Long-term stability and oxidation of ferroelectric AlScN devices: An operando HAXPES study

This study utilizes operando hard X-ray photoelectron spectroscopy (HAXPES) to investigate the long-term oxidation mechanisms of tungsten-capped and uncapped AlScN thin films, revealing that oxygen preferentially replaces nitrogen by bonding with scandium and proposing a model to explain these element-specific degradation processes.

The Gist

Imagine you have built a tiny, super-efficient switch for your computer. This switch is made of a special material called Aluminum Scandium Nitride (AlScN). Think of this material as a "smart sponge" that can remember whether it's been squeezed (electrically charged) or not, making it perfect for next-generation memory and processors.

However, there's a catch: this smart sponge is very sensitive to the air around it. Just like a fresh apple turns brown when you slice it and leave it on the counter, this material starts to "rust" (oxidize) the moment it touches oxygen. If it rusts, the switch stops working properly.

This paper is like a detective story where scientists used a super-powered microscope (called HAXPES) to figure out exactly how this material rusts, why it happens so fast, and how to stop it.

Here is the breakdown of their findings using simple analogies:

1. The Problem: The "Scandium" Weak Link

The material is mostly Aluminum Nitride, but scientists added a little bit of Scandium to it.

• The Analogy: Imagine a brick wall made of strong, stable bricks (Aluminum). To make the wall flexible enough to switch states, they swapped some bricks for a special, slightly weaker type (Scandium).
• The Issue: While this makes the wall flexible, those Scandium bricks are like "magnets for rust." When air touches the wall, the oxygen in the air attacks the Scandium bricks first, pulling them apart and replacing the Nitrogen inside with Oxygen.

2. The Investigation: The "X-Ray Flashlight"

The scientists didn't just look at the surface; they used a special X-ray flashlight (HAXPES) that can see deep inside the material, not just the top layer.

• The Experiment: They took samples and left them in the air for two weeks and six months. They also had some samples covered with a protective "helmet" (a thin layer of Tungsten metal).
• The Discovery:
  • The Rust is Deep: Unlike some materials that form a thin, protective layer of rust that stops further damage (self-limiting), this material keeps rusting all the way through. It's like a sponge that keeps soaking up water until it's completely soggy, not just the surface.
  • Scandium is the Culprit: The oxygen loves the Scandium bricks much more than the Aluminum ones. It replaces the Nitrogen next to the Scandium first.
  • The "Ghost" Signal: They found a strange new signal in their data (a peak at 404 eV). They realized this was actually Nitrogen gas (N₂) getting trapped inside the material like bubbles in a cake as the oxygen pushed the nitrogen out.

3. The Model: The "Musical Chairs" of Atoms

The scientists built a simple computer model to explain what was happening.

• The Scenario: Imagine a dance floor where Nitrogen atoms are holding hands with Scandium and Aluminum.
• The Move: When Oxygen enters the room, it wants to dance with Scandium the most because it's the most energetic match. So, Oxygen kicks the Nitrogen out of the Scandium's hand.
• The Result: The Nitrogen that gets kicked out turns into a gas (N₂) and tries to escape. Some of it leaves the building, but some gets stuck in the walls (interstitial sites), creating bubbles that weaken the structure. Because Oxygen loves Scandium so much, it creates a chain reaction that eats deeper and deeper into the material.

4. The "Live" Test: Applying Voltage

The most exciting part was testing the material while it was working (operando).

• The Uncapped Sample: When they applied electricity to the unprotected (rusty) sample, the rusting got much worse, much faster. It's like blowing on a fire; the electricity fanned the flames of oxidation. Even a tiny voltage made it degrade quickly.
• The Capped Sample: The samples wearing their "Tungsten Helmet" stayed perfectly stable, even under high voltage. The helmet acted as an impenetrable shield, keeping the air out completely.

The Big Takeaway

The main lesson from this paper is that AlScN is a brilliant material for future electronics, but it is incredibly fragile when exposed to air.

• Don't leave it out: You cannot just build these devices and leave them sitting in a lab. They need to be built in a vacuum (a sealed, air-free environment) from the bottom up to the top.
• Wear a helmet: The top layer (the electrode) must be a perfect shield (like the Tungsten used here) to stop oxygen from sneaking in.
• Electricity makes it worse: If the shield is broken, applying electricity will make the material rot away almost instantly.

In short: To make these super-fast, memory-saving switches work in the real world, we have to treat them like a delicate flower in a greenhouse—keep the air out, and they will bloom. Let the air in, and they will wither.

Technical Summary

1. Problem Statement

Aluminum scandium nitride (Al$_{1-x}$Sc$_x$N) is a promising material for next-generation ferroelectric devices (e.g., FeRAM, FeFET) due to its large remanent polarization and CMOS compatibility. However, the introduction of Scandium (Sc) to destabilize the polar AlN lattice for switching introduces a critical flaw: severe susceptibility to oxidation.

• While pure AlN is stable up to ~800°C, AlScN oxidizes rapidly even at room temperature.
• This oxidation degrades ferroelectric switching performance and limits device reliability.
• Previous studies suggested oxidation might be "self-limiting," but the microscopic chemical mechanisms and the specific roles of Sc vs. Al in this process remain unclear.
• There is a lack of understanding regarding how oxidation evolves over time (weeks to months) and how it behaves under operational electric fields (in operando).

2. Methodology

The authors employed Hard X-ray Photoelectron Spectroscopy (HAXPES) to investigate the oxidation mechanisms with high chemical state selectivity and significant information depth (ID).

• Sample Preparation:
  • Material: Al$_{0.83}$Sc$_{17}$N thin films (60 nm) grown on p-Si with bottom electrodes (Ti, TiN, or W).
  • Conditions:
    1. Reference: In-situ capped with 3 nm Tungsten (W) to prevent oxidation.
    2. Exposed: Uncapped samples exposed to air for 2 weeks and 6 months.
    3. Operando: Capacitor stacks with "finger" gold electrodes to apply DC voltage during measurement.
• Experimental Setup:
  • Beamline: P22 at PETRA III (DESY, Hamburg).
  • Photon Energies: 2.8 keV and 6 keV, providing information depths of ~9 nm, ~15 nm, and ~18 nm (via angle variation).
  • Analysis: Core level spectra of Sc 2p, N 1s, Al 2s, and O 1s were analyzed using peak fitting to quantify chemical states (Nitride vs. Oxide).
• Modeling: A theoretical atomistic model was developed to predict the ratios of Sc-O, Al-O, and released Nitrogen (N$_2$) based on bond energy differences.

3. Key Contributions

1. Element-Specific Oxidation Mechanism: The study definitively proves that oxidation in AlScN is site-selective. Oxygen preferentially replaces nitrogen atoms bonded to Scandium (Sc-N) rather than Aluminum (Al-N) due to a significantly higher energy gain (207.4 kJ/mol for Sc-N $\to$ Sc-O vs. 133.9 kJ/mol for Al-N $\to$ Al-O).
2. Rejection of Self-Limiting Oxidation: Contrary to previous claims, the study demonstrates that oxidation is not self-limiting. It forms a continuous oxidation gradient extending from the surface into the bulk, eventually affecting the entire layer.
3. Identification of N$_2$ Spectral Feature: The authors identified a specific N 1s spectral feature (~404 eV) corresponding to N$_2$ molecules trapped in interstitial sites within the lattice, directly linking the oxidation process to nitrogen release.
4. Operando Stability Analysis: The study reveals that applying an electric field to an uncapped (oxidized) sample accelerates oxidation, whereas a capped sample remains chemically stable even near dielectric breakdown voltages.

4. Key Results

• Spectroscopic Evidence:
  • Sc-O vs. Al-O: The intensity of Sc-O bonds is disproportionately higher than Al-O bonds in oxidized samples, confirming Sc-driven oxidation.
  • Nitrogen Release: A distinct peak at ~404 eV in the N 1s spectrum appears only in oxidized samples. This is attributed to N$_2$ formed when oxygen displaces nitrogen.
  • Depth Profiling: Angle-dependent scans show that oxide contributions (Sc-O, Al-O) increase with surface sensitivity but persist deep into the bulk, confirming a gradient rather than a thin surface layer.
• Time Evolution:
  • After 2 weeks, early-stage oxidation is visible with Sc-O formation.
  • After 6 months, the oxidation is severe, with significant Sc-O and Al-O content and a clear N$_2$ signature.
• The Oxidation Model:
  • The model successfully predicts the ratio of Al-O to Sc-O based on Sc concentration ($c_{Sc}$) and Sc-oxidation fraction ($c_{ScO}$).
  • It confirms that for every Sc-O bond formed, multiple Al-O bonds are created due to lattice coordination, explaining the observed spectral weights.
  • The model suggests that while most released N$_2$ escapes, a small fraction remains trapped as interstitials.
• Operando Findings:
  • Uncapped Sample: Applying a small voltage (-1.5 V) caused a significant intensification of oxidation signals (increased O 1s, Sc-O, Al-O).
  • Capped Sample (W-capped): Remained chemically stable up to -38 V (near breakdown). No new oxide signals appeared, proving the W capping effectively isolates the AlScN from air and electric-field-induced oxidation.

5. Significance and Implications

• Device Reliability: The findings highlight that the long-term stability of AlScN ferroelectric devices is critically dependent on preventing air exposure during fabrication. The "self-limiting" assumption is incorrect; without protection, the entire device layer will eventually oxidize.
• Process Optimization: To achieve reliable ferroelectric switching, devices require complete in-situ growth (bottom electrode to top electrode) without breaking vacuum.
• Electrode Selection: The choice of top electrode is crucial. Tungsten (W) acts as an effective diffusion barrier, whereas damaged or permeable electrodes lead to rapid degradation.
• Fundamental Understanding: The study provides a validated chemical model for AlScN oxidation, explaining the role of nitrogen vacancies and Sc-selective bonding. This is essential for designing robust ferroelectric memory and logic devices using nitride electronics.

In conclusion, this work establishes that while AlScN is a superior ferroelectric material, its practical application is currently bottlenecked by rapid, non-self-limiting oxidation driven by Scandium. The solution lies in strict in-situ processing and robust capping layers to preserve the material's chemical and functional integrity.