Benchmarking Single-Qubit Gates on a Neutral Atom Quantum Processor

This paper presents a comprehensive benchmarking study of single-qubit gates on a neutral atom quantum processor using Direct Randomized Benchmarking and Gate Set Tomography, achieving an average fidelity of 99.963% and introducing a novel gauge optimization procedure on the Stiefel manifold to ensure physically consistent and comparable results.

The Gist

Imagine you've just built a brand new, incredibly complex robot army. Each robot is a tiny "quantum bit" (or qubit) made of a single atom of Rubidium, floating in a vacuum and held in place by invisible beams of light (like a high-tech version of Star Wars lightsabers).

Your goal is to make these robots perform simple tasks, like flipping a switch (a "gate"). But, just like real robots, they sometimes glitch, get tired, or misread your instructions.

This paper is about two different ways the scientists checked how well their robot army is working and how they fixed the glitches to make them nearly perfect.

The Problem: The "Noisy" Workshop

In the quantum world, things are messy.

1. The Atoms: They are floating in a trap, but sometimes they wiggle too much (noise) or even fall out of the trap entirely.
2. The Measurement: When you ask the robot "Did you flip the switch?", the robot might say "Yes" when it actually said "No," or the robot might have died before you could ask. This is called SPAM (State Preparation and Measurement) error. It's like trying to judge a chef's cooking skills, but your taste buds are broken, or the chef is too shaky to hold the spoon.

The scientists needed a way to measure how good the "flipping" (the gate) really is, without being fooled by the broken taste buds or shaky hands.

The Two Testers: DRB and GST

To solve this, they used two different "testers" (benchmarking protocols).

1. The "Randomized Obstacle Course" (Direct Randomized Benchmarking - DRB)

Imagine you want to test a gymnast's balance. Instead of asking them to do one perfect handstand (which might fail just because they were nervous at the start), you make them run through a long, random obstacle course.

• How it works: You tell the atom to do a random sequence of 100 flips, twists, and turns. Then, you ask it to return to its starting position.
• The Magic: If the atom is perfect, it comes back exactly where it started. If it's noisy, it gets lost. By running this "obstacle course" thousands of times with different random paths, the scientists can see how fast the atoms get "lost" as the course gets longer.
• The Benefit: This method is great because it ignores the "broken taste buds" (SPAM errors). Even if your measurement is a bit off, the rate at which the atoms get lost tells you exactly how bad the flips themselves are.
• The Result: They found their atoms were doing a fantastic job, with a success rate of 99.963%. That's like flipping a coin 10,000 times and only getting it wrong 3 or 4 times.

2. The "Full Medical Scan" (Gate Set Tomography - GST)

While DRB is like checking a car's speed on a track, GST is like giving the car a full MRI and X-ray.

• How it works: GST doesn't just look at the flips; it reconstructs the entire picture. It figures out exactly how the atom starts, exactly how it flips, and exactly how the measurement machine reads the result. It builds a 3D map of every tiny error.
• The Benefit: It tells you why things are going wrong. Is the laser too strong? Is the timing off by a fraction of a second?
• The Catch: It's slow and computationally heavy. It's like doing a full body scan instead of just checking your pulse.

The "Tuning" Fix

Here is the coolest part of the story.

When they first tested a single atom, the "obstacle course" (DRB) showed that while the average performance was good, the atoms were getting confused in a specific way. It was like a robot that was consistently turning left when it should have turned right, or spinning a little too fast.

The scientists realized this was a calibration error. The laser pulses they used to flip the atoms were slightly the wrong length or the wrong angle.

• The Solution: They created a new "tuning knob" system. They ran the DRB test, analyzed the specific pattern of errors, and mathematically calculated exactly how much to adjust the laser.
• The Result: They turned the knobs, re-ran the test, and boom! The fidelity jumped from 99.36% to 99.963%. They fixed the robot's brain without changing the robot itself.

Scaling Up: The 25-Robot Army

Finally, they didn't just test one robot. They tested a whole squad of 25 atoms at the same time.

• The Challenge: Usually, when you control a whole group, some robots are in the "hot zone" and some are in the "cold zone," making them perform differently.
• The Result: They applied the same global control to all 25 atoms. The average performance remained incredibly high (99.946%). This proves that their "lightsaber" control system is strong enough to manage a whole team of robots without any of them getting lost.

The Big Picture

This paper is a victory lap for Neutral Atom Quantum Computing.

1. Validation: They proved that their specific type of quantum computer (using floating Rubidium atoms) can perform single-qubit gates with near-perfect accuracy.
2. Methodology: They showed that using DRB (for quick, reliable checks) and GST (for deep, detailed diagnosis) together is the best way to build and fix quantum computers.
3. Scalability: They showed that you can control a small army of these atoms (25 of them) just as well as a single one, which is a crucial step toward building a massive, useful quantum supercomputer in the future.

In short: They built a fleet of floating atomic robots, figured out exactly how to measure their skills without being tricked by their shaky hands, tuned their controls to make them nearly perfect, and proved that a whole squad of them can work together flawlessly.

Technical Summary

1. Problem Statement

Quantum computers based on neutral atoms (specifically Rubidium-87) are a promising platform for scalable quantum computing. However, the practical deployment of these systems is hindered by decoherence and operational errors. A critical challenge in characterizing these systems is the State Preparation and Measurement (SPAM) error. Traditional tomography methods assume perfect SPAM, which leads to systematic errors in fidelity estimation. Furthermore, as systems scale to larger qubit arrays, benchmarking protocols must remain efficient while distinguishing between stochastic noise (e.g., decoherence) and coherent control errors (e.g., miscalibrated pulse durations or phases). The authors aim to rigorously benchmark single-qubit gates on a neutral atom processor using protocols that are robust to SPAM errors and capable of diagnosing specific error mechanisms.

2. Methodology

The study employs two complementary benchmarking protocols: Direct Randomized Benchmarking (DRB) and Gate Set Tomography (GST).

A. Experimental Platform

• Hardware: A neutral atom quantum processor using $^{87}\text{Rb}$ atoms cooled in a magneto-optical trap (MOT) and loaded into a 2D array of optical tweezers (holographically generated).
• Qubit Encoding: Hyperfine sublevels of the ground state: $|0\rangle = |5S_{1/2}, F=2, m_F=0\rangle$ and $|1\rangle = |5S_{1/2}, F=1, m_F=0\rangle$.
• Control: Single-qubit gates are implemented via global microwave-driven transitions.
• Readout: Site-resolved measurement using a "push-out" beam that ejects atoms in the $|0\rangle$ state, followed by fluorescence imaging.

B. Benchmarking Protocols

1. Direct Randomized Benchmarking (DRB):

   • Procedure: Prepares stabilizer states, applies $m$ layers of randomly selected Clifford gates (generated from $\{R_x, R_y\}$ rotations), and measures in the computational basis.
   • Advantage: Provides an efficient, SPAM-robust estimate of average gate fidelity by fitting the decay of success probability against circuit depth.
   • Calibration Innovation: The authors introduced a two-parameter calibration model embedded within the DRB protocol. This model extracts coherent control errors without auxiliary measurements:
     • $k$: An overrotation factor (due to imperfect $\pi$-pulse duration).
     • $\phi$: An azimuthal offset angle (due to phase misalignment between $R_x$ and $R_y$ gates).
   • Process: DRB is performed iteratively to extract $k$ and $\phi$, which are then used to correct the physical control parameters.

2. Gate Set Tomography (GST):

   • Procedure: A self-consistent protocol that reconstructs the quantum gates, input states, and measurement operators simultaneously. It uses "germs" (short gate sequences repeated multiple times) to amplify and identify specific gate imperfections.
   • Gauge Optimization: The authors developed a novel gauge optimization procedure to resolve the inherent gauge freedom in GST. This involves:
     • Joint diagonalization of reconstructed input states and POVM elements to align them to a canonical frame.
     • Maximizing fidelity between reconstructed gates and ideal targets via a rotation $R_z(\delta)$.
     • Enforcing physical constraints (complete positivity and trace preservation) by performing optimization over the Stiefel manifold.

C. Simulation

Before experimental validation, the authors developed a numerical simulator incorporating realistic noise models (longitudinal $T_1$ and transverse $T_2$ relaxation, and asymmetric SPAM errors) to validate the DRB and GST protocols and their theoretical predictions.

3. Key Contributions

• Comprehensive Benchmarking Framework: Demonstrated the combined use of DRB (for speed and SPAM robustness) and GST (for detailed process reconstruction) on a neutral atom platform.
• Coherent Error Calibration: Introduced a novel, auxiliary-free method to extract and correct coherent control errors ($k$ and $\phi$) directly from DRB data, significantly improving gate fidelity.
• Gauge Optimization for GST: Developed a rigorous gauge-fixing procedure based on joint diagonalization and Stiefel manifold optimization, enabling meaningful fidelity comparisons for reconstructed gates.
• Scalability Validation: Successfully extended DRB to a 25-qubit array under global control, demonstrating that global control does not significantly degrade individual qubit performance compared to isolated qubits.

4. Results

• Simulation vs. Theory: Simulations confirmed that both DRB and GST accurately predict theoretical fidelities across a range of $T_2$ relaxation times and SPAM error rates.
• Single-Qubit Performance (Pre-Calibration): Initial DRB measurements on a single qubit showed an average fidelity of 99.360%, with significant variability between $|0\rangle$ and $|1\rangle$ outcomes, indicating coherent errors.
• Post-Calibration Performance: After applying the two-parameter calibration derived from DRB:
  • The average gate fidelity improved to 99.963% ($^{+0.015}_{-0.013}\%$).
  • SPAM errors were characterized as $p_{0\to1} \approx 1.5\%$ and $p_{1\to0} \approx 11.9\%$ (the latter high value attributed to atom loss during the push-out measurement).
• GST Verification: Independent GST measurements on the same qubit confirmed:
  • Input state fidelity: 99.98%.
  • Gate fidelities: $F_{R_x} \approx 99.94\%$ and $F_{R_y} \approx 99.38\%$ (consistent with DRB, with the lower $R_y$ fidelity attributed to residual coherent miscalibration).
• 25-Qubit Array: DRB was applied to a 25-qubit array under global control.
  • The average fidelity across all sites was 99.946% ($^{+0.002}_{-0.001}\%$).
  • While spatial inhomogeneities were observed, the global control achieved performance nearly identical to isolated qubit control, validating the scalability of the architecture.

5. Significance

This work establishes the viability of neutral atom architectures for high-fidelity, scalable quantum computing. By demonstrating that global control can maintain high fidelity across a 25-qubit array, the study addresses a major bottleneck in scaling neutral atom systems. The development of SPAM-robust benchmarking techniques that can simultaneously diagnose and correct coherent control errors provides a critical toolkit for hardware optimization. Furthermore, the introduction of a gauge-optimized GST framework offers a more rigorous method for characterizing quantum processes, ensuring that fidelity metrics are physically meaningful and comparable across different experimental setups. These results pave the way for more complex quantum algorithms and error correction schemes on neutral atom processors.