Structural control of two-level defect density revealed by high-throughput correlative measurements of Josephson junctions

This study establishes a high-throughput, data-driven methodology that correlates fabrication parameters and microstructural features across thousands of Josephson junctions to identify specific structural origins of two-level system defects, ultimately achieving a two-thirds reduction in defect density through optimized electrode fabrication.

The Gist

Imagine you are trying to build a super-fast, super-smart computer that uses the laws of quantum physics. This computer, called a quantum processor, is incredibly powerful, but it's also very fragile. It's like trying to balance a house of cards in a hurricane.

The "cards" in this house are tiny circuits made of superconducting materials. The "hurricane" is caused by tiny, invisible defects in the materials themselves. In the world of quantum computing, these defects are called Two-Level Systems (TLS).

Think of TLS as tiny, mischievous ghosts living inside the walls of your computer. Most of the time, they just whisper, causing a little bit of static noise. But sometimes, a ghost gets really close to a "card" (a qubit) and starts shouting. When this happens, the card falls over, and the computer loses its memory. This is called decoherence, and it's the biggest reason why quantum computers are currently so hard to scale up.

For a long time, scientists knew these ghosts existed, but they didn't know where they lived or how to get rid of them. They were like trying to catch a ghost in a dark room without a flashlight.

The Big Breakthrough: A High-Throughput Detective Agency

The researchers in this paper decided to stop guessing and start investigating. They built a massive "detective agency" to find the ghosts and figure out what makes them appear.

Here is how they did it, broken down into simple steps:

1. The Trap (The Resonator)

Instead of looking for ghosts one by one in a single room (a single qubit), they built a giant warehouse with 6,000 tiny rooms (Josephson junctions) all connected together. They tuned these rooms to act like a giant net. If a ghost (TLS) was anywhere in the warehouse, it would get caught in the net and make a distinct "ping" sound. This allowed them to catch hundreds of ghosts at once, rather than just one.

2. The Microscope (The STEM)

Once they caught the ghosts, they needed to see where they were hiding. They took the same chips they used for the experiment and sliced them open to look at them under a super-powerful microscope (called a Transmission Electron Microscope). This let them see the grain structure of the metal—imagine looking at a wall made of bricks. Are the bricks small and jagged? Or are they large and smooth?

3. The Connection (The "Aha!" Moment)

By comparing the "ghost count" (how many TLS they found) with the "brick structure" (the metal grains), they found a massive correlation.

• The Old Way: When they made the metal layers thin, the "bricks" (aluminum grains) were small and jagged. This created lots of cracks and rough edges where the "ghosts" loved to hide.
• The New Way: When they made the metal layers thicker, the "bricks" grew much larger and smoother. This meant fewer cracks and fewer hiding spots for the ghosts.

The Magic Trick: Thicker is Better

The most exciting part of their discovery is a simple rule: Make the metal walls thicker.

When they increased the thickness of the aluminum layers in their circuits, the number of "ghosts" (TLS) dropped by two-thirds.

Think of it like this:

• Thin walls are like a cheap, flimsy fence made of small, broken sticks. A fox (the defect) can easily squeeze through the gaps.
• Thick walls are like a solid, smooth concrete barrier. The fox has nowhere to hide and can't get through.

Why This Matters

Before this paper, scientists thought the only way to reduce these defects was to make the circuits smaller (which is hard and limits how powerful the computer can be). They thought the defects were just a natural, unchangeable part of the materials.

This paper proves that we can control the ghosts. By simply changing how we build the circuits (making the metal thicker), we can significantly reduce the number of defects.

The Bottom Line:
This research gives us a clear blueprint for building better quantum computers. Instead of trying to catch ghosts after they ruin our calculations, we can now build our "houses" in a way that makes it impossible for the ghosts to get in. This is a huge step toward making quantum computers reliable enough to solve real-world problems, from designing new medicines to cracking complex codes.

Technical Summary

1. Problem Statement

Superconducting quantum processors are limited by Two-Level Systems (TLS), which are ubiquitous material defects that couple to circuit modes.

• Impact: Strongly coupled TLS cause excess dielectric loss and decoherence, rendering qubits effectively unusable.
• The Challenge: While TLS are known to exist in amorphous materials and interfaces (specifically the Al/AlOx/Al tunnel barrier in Josephson Junctions or JJs), their microscopic origins remain poorly understood.
• Current Limitations: Existing mitigation strategies (e.g., thermal cycling, strain application) are post-fabrication fixes that increase device complexity and are not scalable. Previous attempts to link fabrication parameters to TLS density have failed due to a lack of large, correlated datasets connecting microstructural features with TLS statistics.

2. Methodology

The authors developed a high-throughput, correlative workflow to bridge the gap between fabrication parameters, microstructure, and TLS density. The approach consists of three integrated stages:

A. Device Design and Fabrication

• Resonator Arrays: Instead of using single qubits (which have small junction areas), the team fabricated frequency-tunable resonators containing arrays of JJs.
  • Each array covers ~100 µm² of junction area (vs. <1 µm² for transmons), increasing the probability of detecting strongly coupled TLS.
  • Eight resonators are frequency-multiplexed on a single chip, allowing data accumulation in a single cooldown.
• Fabrication Variations: Five distinct fabrication treatments were tested:
  • A/A': Standard recipe (Control).
  • B: Post-fabrication annealing (250°C, 20 min).
  • C: Slower aluminum deposition rate (0.6 Å/s).
  • D: Thicker aluminum electrodes (Bottom: 100 nm, Top: 150 nm).

B. Cryogenic Measurement & TLS Extraction

• Automated Spectroscopy: A fully automated system sweeps the resonator frequency via external magnetic flux.
• Detection Algorithm: TLS are identified by analyzing the residuals of the fit between measured data and a standard "hanger" resonator model. Strongly coupled TLS create sharp, localized peaks in these residuals.
• Statistical Inference: To handle false positives/negatives, the team used an empirical-Bayesian framework. They simulated resonator responses with and without TLS to calibrate detection thresholds, then inferred the posterior probability distribution of TLS counts for each device.

C. Microstructural Characterization (STEM)

• Correlative Imaging: Devices from the same fabrication run (identical geometry) were split: one set for cryogenic measurement, the other for Scanning Transmission Electron Microscopy (STEM).
• Metrics Extracted: From 709 atomic-resolution STEM images across 5 devices, the team quantified:
  • Lateral Al grain size ($E_{gw}$).
  • Electrode thickness ($E_t$).
  • Junction roughness and interface morphology.

3. Key Contributions

1. High-Throughput Correlative Dataset: The study assembled a massive dataset correlating TLS statistics from 6,000 JJs with 600+ atomic-resolution STEM images, overcoming the statistical limitations of previous studies.
2. Automated Detection Workflow: They demonstrated a resonator-spectroscopy-based method that detected 393 distinct TLS across 5 samples, significantly outperforming previous qubit-swap spectroscopy methods in efficiency and scale.
3. Identification of Structural Control Knobs: The study successfully identified specific microstructural features (grain size and electrode thickness) that directly correlate with TLS density, moving beyond the previous belief that only junction area matters.

4. Key Results

• Fabrication Impact:
  • Treatments A, A', B, and C showed similar TLS densities ($\rho_{TLS} \approx 0.20 \pm 0.10$ TLS GHz$^{-1}$ µm$^{-2}$).
  • Treatment D (Thicker Electrodes) resulted in a two-thirds reduction in TLS density ($\rho_{TLS} \approx 0.07 \pm 0.04$ TLS GHz$^{-1}$ µm$^{-2}$).
• Microstructural Correlations:
  • Grain Size: Treatment D exhibited significantly larger Al grains (mean ~128 nm) compared to the thinner samples (40–60 nm).
  • Correlation Strength: Statistical analysis (Pearson correlation and Ridge regression) confirmed that mean electrode thickness and mean grain size are the only statistically significant predictors of TLS density.
  • Mechanism: The data suggests that TLS formation is linked to grain boundaries or the AlOx adjacent to them. Thicker electrodes promote larger grains, thereby reducing the density of grain boundaries where defects form.
• Statistical Rigor: The reduction in TLS for Treatment D was confirmed to be statistically significant ($p \leq 0.05$) compared to all other treatments using Kruskal-Wallis tests.

5. Significance

• Paradigm Shift: This work challenges the prevailing view that TLS density in Al/AlOx/Al junctions is an intrinsic, unchangeable limit determined solely by junction area. It proves that fabrication choices (specifically electrode thickness) can actively control defect populations.
• Scalability: A two-thirds reduction in TLS density is critical for scaling quantum processors. Since the probability of a multi-qubit device functioning without errors scales exponentially with the number of qubits, reducing the defect density per qubit dramatically increases the likelihood of building large-scale, functional quantum computers.
• Methodology: The paper establishes a robust, data-driven methodology for "grain engineering" in quantum devices. It provides a blueprint for how to systematically correlate microstructure with quantum performance, paving the way for future optimization of superconducting qubits through materials science rather than just circuit design.