ physical qubits, each needing its own control lines.

Google's Willow processor and its surface code error correction results; IBM's Heron and Condor devices; Rigetti's multi-chip modules; the 2019 Sycamore random circuit sampling experiment that started the current investment cycle.

Superconducting qubits attract the most capital because they look like a semiconductor business: design, fabricate, test, repeat. The cost structure is dominated by the dilution refrigerator and the control electronics rather than by the chip, which is why a 1,000-qubit system costs roughly what a 100-qubit one does plus wiring. That also means the economics improve sharply if cryogenic control electronics work, which is why every large player funds that work.

6 minThe transmon qubit | QuTech AcademyQuTech Academy · 10k+ views

11 minHow to Turn Superconductors Into A Quantum Computer | Superconducting Qubits 1Lukas's Lab · 10k+ views

2 minA Qubit in the MakingIBM Research · 50k+ views

Quantum Computing: Progress and Prospects (National Academies) · Quantum information science (NIST)

A fluxonium replaces the transmon's large capacitor with a large inductance, built from a long array of Josephson junctions. That changes the energy landscape: the qubit frequency drops to a few hundred megahertz to 1 GHz, and the two computational states have very different wavefunctions, which suppresses the dielectric loss channel that limits transmon coherence. The result is coherence times an order of magnitude longer than a transmon's, at the cost of a more complicated circuit and harder control.

Coherence of a millisecond or more, against a few hundred microseconds for transmons, is a large enough gap to change the error-correction arithmetic, and demonstrated two-qubit fidelities above 99.9% are among the best on any solid-state platform. Higher anharmonicity also allows shorter pulses without leakage. The obstacles are complexity and consistency. The junction array must be fabricated uniformly, the low frequency makes thermal population and readout harder, and the parameter space is larger, so calibration is more involved. Very few groups have shown it at more than a handful of qubits.

Watch fluxonium as the most credible challenger inside the superconducting family. It matters if coherence rather than count turns out to be the binding constraint, which is the bet several startups have made. Do not plan a near-term deployment around it, since the largest demonstrations are far behind transmon devices. When comparing to transmons, look at the product of coherence time and gate speed rather than either alone, because fluxonium's longer coherence comes with slower gates.

Coherence times reaching 1 ms and beyond, roughly an order of magnitude above transmons · qubit frequency typically 0.1–1 GHz, well below a transmon's 4–8 GHz · two-qubit fidelities demonstrated above 99.9% · requires an array of many Josephson junctions per qubit · demonstrated systems remain in the single or low double digits of qubits.

Google's published comparison of transmon and fluxonium architectures; Atlantic Quantum and Alice & Bob among the startups pursuing non-transmon superconducting qubits; university groups at Maryland and Yale, where much of the fluxonium work originated.

5 minQubit Types | Part 1 | Superconducting QubitsQuantum Leap · 10k+ views

12 minFluxonium QubitZiqi Qu · under 1k views

8 minHow To Make a Quantum BitVeritasium · 1m+ views

About (Quantum, the open journal for quantum science)

Most qubits encode information in a two-level system and then need many physical copies to protect one logical bit. A bosonic qubit instead encodes information in the many levels of a single microwave resonator, choosing states whose errors are easy to detect and correct. Cat qubits use two coherent states of opposite phase, stabilized by a two-photon drive, which makes bit-flip errors exponentially rare in photon number while leaving phase flips as the only error to correct. GKP codes do something similar with a grid of states in phase space.

Biasing the noise so that one error type essentially disappears changes the error-correction budget dramatically: a code that only has to correct phase flips needs far fewer physical qubits than a surface code correcting both. Demonstrated bit-flip times reach minutes. The costs are control complexity and the fact that the theory advantage has to survive engineering. Stabilizing the cat state requires continuous nonlinear drive and careful engineering of loss channels, readout is harder than for a transmon, and the largest demonstrations are single logical qubits rather than systems.

Follow this as a strategy bet rather than a product. It matters because it attacks the overhead problem directly instead of trying to build ten million physical qubits, and if the noise bias holds up at scale, the resource estimates for useful algorithms fall by an order of magnitude. Do not plan a deployment around it before the mid-2030s. When evaluating claims, ask for the bit-flip time and the phase-flip time separately, since the whole argument rests on the ratio between them.

Encodes a logical qubit in one resonator rather than in many two-level systems · demonstrated bit-flip times of minutes, with phase flips remaining the dominant error · noise bias reduces the physical qubit overhead for error correction substantially · requires continuous stabilizing drive · largest demonstrations are single logical qubits.

Alice & Bob's cat qubit processors and their published bit-flip lifetimes; AWS's Ocelot chip, which combines cat qubits with a repetition code; Yale's original bosonic code experiments, where the approach was invented.

5 minCat Qubits and LDPC Codes, a New Step Towards Quantum Error CorrectionAlice & Bob · 5k+ views

4 minWhat is QUANTUM ERROR CORRECTION (QEC)? 💊 Quantum Pill #4QPlayLearn · 1k+ views

Quantum Computers Glossary (QuEra)

A quantum annealer is not a gate-based computer. It encodes an optimization problem as the ground state of a coupled spin system, initializes the hardware in an easy-to-prepare state, and then slowly changes the Hamiltonian toward the problem one, so the system ideally follows its ground state to the answer. The hardware is superconducting flux qubits with programmable couplings, thousands of them on a chip in a dilution refrigerator. It has been commercially available longer than any gate-based machine.

Machines exist, are large by qubit count, and are usable over the cloud today, which no gate-based platform matches for optimization work. For certain problems the mapping is natural and results come back in milliseconds. The weaknesses are the ones that have made this contested for a decade. Annealing addresses a narrow class of problems, connectivity is limited so an embedded problem uses many physical qubits per logical variable, and rigorous evidence of a general speedup against good classical heuristics remains scarce. Most reported advantages have later been matched classically.

Try an annealer when your problem is genuinely a quadratic unconstrained binary optimization and you can benchmark it honestly against simulated annealing and modern classical solvers on the same instances. It is cheap to test over the cloud. Do not assume that a fast answer is a quantum speedup; run the classical baseline first, and check how many physical qubits each logical variable consumed after embedding. For simulating quantum systems, analog neutral-atom and ion platforms have the stronger recent record.

Thousands of flux qubits on a chip, with limited fixed connectivity · operates at 10–20 mK like other superconducting hardware · embedding a densely connected problem costs many physical qubits per logical variable · available commercially over the cloud since 2011 · claims of speedup have repeatedly been matched by improved classical algorithms.

D-Wave's Advantage and Advantage2 systems; logistics and scheduling pilots at Volkswagen, Denso, and several utilities; the long series of academic papers testing and often deflating specific speedup claims.

6 minWhat is Quantum Annealing?D-Wave · 100k+ views

6 minHow The Quantum Annealing Process WorksD-Wave · 100k+ views

5 minAdiabatic Quantum Computer Explanation D-Wave SystemsPlusUltraTech · 5k+ views

Quantum Computing Applications (D-Wave) · Quantum Computing: Progress and Prospects (National Academies)

A trapped-ion computer holds individual charged atoms, usually ytterbium or barium, in ultra-high vacuum using oscillating electric fields from a chip trap. Laser cooling brings them nearly to rest in a line. Qubit states are two long-lived internal levels of the ion, addressed with lasers or microwaves, and two-qubit gates work by coupling internal states through the ions' shared motion in the trap. Every ion of a given species is identical, which removes the calibration problem that dogs solid-state qubits.

Fidelity and connectivity are the advantages. Two-qubit gate fidelities above 99.9% are routine, coherence lasts seconds to minutes, and because all ions share the same motional modes, any pair can interact directly, giving all-to-all connectivity that saves enormous numbers of swap operations. The costs are speed and optics. Gates take tens to hundreds of microseconds, thousands of times slower than superconducting, and every ion needs precisely controlled laser beams, so the system is fundamentally an optics engineering problem that gets harder with each added ion.

Choose trapped ions when circuit depth and gate quality matter more than raw speed, which describes most near-term algorithm work: a machine with 30 excellent all-to-all qubits runs deeper useful circuits than one with 100 noisy nearest-neighbor ones. Assume the scaling path runs through shuttling architectures or photonic links between traps rather than through longer chains, because a single chain becomes unmanageable past a few tens of ions. Check gate speed against your circuit depth, since slow gates plus deep circuits means long wall-clock runtimes.

Two-qubit gate fidelities above 99.9%, the best demonstrated on any platform · coherence times of seconds to minutes · gate times of tens to hundreds of microseconds · all-to-all connectivity within one ion chain · practical chain lengths of a few tens of ions before motional modes become unmanageable.

Quantinuum's H-series systems, which hold most fidelity records; IonQ's Forte and Tempo machines; the NIST and Innsbruck groups where ion trapping for computing was developed; the 2012 Nobel Prize in Physics for the underlying trapping and measurement techniques.

9 minLearn Quantum: What is Trapped Ion Quantum Computing?IonQ · 10k+ views

2 minHow Trapped-Ion Quantum Computers WorkQuantinuum · 10k+ views

3 minWhat is an Ion Trap Quantum Computer?Simon Benjamin · 10k+ views

Benchmarking a trapped-ion quantum computer with 30 qubits (Quantum) · Quantum information science (NIST)

A single ion chain stops working well past a few tens of ions, because the motional modes crowd together and gates get slower and noisier. The quantum charge-coupled device architecture solves this by moving ions instead of adding them to one line. The trap is divided into zones, some for storage and some for gates, and ions are physically shuttled between them by changing the electrode voltages, the way charge is shifted across a CCD sensor. Gates happen in a small zone with a couple of ions, where fidelity stays high.

It preserves trapped-ion fidelity while allowing far more qubits, and it keeps effective all-to-all connectivity because any two ions can be brought together. Shuttling takes tens to hundreds of microseconds and adds little error when done well. The costs are control complexity and heating. Each transport is a carefully designed voltage waveform, ions gain motional energy during transport and must be re-cooled, and the trap becomes a complex multi-layer chip with many electrodes. Throughput drops because much of the machine's time is spent moving rather than computing.

This is the mainstream scaling path for ion traps, so if you are evaluating trapped-ion vendors you are evaluating QCCD implementations. It fits work needing high fidelity at a few tens to a few hundreds of qubits. Watch the ratio of transport time to gate time in published results, because a machine that spends most of its cycle shuttling has a wall-clock problem even if its fidelities are excellent. Beyond a few hundred ions, expect photonic interconnects between separate traps to become necessary.

Gate zones hold a small number of ions, keeping fidelity near single-pair performance · transport operations take tens to hundreds of microseconds each · ions require re-cooling after transport, which consumes cycle time · traps are multi-layer chips with tens to hundreds of electrodes · demonstrated systems in the tens of qubits, with roadmaps to hundreds.

Quantinuum's H2 system, the most developed QCCD machine; Sandia National Laboratories' surface trap fabrication, which supplies much of the field; university demonstrations of junction transport that showed ions can turn corners without losing coherence.

18 minHow to trap ions for quantum computing?Professor Nano · 10k+ views

6 minTILT: Achieving Higher Fidelity on a Trapped-Ion Linear-Tape Quantum Computing Architecture, HPCA21Quantum Computing · under 1k views

Benchmarking a trapped-ion quantum computer with 30 qubits (Quantum)

Neutral atom machines hold uncharged atoms, usually rubidium or a strontium isotope, in arrays of optical tweezers: tightly focused laser spots, each holding one atom, arranged by a spatial light modulator or an acousto-optic deflector into any geometry you like. Because the atoms are neutral they do not repel each other, so thousands can be packed close together. Two-qubit gates use the Rydberg blockade: exciting one atom to a huge orbital shifts its neighbor's energy levels enough to prevent the same excitation, which is a conditional interaction.

Scale is the headline. Arrays of thousands of atoms have been assembled, and rearranging them into arbitrary geometries is a software change rather than a fabrication run, which is why neutral atoms dominate analog quantum simulation. Atoms are identical by nature. The weaknesses are loss and speed. Atoms escape the traps over seconds and must be reloaded, gate fidelities lag trapped ions, and measurement destroys the atom in most schemes, so mid-circuit measurement, which error correction needs, is hard. Everything depends on a large and stable laser system.

Use neutral atoms for analog quantum simulation of many-body physics now, where they are the leading platform and produce results classical methods struggle to match. For gate-based computing, treat them as the fastest-improving platform: the first logical-qubit demonstrations on this hardware appeared in 2024 and the trajectory has been steep. Weigh the laser system honestly, since a machine that needs dozens of stabilized lasers has a reliability profile more like a physics experiment than an appliance.

Arrays of thousands of atoms assembled in arbitrary two- and three-dimensional geometries · atom lifetime in the trap of seconds, requiring reloading · two-qubit fidelities around 99.5%, behind trapped ions · gates in the hundreds of nanoseconds to microseconds via Rydberg blockade · everything rests on a large stabilized laser system.

QuEra's Aquila analog simulator, available over the cloud; Harvard and MIT's logical qubit demonstrations; Atom Computing's arrays exceeding a thousand qubits; Pasqal's systems in Europe.

3 minNeutral Atom Quantum Computers - Concept of Operation | QuEraQuEra Computing · 10k+ views

10 minThe Rise of the Neutral Atom Quantum ComputerThe Quantum Bull · 1k+ views

6 min6 Quantum Computing with Rydberg atoms - Prof Jonathan PritchardQSimFP · under 1k views

Neutral Atom Quantum Computing: Principles, Routes, Progress, and Challenges (arXiv) · Fault-tolerant quantum computation with a neutral atom processor (arXiv)

A silicon spin qubit stores information in the spin of a single electron confined in a quantum dot, an electrostatically defined puddle a few tens of nanometers across in a silicon or silicon-germanium heterostructure. Gate electrodes above the dot hold the electron and tune its energy; a magnetic field splits the spin states; microwave or electrical pulses rotate the spin. Two-qubit gates use the exchange interaction between neighboring dots. The device looks and is fabricated like a very small transistor.

Size and manufacturability are the case. A spin qubit is roughly 100 nm across against a transmon's hundreds of microns, so millions could fit on a die, and isotopically purified silicon-28 removes the nuclear spins that would otherwise decohere it, giving coherence beyond a second. Fabrication runs in a CMOS fab. The problems are uniformity and wiring. Dot-to-dot variability is severe at these dimensions, exchange coupling is very short range so connectivity is strictly nearest-neighbor, and each dot needs several gate electrodes, which recreates the wiring problem at smaller scale.

Follow this as the long-horizon bet with the best endgame: if spin qubits become uniform enough to fabricate by the million, the scaling argument is stronger than any other platform's. Do not expect useful systems before the late 2030s, since demonstrations are still at six to twelve qubits. It is also the platform where progress in semiconductor manufacturing, rather than in physics, moves the needle, so watch foundry involvement as the leading indicator rather than qubit counts.

Quantum dots roughly 50–100 nm across, against hundreds of microns for a transmon · coherence beyond a second in isotopically purified silicon-28 · single and two-qubit fidelities above 99% demonstrated · operates at 10–100 mK, though some devices work above 1 K, which would ease cooling · demonstrated arrays remain in the single digits to low tens of qubits.

Intel's Tunnel Falls 12-qubit research chip made on 300 mm wafers; Delft and QuTech's spin qubit work, including operation above 1 K; Diraq and Quantum Motion commercializing CMOS-compatible designs; UNSW's donor-based qubits in silicon.

6 minQuantum Dot QubitsQuTech Academy · 5k+ views

11 minSpin qubits | QuTech AcademyQuTech Academy · 10k+ views

4 minSpin Qubits Explained: Quantum Computing with SiliconCodeLucky · under 1k views

Quantum information science (NIST)

A nitrogen-vacancy center is a point defect in diamond: a nitrogen atom next to a missing carbon atom. The resulting electron spin can be initialized with green light, read out through the brightness of its red fluorescence, and manipulated with microwaves, all at room temperature. Its energy levels shift with magnetic field, electric field, temperature, and strain, which makes it a sensor as much as a qubit. Diamond's stiff lattice and low spin-orbit coupling are what allow room-temperature operation, and nothing else offers that.

Room-temperature quantum coherence with optical readout is the unique property, and it makes NV centers the most commercially deployed quantum device outside atomic clocks. Sensitivity is combined with nanoscale spatial resolution, since a single defect can sit within nanometers of a sample. The costs are material and photon budget. Diamond has to be grown with controlled nitrogen and isotopic purity, placing defects precisely is still imperfect, and only a few percent of the emitted photons are usable for coherent networking, which is why entanglement rates between NV nodes remain slow.

Use NV centers when you need magnetic or thermal sensing at micron or nanometer scale, at room temperature, with no cryogenics: current imaging in chips, magnetic microscopy of geological and biological samples, and battery and motor diagnostics. Watch them as network nodes, where they combine a long-lived spin memory with an optical interface. Do not choose them as a computing platform, where coherence per defect is good but scaling to many coupled defects has no established path.

Operates at room temperature with optical initialization and readout · coherence up to milliseconds in isotopically purified diamond · magnetic sensitivity in the nanotesla per root hertz range for ensembles, with single defects giving nanometer spatial resolution · responds to magnetic field, electric field, temperature, and strain · only a small fraction of emitted photons are usable for entanglement.

Element Six and Adamas supplying engineered diamond material; SandboxAQ's NV magnetometry for navigation and medical sensing; Delft's entanglement of NV nodes over kilometers, a milestone toward a quantum network; scanning NV microscopes imaging currents in devices.

7 minNV center qubits | QuTech AcademyQuTech Academy · 10k+ views

4 minThe quantum world of diamondsnature video · 50k+ views

7 minOperations on NV center qubits | QuTech AcademyQuTech Academy · 10k+ views

Quantum information science (NIST)

The NV center is one of many optically active defects. Silicon carbide hosts divacancy and silicon-vacancy centers in a material that already has a commercial wafer industry. Group IV defects in diamond, tin-vacancy and silicon-vacancy, emit a much larger fraction of their photons into a sharp line, which is what a network needs, at the cost of requiring a few kelvin rather than room temperature. Rare-earth ions in crystals, erbium especially, emit at 1,550 nm, so their photons travel down existing telecom fiber.

These defects exist to fix the NV center's specific weaknesses: photon indistinguishability for group IV centers, telecom wavelength for erbium, and an industrial substrate for silicon carbide. Erbium at 1,550 nm is the only solid-state spin that can talk to the deployed fiber network without conversion. The weaknesses are that each fix trades something away. Group IV centers need 1–4 K cooling, erbium's optical transition is weak so it needs a resonator to be useful, and defect placement and yield remain unsolved for all of them.

Track these as the materials answer to quantum networking rather than as computing platforms. If your interest is a repeater node that connects to installed fiber, erbium and telecom-compatible defects are the ones to watch. If it is a room-temperature sensor, stay with NV. Judge progress by photon indistinguishability and by entanglement rate between two separated nodes, since those are the numbers that decide whether a network is possible, and both are far below what a useful repeater needs.

Group IV diamond defects emit a large fraction of photons into the zero-phonon line, against a few percent for NV · they require 1–4 K rather than room temperature · erbium emits at 1,550 nm, matching the fiber loss minimum · silicon carbide brings an existing 150 and 200 mm wafer industry · deterministic defect placement remains the limiting fabrication problem.

Tin-vacancy centers in diamond demonstrating high-fidelity spin-photon interfaces; erbium-doped crystals and silicon devices for telecom-band memories; commercial silicon carbide wafers repurposed for defect qubits; Photonic Inc.'s silicon T centers, which emit near the telecom band.

6 minQuantum ColorFermilab · 100k+ views

6 minColor center: Nitrogen vacancy center in diamond정규석 · 1k+ views

Quantum Flagship

A photonic quantum computer encodes qubits in single photons, in polarization, path, or time bin, and processes them with beamsplitters, phase shifters, and detectors on a photonic integrated circuit. Photons do not interact with each other, so two-qubit gates are made probabilistically through measurement: entangle, measure, and keep the runs that worked. The practical architecture is fusion-based or measurement-based computing, which builds a large entangled resource state from small ones and then computes by measuring it in a chosen order.

Photons do not decohere in the usual sense, they run at room temperature in the waveguides, and the chips can be made in existing silicon photonics foundries, which is a manufacturing base no other platform has. Networking is native, since the qubit is already a photon. The costs are loss and detectors. Every optical component loses some photons, and loss is the dominant error, so components must reach fractions of a decibel; the single-photon detectors that make it work are superconducting nanowires needing cryogenics, so the system is not really cryogenics-free.

Follow photonics as the platform whose scaling argument is manufacturing rather than physics: if a foundry can print millions of low-loss components, the architecture works, and if it cannot, it does not. That makes progress legible through component loss figures and photon source efficiency rather than through qubit counts. It is also the natural technology for quantum networking. Do not expect an intermediate small-scale useful machine, because this architecture is designed for fault tolerance at large scale rather than for noisy intermediate devices.

Qubits are single photons; the waveguides run at room temperature while the detectors need cryogenics · loss is the dominant error, so components must reach fractions of a dB · two-qubit gates are probabilistic, requiring many attempts and heralding · fabricated in commercial silicon photonics foundries · architecture targets fault tolerance directly rather than a noisy intermediate stage.

PsiQuantum's foundry-based fusion architecture, built with GlobalFoundries; Xanadu's Aurora and its photonic error-correction demonstrations; Chinese Jiuzhang boson sampling experiments; the same silicon photonics platform used for datacenter transceivers.

2 minHow Xanadu’s Photonic Quantum Computers WorkXanadu · 100k+ views

2 minXanadu Demonstrates World's First Error-Resistant Photonic QubitXanadu · 5k+ views

12 minXanadu [CHAC] Goes Public | The First Photonic Quantum Computing Stock ExplainedThe Quantum Bull · 1k+ views

Quantum Flagship

Instead of counting photons, the continuous-variable approach encodes information in the amplitude and phase quadratures of a light field, the same variables a radio engineer would recognize. Squeezed light sources reduce the quantum noise in one quadrature at the expense of the other, and measurement is homodyne detection with an ordinary fast photodiode rather than a single-photon counter. Large entangled cluster states can be generated deterministically in time or frequency, which sidesteps the probabilistic gate problem of the photon-counting approach.

Deterministic entanglement generation and room-temperature homodyne detection are real advantages, and cluster states with very large numbers of modes have been demonstrated. Squeezed light is also directly useful outside computing: it is what improved LIGO's sensitivity. The weakness is that fault tolerance requires encoding qubits into these continuous variables through GKP states, and generating high-quality GKP states optically is extremely hard. Finite squeezing puts a floor under the error rate, and the squeezing levels demonstrated remain short of the fault-tolerance threshold.

Treat this as research with one deployed application. Squeezed light is already used in gravitational-wave detection and is a credible route to improved optical sensing, so the sensing case stands on its own. For computing, follow the squeezing figure in decibels and the quality of demonstrated GKP states, because those are the two numbers that decide whether the architecture can ever be fault tolerant. Do not compare it to gate-based platforms by qubit count; the resource is modes and squeezing, not qubits.

Information encoded in field quadratures rather than photon number · homodyne detection uses ordinary fast photodiodes at room temperature · entangled cluster states of thousands to millions of modes demonstrated · squeezing of roughly 10–15 dB achieved, below the level fault tolerance is thought to need · squeezed light is already deployed in gravitational-wave detectors.

Xanadu's Borealis and Aurora machines; the University of Tokyo and DTU cluster state experiments; squeezed light injection in LIGO and Virgo, which improved their range measurably; continuous-variable QKD systems using the same detection technique.

1 minQuantum frontiers | Xanadu’s photonic approach to quantum computingXanadu · 1k+ views

About (Quantum, the open journal for quantum science)

A topological qubit would store information in a global property of a system rather than in any local degree of freedom, so local noise could not corrupt it. The leading candidate uses Majorana zero modes, quasiparticles predicted to appear at the ends of a semiconductor nanowire coupled to a superconductor in a magnetic field. Information would be encoded in how those modes are braided around each other, which is a topological operation and therefore insensitive to small perturbations. Microsoft has pursued this route for two decades.

If it works, hardware-level error protection would collapse the overhead that dominates every other platform's roadmap, turning a million-physical-qubit requirement into something far smaller. That payoff is why the bet has been funded so long. The problem is that the existence of Majorana zero modes in these devices remains contested. A 2018 Nature paper claiming evidence was retracted in 2021, and the 2025 Majorana 1 announcement drew immediate scientific criticism about whether the measured signatures are topological or ordinary Andreev states. The materials, epitaxial semiconductor-superconductor hybrids, are also extremely demanding.

Treat this as a high-variance research position rather than a technology to plan around. It is worth tracking because the payoff is qualitatively different from incremental improvement, and worth discounting heavily because the foundational claim is still disputed by working physicists. When a result is announced, look for whether independent groups reproduce the signature and whether the paper distinguishes topological modes from trivial alternatives, because that distinction is the entire question.

Requires millikelvin temperatures and a magnetic field applied to a semiconductor-superconductor nanowire · promise is hardware-level protection, potentially reducing error-correction overhead by orders of magnitude · a prominent 2018 Nature paper claiming Majorana evidence was retracted in 2021 · the 2025 Majorana 1 result remains scientifically contested · no braiding operation has been demonstrated.

Microsoft's Majorana 1 chip and the debate that followed it; Delft's retracted 2018 result and the reanalysis that followed; indium arsenide and aluminum hybrid nanowires as the material system; the theoretical work by Kitaev that started the field.

12 minMajorana 1 Explained: The Path to a Million QubitsMicrosoft · 1m+ views

23 minMicrosoft's Topological Quantum Computer ExplainedDomain of Science · 500k+ views

7 minMicrosoft Announces Breakthrough With Quantum ChipSabine Hossenfelder · 100k+ views

Quantum Computing: Progress and Prospects (National Academies)

An atomic clock counts oscillations of an atomic transition. Cesium microwave clocks, which define the SI second, tick at 9.2 GHz. Optical clocks use a transition in the visible, around 500 THz, so each tick is 50,000 times shorter and the same fractional stability buys far more precision. Strontium and ytterbium lattice clocks hold thousands of atoms in an optical lattice; aluminum and ytterbium ion clocks use one trapped ion. An optical frequency comb converts the optical tick down to a countable microwave signal.

These are the most accurate measuring devices humans have built, reaching fractional uncertainty around 10^-19, which corresponds to gaining or losing under a second over the age of the universe. That sensitivity makes them instruments as well as clocks: at this level the clock rate changes measurably with a centimeter of height in Earth's gravity, so they measure geopotential. The costs are size and fragility. A record-setting clock fills a laboratory, needs several stabilized lasers and an ultra-stable cavity, and takes expert operators. Transportable versions exist but are far from a product.

Optical clocks matter to you if you care about the redefinition of the second, about geodesy, or about tests of fundamental physics. Commercially, they are entering the market as compact optical clocks that outperform rubidium and cesium standards for holdover in telecom and defense timing, which is the real near-term application: staying accurate when GPS is denied. For ordinary timing, a chip-scale or rubidium clock is orders of magnitude cheaper and entirely sufficient.

Optical transitions near 500 THz against 9.2 GHz for cesium, so roughly 50,000 times more ticks per second · fractional uncertainty around 10^-19 in the best ion clocks, 41% better than the previous record set in 2025 · sensitive enough to detect a 1 cm change in elevation through gravitational time dilation · laboratory systems fill a room; transportable versions are demonstration hardware.

NIST's aluminum ion quantum logic clock, the current accuracy record holder; strontium lattice clocks at JILA and PTB; the international effort toward redefining the SI second on an optical transition; transportable clocks used for relativistic geodesy field campaigns.

5 minHow optical clocks are redefining time and physicsNew Scientist · 10k+ views

5 minOperating principle of an optical atomic clockAQuRA · under 1k views

3 minHow does the NIST-F4 fountain clock work?National Institute of Standards and Technology · 1k+ views

NIST Ion Clock Sets New Record for Most Accurate Clock in the World (NIST) · Portable Optical Lattice Clock (NIST)

A chip-scale atomic clock puts a cesium or rubidium vapor cell, a VCSEL, and a photodiode into a package the size of a matchbox. The laser is modulated so that two of its sidebands drive a coherent population trapping resonance in the vapor, which shows up as a sharp change in transmitted light and is used to lock the local oscillator. Everything is made with microfabrication rather than assembled by hand, which is what took atomic timekeeping from a rack to a component.

It gives atomic-clock holdover in a device drawing well under a watt, and that combination has no alternative: a good crystal oscillator drifts by microseconds within an hour, while a chip-scale clock stays within a microsecond for a day. Cost is thousands rather than tens of thousands. The weaknesses are that it is far less stable than a rack-mounted rubidium standard, sensitive to temperature and magnetic field, and the vapor cell ages, so performance drifts over years. It is a holdover device, not a primary standard.

Use a chip-scale clock wherever GPS timing might be lost and the system must keep working: undersea sensors, seismic arrays, military radios, and network equipment needing holdover. It is also the standard choice for platforms too small for a rack-mounted standard. Use a rubidium or cesium standard where stability over days matters and there is room and power for it. And be clear about the requirement: many designs specify an atomic clock when a disciplined oscillator with occasional GPS lock would do.

Package volume of roughly 15–20 cm3 and power under 1 W · holdover typically within a microsecond over a day, against microseconds per hour for a good crystal oscillator · unit cost of a few thousand dollars · far less stable than a rack rubidium standard · vapor cell aging causes slow drift over years.

Microchip's SA.45s chip-scale atomic clock, the first commercial device of this type, developed from DARPA-funded work at NIST and Sandia; undersea seismic and sonar nodes; military handheld radios needing timing without a GPS lock; timing holdover in telecom base stations.

20 minTESTED: Chip Scale Atomic Clock (Precision Timing & Frequency Reference)Baltic Lab · 10k+ views

3 minChip-Scale Atomic Clocks Serve Low-Power & Remote DutiesMicrowaves & RF · under 1k views

1 minChip Scale Atomic Clock (CSAC) Designed for Extreme EnvironmentsMicrochip Technology, Inc. · 1k+ views

Atomic Devices and Instrumentation Group (NIST)

An optically pumped magnetometer measures magnetic field through atomic spins in a vapor cell. A laser polarizes the alkali atoms, usually rubidium or cesium, and the field causes those spins to precess, which changes how much light the vapor transmits. Run the cell hot and in a very low field and spin-exchange collisions stop dephasing the atoms, the regime called SERF, where sensitivity reaches a few femtotesla per root hertz. That is comparable to a SQUID, without liquid helium.

Matching SQUID sensitivity without cryogenics is what made these commercially important: a magnetoencephalography system built from OPM sensors can sit on a subject's head like a helmet instead of requiring them to stay motionless under a fixed dewar, which finally allows brain imaging of children and of people moving. Sensors are small and cheap enough to array. The limits are field range and environment. SERF operation needs the ambient field nulled to nanotesla, so shielding or active compensation is mandatory, and the cells run at 150 °C, which complicates wearable designs.

Use OPMs where you need very weak magnetic fields measured without cryogenics: biomagnetic imaging, magnetic anomaly detection, and non-destructive testing. They are the clear successor to SQUIDs in magnetoencephalography. Use a fluxgate or Hall sensor where the field is strong and the requirement is ruggedness rather than sensitivity, since those are orders of magnitude cheaper. And plan the shielding, because in an unshielded room a SERF magnetometer measures the building, not the sample.

Sensitivity of a few femtotesla per root hertz in SERF operation, comparable to a SQUID · needs no cryogenics, against liquid helium for SQUIDs · requires ambient field nulled to the nanotesla level · vapor cells operate near 150 °C · sensor heads small enough to array by the hundreds on a wearable helmet.

Cerca Magnetics and QuSpin wearable magnetoencephalography systems; magnetic anomaly detection for submarine and unexploded ordnance survey; SandboxAQ's magnetic navigation work; the SERF technique developed at Princeton, which made the sensitivity possible.

5 minMagnetoencephalography: measuring brain activity with magnetismAlt Shift X · 100k+ views

2 minQuantum Sensing Explained: Optically Pumped Magnetometers, Vapor Cells, and Brain ImagingHAMAMATSU PHOTONICS · under 1k views

8 minMEGIN discusses Optically Pumped Magnetometers (OPMs) and conventional Squid-based MEG utilization.MEGIN · 1k+ views

Atomic Devices and Instrumentation Group (NIST)

NV magnetometry uses nitrogen-vacancy centers in diamond as magnetic field sensors. The defect's spin resonance frequency shifts with field, and because the spin can be read optically, a microwave sweep plus a photodiode gives a field measurement. Two modes exist. A dense ensemble in a millimeter of diamond gives good sensitivity in a small solid-state package. A single NV at the tip of a scanning probe gives nanometer spatial resolution, trading sensitivity for the ability to map fields inside a working device.

It works at room temperature in a solid, so the sensor can be pressed against a sample, dropped into a downhole tool, or built into a chip tester. Nothing else combines nanotesla sensitivity with nanometer resolution. Vector information comes free, because the four possible defect orientations in the crystal resolve field direction. The costs are material and sensitivity. Ensembles reach picotesla per root hertz, a hundred times worse than an OPM, engineered diamond with the right nitrogen content and isotopic purity is expensive, and placing defects at a controlled depth is still imperfect.

Use NV magnetometry when proximity or spatial resolution matters more than absolute sensitivity: imaging current paths in a packaged chip, measuring magnetic domains, monitoring battery cells and motors in place, and downhole or field instruments where a shielded room is impossible. Use an OPM when you need the last two orders of magnitude of sensitivity and can control the environment. For simple field measurement at strength, a Hall probe costs a dollar.

Ensemble sensitivity of picotesla per root hertz, roughly a hundred times behind SERF magnetometers · single-defect probes give nanometer spatial resolution · operates at room temperature in a solid-state package · vector field measurement comes from the four defect orientations in the diamond lattice · engineered diamond material is a significant share of sensor cost.

Scanning NV microscopes imaging current distribution in integrated circuits; SandboxAQ's diamond magnetometers for navigation and cardiac sensing; Element Six supplying engineered diamond; geological and paleomagnetic imaging of rock samples at micron scale.

17 minQuantum Sensing With a Special Synthetic DiamondAsianometry · 50k+ views

4 minQuantum Sensing Explained | SandboxAQSandboxAQ · 100k+ views

UK National Quantum Technologies Programme

An atom interferometer treats atoms as waves. A cloud of laser-cooled atoms is released, and laser pulses act as beamsplitters and mirrors for the matter wave, sending it along two paths that are then recombined. The phase difference between the paths depends on acceleration and rotation along them, so the interference fringes read out gravity or motion directly. Because the measurement is referenced to the atoms' mass and to the laser wavelength, it does not drift the way a mechanical sensor does.

No drift is the point. A quantum gravimeter holds its calibration indefinitely, which conventional spring gravimeters cannot, and a quantum inertial unit would let a vehicle navigate for hours without GPS instead of minutes. Sensitivity is excellent. The costs are size, speed, and dynamics. A commercial gravimeter is a cubic meter and tens of kilograms, the measurement is a cycle of cooling and interrogation rather than a continuous reading, and under vibration or high acceleration the atoms leave the interrogation region, which is exactly the condition a navigation system faces.

Use a quantum gravimeter today where absolute drift-free gravity measurement is worth the size: geodesy, volcano and aquifer monitoring, and civil survey for voids and tunnels. Track quantum inertial navigation as a real but not-yet-deployed capability; the physics works in the laboratory and the engineering problem is operating under motion. For any application inside a moving vehicle, ask specifically about performance under vibration, since that is where laboratory numbers and field numbers diverge most.

Gravimeter sensitivity in the microgal range with no long-term drift · a measurement cycle of cooling and interrogation, typically a few hertz rather than continuous · commercial units are roughly a cubic meter and tens of kilograms · performance degrades sharply under vibration and acceleration · navigation-grade inertial systems remain in field trials.

Muquans and AOSense commercial atom gravimeters used in geophysical survey; UK National Quantum Technologies Programme trials of gravity sensing to find buried infrastructure; shipborne and airborne quantum inertial navigation trials by several navies; the 1997 Nobel Prize for laser cooling that made all of it possible.

3 minHow does Gravio, a quantum gravimeter, work?Atomionics · 10k+ views

3 minHow can we use atoms and photons as quantum sensors?Science Animated · 10k+ views

UK National Quantum Technologies Programme · Quantum Flagship

Quantum illumination is a real theoretical result: entangle two microwave modes, send one at a target, keep the other, and a joint measurement of the returned signal against the retained idler can beat any classical detector using the same transmitted energy, by up to 6 dB in the limit of very weak signals and heavy background noise. Quantum radar is the proposal to build a radar on that idea. The physics is sound; the engineering is where it comes apart.

The theoretical advantage is genuine and is one of the clearest examples of an entanglement-enabled sensing gain. In the laboratory it has been demonstrated at short range. The problems are decisive for radar. Storing the idler mode requires a quantum memory good enough to survive the round-trip time, which for kilometers means milliseconds and does not exist at microwave frequencies; the entangled source must sit in a dilution refrigerator; and the 6 dB ceiling is small compared with what an extra decade of classical processing gain provides. Published analyses conclude a practical quantum radar would underperform a conventional one.

This entry exists to be a check on a claim you will encounter. When a quantum radar capability is announced, ask three questions: what quantum memory holds the idler for the round trip, what temperature the source runs at, and what the range is. So far the honest answers are none, millikelvin, and meters. Quantum-enhanced sensing is real and productive elsewhere on this sheet, in clocks, magnetometers, and gravimeters, where the advantage does not depend on preserving entanglement across a long noisy channel.

Theoretical advantage capped at about 6 dB in the ideal weak-signal limit · requires a quantum memory to hold the idler for the full round-trip time, which no microwave technology provides · entangled microwave sources need millikelvin cooling · laboratory demonstrations operate over meters in shielded conditions · published analyses find conventional radar outperforms it at any practical range.

Laboratory quantum illumination demonstrations at IST Austria and elsewhere; periodic media reports of Chinese quantum radar systems, none with published performance data; the sustained skeptical literature analyzing why the laboratory advantage does not transfer to a field radar.

10 minQUANTUM RADAR: what is it? Will it defeat STEALTH?Millennium 7 * HistoryTech · 10k+ views

12 minI Saw the END of STEALTH - The New Chinese Quantum RadarMillennium 7 * HistoryTech · 50k+ views

Quantum networking technologies (UK National Cyber Security Centre)

Quantum key distribution lets two parties generate a shared secret key whose security rests on physics rather than on computational hardness. In the BB84 protocol the sender encodes random bits on single photons in randomly chosen bases and the receiver measures in randomly chosen bases; they compare bases publicly, keep the matching cases, and estimate the error rate. Any eavesdropper measuring the photons disturbs them and raises that error rate, so interception is detectable. Commercial systems have existed for two decades.

The security argument does not weaken as computers improve, which is the entire appeal against a future quantum attacker. Systems are commercially available and interoperate over standard fiber. The weaknesses are practical and have kept adoption narrow. Fiber loss caps range at 100–200 km without a trusted node, key rates fall exponentially with distance, and the security proof covers the protocol rather than the hardware, so side-channel attacks on real detectors and sources have repeatedly been demonstrated. Trusted-node networks reintroduce exactly the trust QKD was meant to remove.

Consider fiber QKD for a short, fixed, high-value link where you control both endpoints and the fiber between them, and where a physics-based argument has institutional value: a bank between two datacenters, a government link between two buildings. For anything else, national security agencies including the NSA and the UK NCSC recommend post-quantum cryptography instead, on the grounds that it needs no new hardware and no trusted nodes. Treat QKD as a complement for specific links, not as a replacement for cryptography.

Practical range 100–200 km of fiber without trusted nodes, with key rate falling exponentially with distance · key rates of kilobits to megabits per second at short range, dropping to bits per second at the limit · needs dedicated or carefully managed fiber · single-photon detectors usually require cooling · security proofs cover protocols, and implementation side channels have been exploited repeatedly.

ID Quantique and Toshiba commercial systems; China's 2,000 km Beijing-Shanghai backbone, which relies on trusted relay nodes; banking and government links in Switzerland, Korea, and Singapore; the published guidance from NSA and NCSC recommending post-quantum cryptography over QKD for general use.

13 minHow Quantum Key Distribution Works (BB84 & E91)Improbable Matter · 50k+ views

17 minOutsmarting Hackers: Quantum Key Distribution ExplainedQiskit · 10k+ views

4 minQuantum Key Distribution (QKD) Explained step by step, Request a Demo from QNu LabsQNu Labs · 10k+ views

Quantum networking technologies (UK National Cyber Security Centre)

Fiber attenuates photons exponentially, which caps terrestrial QKD at a couple of hundred kilometers. Free space through vacuum does not, and most of the atmosphere's thickness is in the first ten kilometers, so a satellite link loses far less than the equivalent fiber. A low-Earth-orbit satellite generates entangled photon pairs or prepares BB84 states and sends them to optical ground stations during a pass, distributing keys between stations thousands of kilometers apart without any trusted relay in between.

It solves the range problem and, in the entanglement-based version, removes the need to trust intermediate nodes, which is the strongest security argument in the field. China's Micius demonstrated entanglement distribution over 1,200 km and intercontinental key exchange. The costs are availability and rate. A low-orbit satellite is over a given station for a few minutes per pass, keys accumulate slowly, clouds block the link entirely, and the ground stations are telescopes with tracking systems. Building and launching the satellite is the obvious other cost.

Satellite QKD is a national infrastructure decision rather than an enterprise purchase. It is the credible route to intercontinental key distribution without trusted nodes, and several countries are building programs on that basis. If your requirement is protecting data against a future quantum computer, post-quantum cryptography does it today with a software update; satellite QKD addresses a narrower case where a physics-based guarantee across a long distance is worth an orbital asset.

Loss through the atmosphere is far below the equivalent fiber path, since most of the atmosphere is in the first 10 km · Micius distributed entanglement over 1,200 km and supported intercontinental key exchange · a low-orbit pass gives a few minutes of link time per ground station · cloud cover blocks the link completely · ground segment requires tracking optical telescopes.

China's Micius satellite and the Beijing-Vienna video call it secured; the European Union's EuroQCI and IRIS2 programs; the UK and Singapore SpeQtral missions; several commercial ventures proposing constellations for key delivery.

3 minQuantum satellite achieves 'spooky action' at record distanceScience Magazine · 100k+ views

8 minQuantum Cryptography ExplainedPhysics Girl · 100k+ views

2 minPhD student explains Quantum CommunicationsDepartment for Science, Innovation and Technology · 50k+ views

Quantum Flagship

A classical repeater amplifies a signal, and quantum information cannot be copied, so that approach is unavailable. A quantum repeater instead divides a long link into segments, entangles adjacent nodes over each segment, and then performs entanglement swapping at each node so that the end points end up entangled with each other without any photon having traveled the whole distance. Each node needs a quantum memory that holds its half of the entanglement while neighboring segments are established, and purification to clean up accumulated errors.

Repeaters would remove the trusted-node problem that limits every deployed quantum network and would let distributed quantum computing and long-baseline quantum sensing work at all. Nothing else does that. The obstacles are memory and rate. A memory has to hold coherence for at least the classical communication time across a segment, tens of milliseconds for 100 km, while also having an efficient optical interface at a wavelength fiber transmits, and no material yet gives all three at once. Demonstrated entanglement rates between separated nodes are hertz or below.

Treat repeaters as pre-competitive research whose progress is worth tracking through two numbers: memory coherence time against segment latency, and entanglement rate between physically separated nodes. Those decide feasibility more than any announcement of a new node type. Do not plan a quantum network deployment on repeaters before the 2040s. In the meantime, trusted-node networks and satellite links are what exist, and both make security compromises repeaters are meant to remove.

Memory must hold coherence longer than the classical signaling time across a segment, roughly 0.5 ms per 100 km each way · demonstrated entanglement rates between separated nodes are hertz or below, against the megahertz a useful link needs · requires an optical interface at a fiber-friendly wavelength, usually via frequency conversion · no material yet combines long memory, efficient interface, and telecom wavelength.

Delft's entanglement between three network nodes, the first multi-node quantum network; Harvard's silicon-vacancy memory nodes on deployed Boston fiber; the Quantum Internet Alliance in Europe; erbium and rare-earth memories pursued specifically for telecom compatibility.

4 minQuantum RepeatersMazzi · 5k+ views

9 min12-3 Reaching for distance: Entanglement swappingQ-Leap Edu Quantum Communications · 5k+ views

Quantum information science (NIST)

Cryptography needs randomness, and classical sources are deterministic processes that merely look random. A quantum random number generator takes its entropy from a measurement whose outcome is fundamentally undetermined: which way a photon goes at a beamsplitter, when a photon arrives, or the vacuum fluctuations measured by homodyne detection. The last of these is the commercially dominant approach, because it needs only a laser, a beamsplitter, and a fast photodiode, all of which fit on a chip.

The entropy source is provably unpredictable rather than merely hard to predict, which is a stronger claim than any classical generator can make, and modern devices deliver gigabits per second in a package small enough for a phone. Cost has fallen to the point where the technology is shipping in consumer hardware. The weakness is that the claim applies to the physics, not to the device. Real generators have classical noise mixed in, and a device that is not continuously health-tested can degrade silently. Certification schemes exist precisely because "quantum" on the label is not by itself an assurance.

Use a QRNG where entropy quality is a genuine risk: key generation at scale, hardware security modules, gaming and lottery systems, and any environment where a virtual machine's classical entropy pool is thin. Insist on a certified device with continuous health testing and a documented extractor, since that is what distinguishes a real product from a marketing claim. For most applications a well-seeded operating system generator is entirely adequate, and the failures in practice come from seeding rather than from the algorithm.

Entropy from an inherently indeterminate quantum measurement rather than from a deterministic process · commercial rates from megabits to tens of gigabits per second · chip-scale devices now integrated into phones and security modules · output must pass continuous health tests and a randomness extractor · certification under standards such as NIST SP 800-90B is what makes the claim auditable.

ID Quantique's chip-scale generators, shipped in Samsung phones; Quantinuum and Quside commercial units; QRNG modules inside hardware security modules used by banks; NIST's Randomness Beacon, which publishes public random values.

9 minQuantum Random Number Generation - Do we really need it?Cryptosense · 5k+ views

3 minWhat Is a Quantum Random Number Generator (QRNG)? True Randomness for Modern CryptographyCyberpedia by Palo Alto Networks · under 1k views

3 minQuantum Randomness and Random Number GenerationQuantum Data World · under 1k views

Quantum information science (NIST)

Post-quantum cryptography is classical mathematics chosen to resist a quantum attacker. Shor's algorithm breaks RSA and elliptic curve cryptography, which is nearly all public-key cryptography deployed today, so the response is to replace those algorithms with ones based on problems no efficient quantum algorithm is known for. NIST standardized three in 2024: ML-KEM for key establishment, based on module lattices, and ML-DSA and SLH-DSA for signatures, from lattices and hash functions respectively.

It runs on existing hardware as a software change, works over the internet as it is, and needs no trusted nodes, satellites, or cryogenics, which is why the security agencies that considered both recommend it over QKD for general use. Deployment is already underway in browsers and messaging. The costs are size and confidence. Keys and signatures are considerably larger than elliptic curve equivalents, which stresses protocols with tight packet budgets, and the assumption that these problems are hard is a mathematical belief rather than a proof, as the break of the SIKE candidate in 2022 demonstrated.

Start now, for two reasons. Migration of a large estate takes years, and "harvest now, decrypt later" means traffic captured today is exposed whenever a cryptographically relevant quantum computer arrives. Prioritize long-lived secrets and anything with a decade-plus confidentiality requirement. Deploy in hybrid mode, combining a post-quantum algorithm with a classical one, so a break in either leaves you no worse off. Build crypto-agility into the design, because the standards will evolve.

NIST standardized ML-KEM, ML-DSA, and SLH-DSA in August 2024 · ML-KEM public keys are roughly 1.2 kB against 32 bytes for X25519 · hybrid deployment combining classical and post-quantum algorithms is the recommended transition · the SIKE candidate was broken classically in 2022, after years of analysis · large enterprise migrations are expected to take five to ten years.

Chrome and Cloudflare deploying hybrid X25519 with ML-KEM at internet scale; Signal's PQXDH protocol; NSA's Commercial National Security Algorithm Suite 2.0 timeline; national migration mandates in the US and Europe.

2 minNIST's Post-Quantum Cryptography Standardization ExplainedSandboxAQ · 1m+ views

3 minPost-Quantum Cryptography: the Good, the Bad, and the PowerfulNational Institute of Standards and Technology · 100k+ views

13 minPost Quantum Cryptography - ComputerphileComputerphile · 100k+ views

Post-Quantum Cryptography (NIST Computer Security Resource Center)

A dilution refrigerator reaches 10 mK by exploiting a quirk of helium isotopes: below 0.87 K a mixture of helium-3 and helium-4 separates into two phases, and moving helium-3 across the boundary from the concentrated into the dilute phase absorbs heat, much as evaporation does. Circulating helium-3 continuously gives continuous cooling with no moving parts in the cold section. Modern cryogen-free systems use a pulse tube cooler to reach 4 K first, so no liquid helium bath is needed.

It is the only technology that reaches millikelvin continuously, so every superconducting and spin qubit machine sits inside one. Cryogen-free designs run for months without intervention, which turned the dilution fridge from a specialist instrument into infrastructure. The costs are money, space, and cooling power. A system runs several hundred thousand dollars to over a million, fills a room with compressors and gas handling, and provides only microwatts to a milliwatt of cooling at base, which is what really limits how many control lines a fridge can accept.

If your platform needs millikelvin, you need one, so the design questions are cooling power at base and how many coaxial lines the fridge can accept without overwhelming it. That second number, not the qubit chip, is what caps a single-fridge machine at a few thousand qubits today. Watch helium-3 supply, which comes from tritium decay in weapons stockpiles and is genuinely scarce. If a platform can run at 1 K or above, the cooling problem gets dramatically easier, which is a large part of why spin qubits above 1 K are interesting.

Base temperature 6–20 mK, with cooling power of roughly 10–500 microwatts at 20 mK · cryogen-free systems run months without intervention · system cost from several hundred thousand dollars to over a million · helium-3 supply comes from tritium decay and is scarce · thermal load from control wiring, not the qubit chip, limits how many lines a fridge can take.

Bluefors and Oxford Instruments as the dominant suppliers; IBM's Goldeneye, a large-volume fridge built for scaling; Fermilab and academic groups using them for dark matter detection as well as qubits; the multi-fridge halls now being built by every large quantum computing effort.

6 minWorking Principle of a Dilution Refrigerator.QEL@UCL · 10k+ views

3 minWhat is a dilution fridge? | Quantum science at FermilabFermilab · 10k+ views

3 minBehind The Tech : CryostatsAlice & Bob · 5k+ views

Quantum Computers Glossary (QuEra)

Every superconducting or spin qubit needs microwave control and readout, and today that comes from room-temperature instruments connected by coaxial cables running down into the refrigerator. Each line carries heat in and takes space, and a few thousand lines is roughly where a fridge gives up. Cryogenic control electronics moves the signal generation and digitization to the 4 K stage, where the cooling power is watts rather than microwatts, so one chip can drive many qubits over a handful of digital fibers.

It attacks the wiring bottleneck directly, and that bottleneck, not qubit fabrication, is what caps a single-fridge system today. Cryo-CMOS also shortens the signal path, which reduces latency for feedback-based error correction. The costs are power and noise. Even a few milliwatts per qubit multiplies badly against a 4 K stage's cooling budget, so the design constraint is power per channel rather than performance, and a noisy controller sitting near the qubits can degrade the very coherence it was meant to enable. Transistor behavior also changes at 4 K, so the models have to be rebuilt.

This is infrastructure rather than a purchase decision, but it is the item to watch if you are assessing whether a superconducting roadmap is credible past a few thousand qubits. Ask any vendor claiming a path to a million qubits how the control lines are handled, because the honest answers are cryogenic electronics, multiplexing, or photonic links, and each has a different maturity. Progress is measured in milliwatts per qubit and in demonstrated fidelity with the cryogenic controller in the loop.

The 4 K stage offers watts of cooling against microwatts at 20 mK, which is why controllers go there · power budgets of a few milliwatts per qubit channel are the design constraint · replaces one coaxial line per qubit with a few digital fibers · transistor models must be recharacterized at cryogenic temperature · demonstrated controlling small numbers of qubits at full fidelity.

Intel's Horse Ridge cryogenic control chip; Google and QuTech cryo-CMOS demonstrations controlling transmons; SemiQon and other startups building cryogenic CMOS; multiplexed readout schemes that share one line among many qubits as the nearer-term alternative.

10 minSolving the Control Electronics Bottleneck for Quantum Computing: CryoCMOSAri Noori · under 1k views

6 minWhy Quantum Computers Need Cryo-CMOS Controllers (The Cable Problem Solved)The Honeyed Truth · under 1k views

Quantum Computers Glossary (QuEra)

Every atomic platform, ions, neutral atoms, clocks, and magnetometers, runs on lasers. A trapped-ion machine needs sources for cooling, repumping, state preparation, gates, and readout, each at a specific wavelength, each frequency-stabilized to a reference, each routed through acousto-optic modulators to individual atoms. A research system fills several optical tables. Making these platforms into products means turning that table into a rack, which is the work of integrating lasers, modulators, and switches into photonic chips and delivering light through fiber and waveguides.

This is where the reliability of every atomic platform actually lives, and the trend is favorable: integrated photonics can put beam routing, switching, and delivery on a chip, and trap chips with built-in waveguides have delivered light to ions without any free-space alignment. That removes the drift and the daily realignment that make laboratory systems fragile. The difficulty is that the requirements are severe. Linewidths of hertz, absolute frequency stability, and precise power control at each site are all needed at once, and integrated components have loss and thermal sensitivity that discrete optics do not.

Judge any atomic-platform company partly on this stack, because it predicts uptime better than qubit count does. A machine that needs a laser expert on site every morning is not a product. The questions worth asking are how many discrete lasers the system uses, how they are stabilized, whether beam delivery is free space or integrated, and what the demonstrated continuous operating time is. Photonic integration is the direction of travel and is the reason atomic platforms may become rack-mountable this decade.

A trapped-ion system typically needs five or more distinct wavelengths, each stabilized · linewidths of kilohertz to hertz depending on the transition · research systems occupy several optical tables; product versions target racks · integrated trap chips have delivered gate light to ions with no free-space alignment · continuous operating time without realignment is the honest measure of maturity.

Sandia and university trap chips with integrated waveguide delivery; Vescent, Toptica, and M Squared supplying stabilized laser systems to the field; QuEra and Pasqal's laser subsystems for tweezer arrays; the compact laser packages developed for chip-scale atomic clocks, which showed the path.

Quantum Flagship · UK National Quantum Technologies Programme

Physical qubits fail far too often to run a useful algorithm, so information is spread across many of them in a code whose syndrome measurements reveal what error occurred without revealing the data. The surface code is the leading choice for planar hardware because it needs only nearest-neighbor connections and tolerates physical error rates below about 1%, the highest threshold of any practical code. Its cost is overhead: roughly a thousand physical qubits per logical qubit at realistic error rates, and more for the magic states that non-Clifford gates require.

It is what turns a noisy device into a computer, and 2023–2024 brought the first demonstrations that adding qubits actually reduces logical error rather than increasing it, which was the load-bearing assumption. Newer qLDPC codes cut the overhead severalfold at the cost of requiring longer-range connections, which some platforms can provide. The costs are the numbers. A thousand-to-one overhead means a machine factoring RSA-2048 needs millions of physical qubits, and the decoder has to identify errors faster than they accumulate, which is a real-time classical computing problem in its own right.

Error correction is the lens for reading any quantum computing claim. A useful machine is described by logical qubits and logical error rate, not by physical qubit count, and a platform that improves fidelity reduces overhead superlinearly, which is why a 99.9% platform with fewer qubits can beat a 99% platform with many more. When someone announces a milestone, ask whether the logical error rate went down as the code grew, since that is the only result that indicates a path to scale.

Surface code threshold near 1% physical error rate, the highest of any practical code · roughly 1,000 physical qubits per logical qubit at realistic error rates · qLDPC codes cut that severalfold but need longer-range connectivity · decoders must run in real time, within microseconds for superconducting hardware · cryptographically relevant algorithms imply millions of physical qubits.

Google's Willow result showing logical error falling as the surface code distance grew; Quantinuum and Microsoft's logical qubit demonstrations on trapped ions; Harvard and QuEra's logical operations on neutral atoms; IBM's shift toward qLDPC codes in its published roadmap.

17 minSuppressing quantum errors by scaling a surface code logical qubitGoogle Quantum AI · 5k+ views

5 minQuantum error correction codes | QuTech AcademyQuTech Academy · 10k+ views

5 minParameters of a surface code | QuTech AcademyQuTech Academy · 10k+ views

Quantum Computing: Progress and Prospects (National Academies)

Qubit count is the number quoted in headlines and the least informative one available. A machine's usefulness depends on how many operations it can run before errors accumulate, which means gate fidelity, connectivity, coherence, speed, and how well the whole system holds calibration. Several metrics try to capture this. Quantum volume combines width and depth into a single number that rewards fidelity and connectivity. CLOPS measures circuit throughput. Algorithmic benchmarks time actual problems. Randomized benchmarking isolates average gate error from state preparation and measurement error.

Composite metrics stop the qubit-count arms race from driving hardware decisions in the wrong direction, and randomized benchmarking gives a clean, platform-independent gate error figure. Algorithmic benchmarks measure what a user would actually experience. The weaknesses are gaming and scope. Any single number can be optimized for rather than earned, quantum volume saturates for machines with many qubits, and vendors naturally report the metric on which they lead. There is also no agreed way to compare an analog simulator against a gate-based machine.

Ask for three numbers together and refuse to evaluate on one: two-qubit gate fidelity, connectivity, and gate time. Those determine how deep a circuit can run and how long it takes. Then ask for an algorithmic benchmark on a problem resembling yours, run end to end including compilation and calibration overhead. Treat any comparison that leads with qubit count as marketing. DARPA's Quantum Benchmarking Initiative exists because the field needed independent assessment rather than vendor self-report.

Two-qubit gate fidelity, connectivity, and gate time together determine achievable circuit depth · quantum volume combines width and depth but saturates on wide machines · randomized benchmarking separates gate error from state preparation and measurement error · algorithmic benchmarks include compilation and calibration overhead, which vendor numbers often exclude · no accepted metric compares analog simulators with gate-based machines.

IBM's scale, quality, and speed framing with quantum volume and CLOPS; the QED-C and Metriq benchmark suites; DARPA's Quantum Benchmarking Initiative assessing vendor claims independently; the repeated pattern of quantum advantage claims later matched by improved classical simulation.

10 minThree Metrics for Quantum Computing Performance: Scale, Quality and SpeedIBM Research · 5k+ views

3 minQuantum Benchmarking Initiative - OverviewDARPAtv · 5k+ views

13 minBenchmarking near-term quantum computers*IBM Research · 1k+ views

Benchmarking a trapped-ion quantum computer with 30 qubits (Quantum)