High-fidelity geometric quantum gates exceeding 99.9% in germanium quantum dots

AI summary

Overview

Research area: Experimental semiconductor quantum computing — single-hole spin qubits in strained germanium double quantum dots, with a focus on gate fidelity benchmarking (gate set tomography, randomized benchmarking) and noise-robust "geometric" quantum control.

Technical level: Advanced. The paper assumes familiarity with spin qubits, electric dipole spin resonance, Rabi/Ramsey/CPMG measurements, gate set tomography, and geometric phases, though its central claim is stated in accessible terms.

Scope: The paper benchmarks single-qubit gate fidelities of a germanium hole-spin qubit, identifies off-resonance noise as the dominant limit, and demonstrates that non-adiabatic geometric gates (arXiv:2609.33251v1, 27 Sep 2026) reach fidelities above 99% across a wide range of Rabi frequencies, with a maximum surpassing 99.9%.

What This Paper Is About

Spin qubits made from holes in germanium are attractive because they can be driven electrically, but the same strong spin-orbit interaction that enables fast control also exposes them to complex noise, so operations drift and degrade. The authors ask which noise source actually limits standard "dynamical" single-qubit gates, and whether a different control scheme — one that uses geometric phases rather than dynamical phases — can deliver high fidelity that survives that noise without constant recalibration.

Key Contributions

1. A characterized hole-spin qubit in a strained germanium double quantum dot. The device is defined in an undoped strained germanium wafer, read out with a nearby charge sensor using enhanced latching readout, and driven by electric dipole spin resonance at an in-plane field of 1333 mT, with resonance frequencies of 5.60 GHz and 7.53 GHz (g-factors 0.29 and 0.39).

2. A systematic fidelity benchmark of dynamical gates versus Rabi frequency. Using gate set tomography (sequence depth 8, 448 sequences, analyzed with pyGSTi) at a series of Rabi frequencies, the authors show that X/2 and Y/2 fidelities climb toward roughly 99.9% while the I gate saturates at 97.48%, isolating idling/off-resonance noise as the bottleneck rather than Rabi decay.

3. Implementation of non-adiabatic geometric gates (Abelian phase) and their fidelity comparison. Two evolution loops are implemented: "Path1," designed against systematic (Rabi-frequency) noise, and "Path2," designed against off-resonance noise, plus an extended "Path3π" loop. Path2 keeps average gate-set fidelity above 99% across the whole measured Rabi-frequency range, with maximum fidelities of 99.98%, 99.80%, and 99.97% for I, X/2, and Y/2.

4. Experimental demonstration of noise resilience plus a noise model. By deliberately detuning the microwave frequency, geometric X/2 and Y/2 (I) gates stay above 99% fidelity at ±2.5 MHz (±1.2 MHz) detuning, and measured frequency drift over 55 hours stays inside that window; numerical simulation with extracted noise strengths (δf_Larmor = 1.80 MHz, δf_Rabi = 58.48 kHz) reproduces the experimental trends.

Main Findings

• Qubit parameters: Rabi frequency reaches 19 MHz at maximum driving power with a quality factor Q = 39.78 (Q defined as 2·T2^Rabi·f_Rabi); a fit at f_Rabi = 11.56 MHz gives T2^Rabi = 1.56 μs and Q = 18.04. Dephasing time T2* = 136 ns, and CPMG refocusing extends coherence to T2^CPMG = 6.75 μs at Nπ = 230, about 50 times longer than T2*.

• Randomized benchmarking result: The average single-qubit gate fidelity measured by RB is 99.49%, with I, X/2, and Y/2 gates all exceeding 99%. Table 1 lists RB values of 99.34(8)% for the reference, 99.21(3)% for I, 99.63(2)% for X/2, and 99.62(2)% for Y/2.

• Dynamical gates limited by idling: At f_Rabi = 19 MHz, GST gives 97.48% for I, 99.81% for X/2, and 99.88% for Y/2. The I-gate fidelity stays substantially below 99% across the entire Rabi-frequency range and appears to saturate above 15 MHz; the authors attribute this to GST sequences containing idle times much longer than T2*, and note the I-gate fidelity improves by 3.46% as f_Rabi is raised.

• Geometric gates clear 99% everywhere: For Path2, average control fidelity exceeds 99% at every measured Rabi frequency, with maxima of 99.98% (I), 99.80% (X/2), and 99.97% (Y/2) — notably, the I gate improves most at low Rabi frequency, where dynamical idling performs worst.

• Scheme choice matters: Path1, which targets systematic Rabi-frequency noise, yields gate-set fidelities well below 99%, which the authors suggest is because the targeted systematic noise is weaker than the off-resonance noise in this device. Path3π, a 3π loop 1.5 times longer than Path1 and Path2, performs significantly worse than Path2, probably because of its longer operation time.

• Detuning robustness: Geometric X/2 and Y/2 (I) fidelities remain above 99% at detunings of ±2.5 MHz (±1.2 MHz), and the geometric-versus-dynamical fidelity gap widens as |Δf| grows, confirming faster degradation for dynamical gates.

• Cost of robustness: Geometric gates here take four or six times longer than the corresponding dynamical gates, and the paper explicitly frames this as a trade-off between operation time and noise resistance.

• Readout details: Single-shot enhanced latching readout in the (1,1)–(2,0) region, based on Pauli spin blockade and asymmetric dot–reservoir coupling, with a singlet–triplet energy difference of 0.8 meV and a lever arm of 0.13 eV/V.

Methodology in Plain English

The team fabricated a pair of coupled quantum dots in strained germanium and confined single holes in them. A magnetic field splits the spin states, and microwave pulses applied to a gate drive the spin between states via the material's intrinsic spin-orbit coupling. The spin state is read out by exploiting Pauli blockade: whether a hole can tunnel out of the array depends on whether the two-spin state is a singlet or a triplet, which produces a detectable charge signal.

To measure how good the gates really are — separating gate errors from imperfect state preparation and readout — the authors used two complementary benchmarking protocols, randomized benchmarking and gate set tomography, the latter analyzed with pyGSTi. They repeated the GST experiments at several Rabi frequencies so they could see which error channel dominates. Gate set tomography's error analysis split errors into coherent (Hamiltonian-like) and stochastic components; the stochastic, off-resonance part motivated switching to geometric control.

Geometric gates work by steering the qubit around a closed loop in state space in three timed microwave segments with different phases and amplitudes, so the resulting operation depends on the enclosed geometry rather than on the elapsed time. By choosing the phase pattern of the middle segment, the loop can be made insensitive to either amplitude noise (Path1) or detuning noise (Path2). The authors then repeated the fidelity measurements while intentionally shifting the microwave frequency away from resonance to test that insensitivity directly, and cross-checked with a numerical simulation of the GST sequences using extracted noise strengths.

Why This Matters

Impact on research. The paper moves beyond simply reporting a high number: it uses GST to identify a specific noise channel (off-resonance/detuning noise during idling) as the practical limiter for germanium hole-spin gates, then shows an experimentally viable control scheme that tolerates it. It also supplies the missing experimental verification of noise resilience for Abelian-phase non-adiabatic geometric gates in semiconductors, which the authors note had been implemented before but not shown to be robust, with fidelities previously far below fault-tolerant thresholds.

Real-world applications (as framed by the paper):

• Fault-tolerant quantum computing with surface codes, since the reported fidelities for X/2 and Y/2 gates exceed the 99% threshold the paper cites for surface-code error correction.
• Scalable quantum-dot processors built from two-dimensional arrays, where reproducible, uniform operation across many qubits is required.
• Reduced calibration burden in laboratory and systems settings — the tolerance to ±2.5 MHz detuning and 55 hours of drift means fewer recalibration cycles.
• Semiconductor manufacturing pathways, since the device is fabricated on strained germanium with standard gate-defined architectures and involves microelectronics fabrication partners.

Industry relevance. The work sits at the intersection of quantum device physics and semiconductor process technology: the heterostructure was supplied by microelectronics/memory-technology institutes, and one author is affiliated with a quantum computing company. Germanium's low effective mass and CMOS-adjacent fabrication make this a plausible route toward foundry-compatible spin-qubit processors, and calibration overhead is a direct operating-cost issue for any multi-qubit system.

Future Directions

• Shortening geometric gate times. The paper notes a recently proposed noncyclic non-adiabatic geometric gate that shortens evolution time, which would reduce the four-to-six-times time penalty of the current schemes.
• Automatic noise-type selection. Because Path1 and Path2 target different noise, the authors argue for first characterizing the noise (via coherence measurements, noise spectroscopy, GST error reports, or Rabi decay envelopes) and then choosing the scheme, using Path3π or noncyclic protocols when systematic and off-resonance noise are comparable.
• Extension to two-qubit geometric gates. Combined with existing two-qubit geometric proposals, this could enable robust universal geometric quantum computation in a large-scale qubit array — not demonstrated here.
• Scaling and reproducibility. Whether the geometric advantage persists across many qubits with site-dependent, anisotropic noise in dense arrays remains an open experimental question, as does reducing the calibration resources needed at scale.

Target Audience

Experimental quantum-computing researchers working on semiconductor spin qubits and quantum-dot device engineering; quantum control specialists interested in geometric-phase gate schemes and their practical trade-offs; benchmarking methodologists interested in GST versus RB comparisons; and industry or program-level readers evaluating germanium hole-spin technology for scalable, fault-tolerant hardware. Readers without a quantum-information background will find the device and concept descriptions accessible, but the fidelity analysis and geometric-gate formalism require graduate-level familiarity.

Authors’ abstract

Achieving high-fidelity and robust qubit manipulations is a crucial requirement for realizing faulttolerant quantum computation. Here, we demonstrate a single-hole spin qubit in a germanium quantum dot and characterize its control fidelity using gate set tomography. The maximum control fidelities reach 97.48%, 99.81%, 99.88% for the I, X/2 and Y /2 gate, respectively. These results reveal that off-resonance noise during consecutive I gates in gate set tomography sequences severely limits qubit performance. Therefore, we introduce geometric quantum computation to realize noiseresilient qubit manipulation. The geometric gate control fidelities remain above 99% across a wide range of Rabi frequencies. The maximum fidelity surpasses 99.9%. Furthermore, the fidelities of geometric X/2 and Y /2 (I) gates exceed 99% even when detuning the microwave frequency by +-2.5 MHz (+-1.2 MHz), highlighting the noise-resilient feature. These results demonstrate that geometric quantum computation is a potential method for achieving high-fidelity qubit manipulation reproducibly in semiconductor quantum computation.