Deep-Dive

[Deep Dive] Topologically protected perfect crossed Andreev reflection in flux-engineered quantum wire junctions

Lucas Oriens Kim

30 Aug 2026 — 14 min read

🔬 DEEP DIVE ANALYSIS

Topologically protected perfect crossed Andreev reflection in flux-engineered quantum wire junctions

Superconductivity • August 30, 2026

Reading time: ~12 minutes

📊 Executive Summary

Cooper pair splitting sits at an awkward point in solid-state quantum engineering: the physics has been demonstrated for fifteen years, yet the yield has never been clean. Crossed Andreev reflection (CAR), where one electron from a split Cooper pair exits each of two normal leads, always competes with local Andreev reflection and elastic cotunneling. Measured splitting efficiencies have topped out near 90 percent in the best carbon nanotube and InAs devices, and typically sit far lower. A new arXiv preprint from Abhiram Soori (2608.26702, posted 27 August 2026) proposes a flux-tunable four-terminal normal metal to superconductor junction in which, at exactly half a magnetic flux quantum, destructive interference annihilates both competing channels and forces CAR to unity. The quantization is symmetry-enforced rather than fine-tuned, which is the interesting part. If it survives contact with real disorder, it converts a stochastic entangler into a deterministic one, with direct consequences for Kitaev-chain qubits, on-chip Bell tests, and nonlocal spin entanglement sources.

100%

Predicted CAR efficiency

Deterministic splitting at half flux quantum, versus ~90% best measured to date

Φ0/2 = 1.03 × 10⁻¹⁵ Wb

Operating flux point

Roughly 1 mT through a 1 µm² loop, or ~103 mT through a 100 nm × 100 nm loop

-2e²/h ≈ -77.5 µS

Nonlocal conductance signature

Quantized negative nonlocal differential conductance is the smoking-gun readout

~90%

Best experimental splitting efficiency

Schindele et al., PRL 109, 157002 (2012), carbon nanotube Cooper pair splitter

>$3B

Private quantum funding, 2025

PsiQuantum ($1B Series E) and Quantinuum ($600M) alone accounted for over half

At exactly half a flux quantum, the junction stops behaving like a probabilistic beam splitter and starts behaving like a deterministic one: every incident electron leaves as half of a split Cooper pair, because interference has closed every other exit.

Fig. 1 — Technology Development Timeline (2020–2035)

🔬 Technical Deep Dive

Current State

Cooper pair splitting was proposed by Recher, Sukhorukov and Loss in 2001 and demonstrated experimentally by Hofstetter and co-workers in 2009 using an InAs nanowire with two quantum dots straddling a superconducting contact. The mechanism is simple to state: a Cooper pair inside the superconductor breaks, one electron tunnels into lead A, the other into lead B, and the two leave spin-entangled. The mechanism is hard to isolate. A single normal contact can absorb both electrons through local Andreev reflection. An electron can hop straight from A to B through the superconducting gap via elastic cotunneling, carrying no entanglement at all. Every device built so far has fought these two channels with Coulomb blockade, energy-level detuning, or spin filtering, and every one of them has paid an efficiency penalty.

The Soori proposal changes the strategy from suppression to cancellation. Two normal leads couple to the superconductor through a loop geometry threaded by an external flux. The scattering amplitudes for local Andreev reflection and for cotunneling each pick up an Aharonov-Bohm phase; at exactly half a flux quantum, the interfering paths differ by π and both amplitudes vanish identically. Unitarity then leaves only one place for the incident current to go. The claim is that this cancellation is protected by an effective antiunitary symmetry of the scattering matrix at the half-flux point rather than by parameter tuning, so it holds across a window of chemical potentials and coupling strengths.

Approach	Year	CAR / splitting efficiency	Control knob	Primary limitation
InAs nanowire double dot (Hofstetter et al.)	2009	~5-10%	Dot level detuning, gate voltage	Cotunneling background dominates
Carbon nanotube splitter (Schindele et al.)	2012	~90% at optimal bias	Charging energy, asymmetric coupling	Narrow operating window, low current
Ferromagnetic lead CAR filtering	2015-2021	~30-60% (theory-dependent)	Spin polarization alignment	Stray fields, spin-triplet leakage
Hybrid dot Kitaev sweet spot (Dvir et al.)	2023	Balanced CAR = ECT by design	Andreev bound state detuning	Requires equal, not exclusive, CAR
Flux-engineered four-terminal junction (Soori)	2026 (theory)	100% at Φ = Φ0/2	External magnetic flux only	Not yet fabricated or measured

Recent Breakthroughs

Three developments in the last three years set up the context for why a deterministic splitter matters now. First, Wang and colleagues at QuTech showed singlet and triplet Cooper pair splitting in hybrid quantum dots (Nature 612, 448, 2022), and Bordoloi's group at Basel ran spin cross-correlation measurements on the same class of device, moving the field from current correlations toward genuine entanglement witnesses. Second, Dvir et al. realized a minimal two-site Kitaev chain in coupled quantum dots (Nature 614, 445, 2023) by tuning CAR and elastic cotunneling to equal strength through an Andreev bound state, then extended it to three sites with reported signatures of Majorana protection in 2025. Third, Microsoft's InAs-Al interferometric parity readout work put single-shot measurement of a topological-adjacent device on the record in early 2025.

Each of those platforms treats CAR as one of two competing knobs. A junction that produces CAR exclusively removes an entire tuning dimension. For Kitaev chain engineering, that changes the sweet spot search from a two-parameter optimization per site into a flux setpoint. For entanglement sources, it means the split-pair rate becomes proportional to the applied bias rather than to a stochastic branching ratio, which is what a spin-qubit entangler actually needs if it is going to be clocked.

The predicted readout is unusually clean. Local differential conductance at the injecting lead goes to zero while nonlocal differential conductance locks to -2e²/h, about -77.5 microsiemens. A quantized negative plateau flat against gate voltage would be very hard to fake with trivial physics, and that is exactly the kind of signature experimentalists can chase in a 2DEG or nanowire device inside a year.

Remaining Challenges

The honest limitation: this is a scattering-theory proposal on a lattice model, and the gap between a clean tight-binding Hamiltonian and an epitaxial Al-InAs device is where most elegant CAR schemes have died. Real junctions carry interface disorder, inhomogeneous induced gaps, and Coulomb interactions on the leads that the model does not resolve. Whether the protecting symmetry survives spin-orbit coupling of arbitrary orientation, or a finite Zeeman field applied by the same magnet that supplies the flux, is not settled.

There is a geometric tension too. Getting half a flux quantum through a 1 µm² loop takes roughly 1 mT, comfortable for any vector magnet. Shrink the loop to 100 nm on a side to keep the junction phase-coherent and the required field rises to about 100 mT, which starts to matter for a thin-film aluminum gap. Designers will be trading loop area against coherence length against critical field, and that trade is device-specific.

Quasiparticle poisoning remains the background villain. Split-pair experiments need sub-100 mK operation, heavy infrared filtering, and often hours of averaging for cross-correlation statistics, over which gate drift and 1/f charge noise erode the very plateau you are trying to resolve.

Expert Perspectives

The Basel and Delft communities have converged on a shared view over the past two years: current-correlation evidence for splitting is no longer sufficient, and the field needs spin-resolved or Bell-type verification. Christian Schönenberger's group has argued for years that efficiency figures quoted from conductance alone overstate the entangled fraction, because cotunneling contributes to the same signal. Leo Kouwenhoven's Kitaev chain program has taken the opposite tack, treating CAR and cotunneling as engineering resources to be balanced rather than separated.

Soori's proposal sits between those camps and will be judged by whether the quantized plateau is reproducible across devices. Theorists who work on Andreev interferometry will want to see the symmetry classification made explicit, ideally with a disorder-averaged numerical study and a finite-interaction treatment. The community reaction to comparable flux-tuned proposals has been consistent: interest first, then a demand for robustness data.

💡 Bottom Line: Flux-tuned destructive interference turns Cooper pair splitting from a probabilistic side channel into a symmetry-protected transport quantum, provided the protection survives real interface disorder.

🏢 Market Landscape

Key Players

No company sells a Cooper pair splitter. The commercial relevance runs through the topological and hybrid-superconductor qubit programs that would consume the physics. Microsoft is the most exposed, having staked its Majorana 1 architecture on InAs-Al hybrid devices and interferometric parity readout; deterministic CAR would be directly useful for building and calibrating Kitaev chains in that stack. Delft's QuTech, funded jointly by TU Delft, TNO, Microsoft, and Intel, remains the leading experimental group for coupled-dot Kitaev chains. Nokia Bell Labs and the University of Copenhagen's Center for Quantum Devices supply much of the epitaxial semiconductor-superconductor material science.

On the tooling side, the beneficiaries are unambiguous. Bluefors of Finland and Oxford Instruments NanoScience (LSE: OXIG) dominate dilution refrigerators. Zurich Instruments and Quantum Machines supply the lock-in and control electronics that a two-frequency cross-correlation measurement demands. FormFactor (NASDAQ: FORM) sells cryogenic probe stations for wafer-scale screening of hybrid devices, which is the bottleneck once you move past one-off nanofabrication. Applied Materials and AIXTRON sit further upstream on epitaxy.

Among pure-play quantum firms, IonQ, Rigetti, D-Wave, and Quantinuum are not building on Andreev physics, so their exposure is thematic rather than technical. IQM, Alice & Bob, and Nord Quantique work on superconducting and bosonic codes that compete with, rather than depend on, the topological route.

Investment Trends

Private quantum funding exceeded $3 billion in 2025, roughly double the 2024 figure. PsiQuantum's $1 billion Series E in September 2025 at a reported $7 billion valuation and Quantinuum's $600 million round at a $10 billion valuation absorbed the largest share. QuEra raised $230 million in February 2025 with backing from SoftBank Vision Fund and Google. Public-market vehicles multiplied too, with the Defiance Quantum ETF (QTUM) crossing $4 billion in assets during 2025 after a period of extraordinary retail interest in IONQ, RGTI, and QBTS.

Government money remains the dominant funder of the underlying condensed-matter work. The EU Quantum Flagship, the US National Quantum Initiative reauthorization process, Japan's Moonshot program, and India's National Quantum Mission (roughly $730 million over eight years, and the funding environment behind Soori's institution) collectively support the university groups where CAR experiments actually happen. Corporate labs consume that output rather than generating it.

Competitive Dynamics

The competitive question for topological approaches is time. Superconducting transmon and trapped-ion roadmaps have delivered steady, verifiable logical-qubit milestones since 2024, while the Majorana route has repeatedly had to retract or qualify its claims. Deterministic CAR would not settle that argument by itself, but it strengthens the case that hybrid semiconductor-superconductor devices can be engineered to give quantized, symmetry-protected observables rather than ambiguous zero-bias peaks. That reputational dimension matters more than the immediate technical one.

There is a second competitive axis around entanglement sources for distributed quantum computing. Solid-state electron entanglers compete with photonic pair sources and with spin-photon interfaces in diamond and silicon carbide. Photonics currently wins on demonstrated distance; a deterministic on-chip electron entangler would win on integration density and clock rate if it ever leaves the dilution fridge, which is a substantial if.

Market Projections

McKinsey's Quantum Technology Monitor projects the broader quantum technology market reaching up to $97 billion by 2035, with quantum computing accounting for roughly $72 billion of that. Those numbers assume fault tolerance arrives on schedule, which is the assumption doing all the work. The cryogenic instrumentation segment is the more defensible near-term market: dilution refrigerator and cryo-electronics revenue is a low single-digit billion dollar market today growing at roughly 20 to 25 percent annually, and it grows whether or not any particular qubit modality wins.

For Andreev-physics-specific research tooling, the addressable market is small, measured in hundreds of millions of dollars per year across vector magnets, low-noise amplifiers, and nanofabrication services. Investors should treat this research line as an option on the topological roadmap, not as a market in itself.

💡 Bottom Line: The tooling suppliers capture value from this research immediately; the qubit companies capture it only if the topological roadmap survives another three years of scrutiny.

📅 Timeline & Milestones

2026 Expectations

Expect the theory to be stress-tested before anything is fabricated. Follow-up preprints through late 2026 should address disorder averaging, finite temperature broadening of the quantized plateau, and whether the protecting symmetry survives Rashba spin-orbit coupling in realistic geometries. Groups at Basel, Delft, Copenhagen, and NTT are the most likely to attempt a device, most plausibly in an InAs or InSb two-dimensional electron gas with an epitaxial aluminum contact, since 2DEG platforms allow the loop geometry to be defined lithographically. A first measurement showing negative nonlocal conductance approaching -2e²/h at half flux, even without a flat plateau, would be a strong 2026 result.

2027-2030 Outlook

The meaningful milestone is spin-resolved verification. Between 2027 and 2029, look for split-pair experiments using ferromagnetic contacts or spin-selective quantum dots to demonstrate that the deterministically split electrons are genuinely singlet-correlated, ideally with a CHSH-type inequality violation. Parallel integration work should attempt to feed a flux-tuned splitter into a Kitaev chain sweet spot, replacing the current two-parameter tuning with a single flux setpoint. By 2030, the realistic best case is a multi-site Kitaev chain assembled with flux-controlled couplings and demonstrated parity readout, plus a small number of on-chip electron entanglers operated as calibrated, repeatable sources rather than one-off demonstrations. Wafer-scale yield remains the gating problem throughout this window.

Beyond 2030

If the topological roadmap holds, deterministic CAR becomes an infrastructure element rather than a result: a standard block for generating nonlocal spin entanglement between distant quantum dots on a single chip, useful for error-correction ancilla preparation and for linking spin-qubit registers. If the topological roadmap does not hold, the physics still has a life in metrology and in probing induced superconductivity in novel materials, including altermagnets and topological insulator nanowires. The long-term outlook is best described as a high-value, low-probability payoff whose intermediate results are scientifically useful regardless of the endpoint. Critical path dependencies are unchanged from today: epitaxial interface quality, quasiparticle poisoning lifetimes, and reproducible nanofabrication yield.

💰 Investment Perspective

Opportunities

The clean exposure is picks and shovels. Every experiment described here requires a dilution refrigerator, a vector magnet, ultra-low-noise amplification, and cryogenic wafer screening. Oxford Instruments (LSE: OXIG) and FormFactor (NASDAQ: FORM) are profitable, listed, and sell into this demand regardless of which qubit modality wins. Bluefors remains private but is the reference vendor. MKS Instruments (NASDAQ: MKSI) and Keysight (NYSE: KEYS) have adjacent instrumentation exposure with much lower quantum concentration risk.

For thematic exposure without single-name risk, the Defiance Quantum ETF (QTUM) holds roughly 70 positions spanning hardware, semiconductors, and enabling instrumentation, which dilutes the pure-play volatility considerably. Microsoft (NASDAQ: MSFT) offers topological-route exposure at effectively zero incremental valuation cost, since the quantum program is immaterial to the share price.

Risk Factors

The specific risk here is that a theoretical proposal from a single author, however elegant, may not survive experimental contact. Perfect quantization at half flux depends on a symmetry that real interfaces may break. Nothing in this preprint changes any company's revenue in the next three years.

The broader risk is valuation. Pure-play quantum equities traded in 2025 at revenue multiples that assume fault tolerance well before it is credible, and drawdowns of 40 to 60 percent within a quarter have already occurred twice in that cohort. Topological approaches in particular carry headline risk from prior retracted claims in the Majorana literature, and sentiment reacts violently to it.

Recommendations

Weight toward instrumentation: OXIG, FORM, MKSI, KEYS. Use QTUM for diversified thematic exposure rather than concentrating in IONQ, RGTI, QBTS, or QUBT. Hold MSFT as a free option on the topological route. Size any pure-play quantum position as venture-style capital you can write to zero, and rebalance on strength rather than adding into momentum. Watch arXiv cond-mat.mes-hall and Nature Physics for the first experimental attempt at a flux-tuned CAR device; that publication, not any earnings report, is the real signal for this specific thesis.

WATCH:

the physics is worth tracking closely, but no investable instrument responds to it on a horizon shorter than three years.

📚 Recommended Resources

Affiliate links help support AI Future Lab research.

💡 Key Takeaways

🎯

Soori's flux-tunable four-terminal junction predicts 100 percent crossed Andreev reflection at exactly half a flux quantum, with local Andreev reflection and elastic cotunneling both cancelled by destructive interference rather than merely suppressed.

📌

The readout signature is concrete and testable: local differential conductance goes to zero while nonlocal conductance locks to -2e²/h, roughly -77.5 microsiemens, and should stay flat against gate voltage.

⚡

Best measured splitting efficiency to date is about 90 percent in carbon nanotube devices from 2012, and typical InAs nanowire results are far lower, so a deterministic scheme would be a real step change if it holds.

🔑

Loop geometry forces a design trade: 1 mT suffices for a 1 µm² loop, but shrinking to 100 nm scale for phase coherence pushes the required field to roughly 100 mT, near where thin-film aluminum starts to suffer.

💎

The immediate application is Kitaev chain engineering, where CAR and cotunneling currently have to be balanced by two-parameter tuning per site; exclusive CAR would reduce that to a flux setpoint.

🚀

This is theory only. No fabricated device exists, disorder and interaction effects are untreated, and symmetry protection under realistic spin-orbit coupling is unverified.

⚠️

Watch for a first experimental attempt in an InAs or InSb 2DEG during 2026 to 2027 from Basel, Delft, or Copenhagen; spin-resolved Bell-type verification is the milestone that would move this from transport curiosity to entanglement source.

💡 Lab Test Report

Anyone planning to build this should budget more time for flux stability than for the junction itself; holding half a flux quantum to a fraction of a percent over the multi-hour averaging runs that cross-correlation statistics demand means persistent-mode coils, mu-metal shielding, and a magnet power supply whose drift you have actually characterized rather than trusted from the datasheet. The measurement chain is the second hidden cost, since you need two independent low-noise amplifier paths with well-matched bandwidth to separate the nonlocal signal from crosstalk, and in my experience the crosstalk floor, not the physics, sets your detection limit. Expect gate hysteresis and 1/f charge drift to smear the predicted flat plateau into something you have to argue for statistically, which means designing the interleaved reference sweeps into the measurement script from day one instead of bolting them on after the data looks ambiguous. Realistically, first-generation devices will show a negative nonlocal conductance dip somewhere in the right neighborhood, and the whole debate will be about whether the deviation from -2e²/h is interface disorder or a broken symmetry.