UPenn Turns Diamond Crosstalk Into a Four-Qubit Entangling Gate at Room Temperature

September 14, 2026 – Researchers at the University of Pennsylvania have generated a four-qubit entangled state inside a diamond at room temperature using a single entangling pulse sequence. The parallel operation took 14.8 microseconds with an estimated gate fidelity of 0.92 ± 0.04 – about ten times faster and substantially more accurate than the sequential approach tested in the same register, which took 139.9 microseconds and achieved a gate fidelity of 0.69 ± 0.03. Those figures treat preparation and measurement errors separately. The results were published in Nature Nanotechnology on 14 September 2026.

The bottom line for readers tracking the cryptographically relevant quantum computer timeline: this result improves a local quantum-control operation inside one small register. It provides no basis for changing a cryptanalytic threat forecast. Its near-term relevance is to quantum sensing and error-correction circuit design.

The team – Joseph D. Minnella, Mathieu Ouellet, Amelia R. Klein, and Lee C. Bassett of UPenn’s Quantum Engineering Laboratory – used a nitrogen-vacancy (NV) center in diamond as their platform. An NV center is a point defect formed when a nitrogen atom replaces a carbon atom in diamond’s crystal lattice and an adjacent site remains vacant. The defect’s electronic spin can be prepared, controlled, and read out using light and microwave pulses at room temperature.

The electronic spin served as the central qubit. Three nearby carbon-13 nuclear spins served as memory qubits. The researchers created a Greenberger–Horne–Zeilinger (GHZ) state – a form of multipartite entanglement linking the electronic spin and three nuclear spins into a single collective quantum state.

Standard practice in diamond quantum registers creates multipartite entanglement through sequential two-qubit gates: the central electron is entangled with one nuclear spin at a time. Each additional gate adds time and accumulates errors. The UPenn team took a different approach. In an NV center, the electron continuously interacts with the nuclear spins around it. Those interactions – normally treated as crosstalk to be suppressed – were tuned into parallel conditional gates by adjusting the timing and repetition count of an XY8 dynamical decoupling pulse sequence.

The technique builds on an entanglement framework developed by Evangelia Takou, Edwin Barnes, and Sophia Economou, who showed theoretically how interactions that complicate selective control can also support parallel entanglement when pulse parameters are chosen appropriately (PRX 13, 011004, 2023; Quantum 8, 1304, 2024). The UPenn contribution is the experimental implementation, calibration, and benchmarking in an actual room-temperature register.

The result: a single control sequence that entangled all three nuclear spins with the electron simultaneously, producing a four-qubit GHZ state near the fundamental speed limit set by the hyperfine coupling strengths in this register.

The parallel four-qubit gate achieved 0.92 ± 0.04. The sequential four-qubit gate reached 0.69 ± 0.03. For three-qubit parallel gates – tested across different combinations of nuclear qubits – the average fidelity was 0.83, compared with 0.78 for sequential gates, with a roughly two-fold speed improvement, though performance varied across qubit subsets.

The team verified the entangled states by measuring multiple quantum coherences (MQC), a phase-amplification technique in which the oscillation frequency of the read-out signal reveals how many qubits share the entangled state. The parallel gate produced the frequency expected for three entangled nuclear spins, 3.0 ± 0.1; the sequential gate gave a shifted 2.6 ± 0.2, which the authors attribute to its lower fidelity. Because only the electron is read out, the method establishes the size of the entangled state but not its fidelity. The team also performed repeated gate applications to separate gate errors from state-preparation-and-measurement (SPAM) errors – a distinction that matters, because the room-temperature SPAM fidelity remains a significant limitation of the platform.

To assess how broadly the technique could apply, the researchers simulated 500 random, weakly coupled NV registers. Weakly coupled registers are the less common case: about 64% of randomly generated configurations contained at least one strongly coupled nucleus and were set aside, although the authors note such mixed registers could potentially be addressed as well. Among the weakly coupled registers, 87% supported a parallel gate on two nuclear spins and 57% on three, while 15% supported four and 1.4% five. The authors treat these figures as a lower bound, because they deliberately ran a conservative search for gates.

The authors noted that the same principles could extend to other color centers in diamond – such as the silicon- and tin-vacancy (SiV and SnV) centers – and to spin registers in silicon carbide, as well as in silicon. Isotopic engineering of the host crystal could further improve the likelihood of finding suitable nuclear spin arrangements.

The preprint first appeared on arXiv on 11 August 2025. Experimental datasets are openly archived on Zenodo, data-processing code is publicly available on Zenodo and GitHub. The work was funded primarily by the U.S. National Science Foundation (ExpandQISE OSI-2427091).

My Analysis

Parallel control of a coupled register

I find the engineering decision in the UPenn result more interesting than its qubit count or its speed: the group built the gate from the register’s internal couplings instead of suppressing them.

Many quantum platforms fight crosstalk. In superconducting processors, frequency collisions between neighboring transmons are a persistent source of gate errors. In trapped-ion systems, off-resonant coupling to spectator ions limits fidelity as the chain grows. In diamond NV registers, the electron’s hyperfine interactions with surrounding nuclear spins are the local version: when you try to control one nuclear qubit, the others feel it.

The standard solution is suppression – dynamical decoupling sequences that refocus unwanted interactions while selectively driving the target qubit. The UPenn group kept the same dynamical decoupling framework (the XY8 sequence is a workhorse in NV-center control) but worked in the region the field usually overlooks: the first-order resonances, where the nuclear spins’ signals overlap with each other and with the surrounding spin bath. There, instead of one intended gate and two side effects to cancel, they found a single pulse spacing and repetition count that rotates all three nuclei, each conditioned on the electron’s state, in one pass.

The experimental contribution – putting the Takou-Barnes-Economou theoretical framework to work and measuring what happens in a real register at room temperature – is a useful addition to the NV control toolkit. The authors expect it to carry over to other central-spin systems in which an electron couples to surrounding nuclear spins, and it trades flexibility in gate choice for speed in producing specific high-value states like GHZ.

Implications for quantum sensing

The near-term payoff of faster GHZ-state preparation in diamond is in sensing.

NV centers are established tools for nanoscale magnetic sensing and are actively used in condensed-matter research, materials characterization, and biomagnetism. Entangled multi-qubit states such as the GHZ state can improve a sensor’s sensitivity beyond the standard quantum limit under the right conditions. Prior work has already demonstrated room-temperature NV-based phase sensing beyond the standard quantum limit (Xie et al., Science Advances, 2021).

The constraint is operational overhead. Decoherence eats into the sensor’s useful measurement window. Room-temperature electron coherence under the dynamical decoupling sequences that the gates actually use can reach hundreds of microseconds to roughly a millisecond – far longer than the unprotected dephasing time of a few microseconds, but still shorter than the coherence of the nuclear spins. Every microsecond of entangling-gate duration is a microsecond not spent measuring the target signal.

Cutting the entangling-gate duration from 139.9 to 14.8 microseconds is operationally meaningful. And the gate fidelity improvement from 0.69 to 0.92 matters equally: a poorly prepared GHZ state does not deliver its promised sensitivity advantage.

Whether this faster gate translates into better sensitivity per unit time in a complete sensing experiment remains to be shown. The gate duration is one component of the full measurement cycle; initialization, readout, coherence, and how the target signal couples to the electron and nuclear spins all contribute. The UPenn result reduces one part of that overhead. Demonstrating the end-to-end sensing improvement is the next step.

From local registers to larger systems

For readers tracking the path to a cryptographically relevant quantum computer, the direct answer: NV diamond is not the primary path to that destination, and this result does not change the assessment.

The register in this experiment contained four qubits – one electron and three nuclear spins – in a single NV center. The researchers’ own simulations suggest that parallel gates involving four or five nuclear spins (five or six total qubits) are possible in some weakly coupled configurations. A 2019 result from the Taminiau group at TU Delft demonstrated a ten-qubit solid-state spin register – one electron spin and nine nuclear spins (one nitrogen, eight carbon-13) – with single-qubit quantum memory exceeding 75 seconds, at cryogenic temperature (3.7 K).

Those are the demonstrated scale numbers for a single NV center. The registers are small, and extending them depends on coupling range, spectral crowding, and control-protocol capabilities rather than on a fixed physical ceiling – but the practical constraints are severe.

One major route to connecting NV registers into a larger system is photonic networking – entangling distant NV electrons via photon interference. QuTech’s research program has demonstrated entanglement between NV centers separated by 1.3 kilometers and multi-node quantum networks. More recently, a May 2026 Nature Communications paper demonstrated an unconditional teleported CNOT gate between remote diamond registers. These are small experimental results on the path to distributed quantum computation, not operational processors – but they demonstrate that the architecture is being actively pursued.

The scaling constraints remain formidable. Photonic-link fidelity and successful-entanglement throughput are major bottlenecks. The consequences depend on memory performance, error-correction overhead, and communication budgets that no current NV system has demonstrated at scale.

For a cross-platform sense of where NV registers stand: a 2025 study from Jörg Wrachtrup’s group at the University of Stuttgart estimated a quantum volume of 8 for a room-temperature NV-center quantum register, using an experimentally calibrated error model on a three-nuclear-qubit register with all-to-all connectivity. QV = 8 is the width limit of a three-qubit computational register – reaching it says the gates are good, not that the platform cannot grow. But the comparison with leading platforms is stark: Quantinuum’s trapped-ion H2-2 system reported a quantum volume of 33.5 million (2²⁵) in September 2025. QV penalizes low qubit count heavily, and NV registers have excellent connectivity, so the numerical ratio overstates the per-gate quality gap. It does, however, illustrate the scale gap.

The UPenn parallel gate makes each small register faster and more accurate. That is valuable for sensing and for small-scale quantum error correction demonstrations, and useful as a building block in distributed architectures. It does not address the fundamental challenge of scaling NV systems to the large, fault-tolerant processors required for practical cryptanalysis – which current resource estimates place at hundreds of thousands to millions of physical qubits depending on the architecture and error rates assumed.

What is genuinely new and what is incremental

The result deserves recognition on its own terms. The authors demonstrated a four-qubit entangling gate operating near the fundamental speed limit set by the physical interactions in this register – not merely faster than before, but close to as fast as the physics allows. The fidelity improvement from 0.69 to 0.92 for the gate operation is substantial by any platform’s standards. The experimental datasets and processing code are openly published on Zenodo and GitHub – a standard of transparency that strengthens the result.

What the experiment does not demonstrate: a scalable quantum computer, a complete error-correction cycle, fault-tolerant operations, or any application algorithm. The supplementary materials discuss and simulate quantum error correction, but the paper reports no experimental QEC cycle. The paper’s own conclusion is more optimistic than a strict reading of the data would support – the authors write that the speedup “opens the door to new quantum error correcting possibilities, to the point that fault tolerance could be achieved at room temperature.” That is a statement of direction, not a demonstrated result.

The SPAM errors at room temperature are significant and unresolved. The supplementary materials report a fitted four-qubit SPAM fidelity of 0.32 ± 0.01 – a number that matters for any real application built on these gates. The authors identify several paths forward: repeated readout at high magnetic fields, spin-to-charge conversion, and dynamic nuclear polarization. Operating at cryogenic temperatures would also improve measurement fidelity, at the cost of the platform’s room-temperature advantage.

The broader picture for NV diamond

NV-center technology in 2026 occupies an unusual position in the quantum landscape. The platform has found use in both quantum sensing and early-stage quantum computing – companies like Quantum Brilliance and SaxonQ have shipped room-temperature NV systems, though with limited entangling register sizes per core. QuTech and TU Delft have pushed toward higher-fidelity NV registers using cryo-CMOS controllers, reporting gate fidelities of 99.3% (electron) and 99.8% (nuclear) at cryogenic temperatures earlier this year. Color-center groups are also developing SiV and SnV centers in diamond, alongside divacancies in silicon carbide and T-centers in silicon.

The strategic value of color centers spans quantum networking, quantum sensing, and – through distributed architectures – potentially quantum computation. As I assess in my CRQC Quantum Capability Framework, the primary paths to a cryptographically relevant quantum computer run through platforms that can manufacture and control large numbers of physical qubits at scale: superconducting circuits, trapped ions, neutral atoms, and potentially silicon spins. NV diamond contributes to that broader ecosystem as a quantum memory, networking node, and sensing platform, and distributed-computing research is exploring whether small optically connected registers can add up to something larger.

The UPenn result makes the local registers faster and more accurate. That is a real engineering advance in a platform with multiple near-term applications. It supplies no basis for revising a cryptanalytic threat forecast. Its contribution is a faster local entangling operation that future sensing, error-correction, and modular-computing experiments can build on.