Germanium-tin (GeSn) avalanche photodiode with up to 2.7 micro cutoff wavelength for extended SWIR detection

This paper experimentally demonstrates a CMOS-compatible GeSn-on-Si avalanche photodiode with a thin 122-nm Ge buffer that enables high-Sn-content growth (up to 12.7%), achieving extended SWIR detection with a 2.7 μm cutoff wavelength and high avalanche gain at cryogenic temperatures.

The Gist

Imagine you are trying to take a picture of something far away, but the air is filled with thick fog, dust, or smoke. Normal cameras (which see visible light) get blinded by this. To see through the haze, you need a special kind of "night vision" that uses Short-Wave Infrared (SWIR) light. This is like using a flashlight that shines a color of light our eyes can't see, but which passes right through fog and smoke.

This paper is about building a super-sensitive "eye" for these special cameras, specifically one that can see even further into the infrared spectrum (up to 2.7 micrometers). This is crucial for things like self-driving cars, military surveillance, and medical imaging.

Here is the story of how they built this new eye, explained with some everyday analogies:

1. The Problem: The "Mismatched Puzzle"

To make these detectors, scientists usually stack layers of different materials on top of a silicon chip (the same stuff computer chips are made of).

• The Goal: They wanted to use a material called Germanium-Tin (GeSn) because it's great at absorbing infrared light.
• The Issue: Think of the silicon chip as a wooden floor and the GeSn layer as a heavy carpet. If the carpet is too big for the floor, it bunches up and creates wrinkles (cracks). In science, these wrinkles are called "defects."
• The Old Solution: Usually, to fix this, scientists put a very thick "buffer" layer (like a thick, soft underlay) between the floor and the carpet to smooth out the wrinkles. But this paper's team wanted to use a very thin underlay to make the device work better electrically.
• The Risk: Using a thin underlay usually means the carpet bunches up badly, ruining the device.

2. The Breakthrough: The "Magic Stretch"

The team took a risk. They used a GeSn layer with a very high amount of Tin (12.7%) and placed it on a surprisingly thin Germanium buffer (only 122 nanometers thick).

• The Analogy: Imagine stretching a rubber band. When you stretch it tight, it changes shape.
• What Happened: Because the buffer was so thin, the GeSn layer was forced to stretch and relax in a unique way. Instead of bunching up and breaking, this "stretching" actually helped the Tin atoms fit in better than anyone expected.
• The Result: They ended up with a material that could detect light at 2.7 micrometers. That is a huge leap forward, allowing the "eye" to see much further into the infrared spectrum than previous versions.

3. The Device: The "Snowball Effect" (Avalanche Photodiode)

The device they built is called an Avalanche Photodiode (APD).

• How it works normally: When a photon (a particle of light) hits the detector, it knocks loose one electron. That's a tiny signal.
• The Avalanche: This device is designed so that when that first electron moves, it hits other electrons, knocking them loose too. Then those hit more. It's like a single snowball rolling down a hill, picking up more snow until it becomes a massive avalanche.
• Why it matters: This "avalanche" effect amplifies the signal thousands of times, making the detector incredibly sensitive. Even a single photon of light can be detected.

4. The Challenges They Overcame

Even with the breakthrough, there were two main hurdles:

• The "Traffic Jam" (Doping): The materials had too many "impurities" (like too many cars on a highway), which slowed down the electrons. The team realized that by using a slightly thicker buffer in the future, they could clean up this traffic jam and make the device even faster and more efficient.
• The "Noise" (Dark Current): When the device is in the dark, it sometimes creates its own fake signals (noise). Because their device was new and unpolished, it had a bit more noise than the best commercial detectors. However, they proved it works, and they have a clear plan to reduce the noise by improving the "surface finish" of the device.

5. The Bottom Line

This paper is a "proof of concept." It's like showing that a new type of engine can actually run, even if it's not perfectly tuned yet.

• What they achieved: They built a detector on a standard silicon chip that can see light up to 2.7 micrometers (farther than before) and amplify that signal massively.
• Why it's exciting: It proves we can make these high-tech infrared eyes using standard computer manufacturing techniques (CMOS compatible). This means we could eventually mass-produce them cheaply for use in everything from self-driving cars to smartphones, allowing them to "see" through fog and smoke better than ever before.

In short: They figured out how to stretch a new material just right so it fits on a silicon chip, creating a super-sensitive eye that can see through the darkest fog, paving the way for cheaper and better infrared technology.

Technical Summary

1. Problem Statement

The development of high-sensitivity photodetectors for the Short-Wave Infrared (SWIR) and Extended SWIR (e-SWIR) ranges (up to 2.5 µm and beyond) is critical for applications like LiDAR, particularly in adverse conditions (dust, fog) and for "eye-safe" wavelengths (1.55 µm and 2 µm).

• Current Limitations: While Silicon (Si) offers excellent multiplication properties (low excess noise) and CMOS compatibility, it lacks absorption in the SWIR/e-SWIR range. Germanium (Ge) absorbs up to ~1.6 µm, but extending this further requires Germanium-Tin (GeSn) alloys with high Tin (Sn) content (>10%).
• The Specific Challenge: Growing high-Sn-content GeSn on Si substrates typically requires a thick Ge buffer layer (700–900 nm) to relax lattice mismatch and reduce threading dislocations. However, for Avalanche Photodiodes (APDs) using a Separate Absorption, Charge, and Multiplication (SACM) design:
  • A thick Ge buffer causes a significant drop in the electric field across the buffer due to high background p-doping ($10^{16}–10^{17} \text{ cm}^{-3}$).
  • This field drop prevents efficient transport of photocarriers from the GeSn absorber to the Si multiplication layer.
  • Consequently, previous designs struggled to achieve high Sn content (needed for >2 µm cutoff) while maintaining a thin enough buffer to sustain the necessary electric field.

2. Methodology

The researchers employed a novel approach to balance crystal quality with electric field efficiency:

• Structure Design: They designed a SACM APD with a significantly thinner Ge buffer (122 nm) compared to the standard 700–900 nm.
  • Layers: n+ Si substrate $\rightarrow$ 583 nm i-Si multiplication layer $\rightarrow$ p+ Si/Ge charge layer $\rightarrow$ 122 nm i-Ge buffer $\rightarrow$ 250 nm i-GeSn absorber $\rightarrow$ 25 nm p+ GeSn contact.
• Growth Process: Epitaxial growth was performed using Reduced Pressure Chemical Vapor Deposition (RPCVD) with SiH$_4$, GeH$_4$, and SnCl$_4$ precursors. A nominal 8% Sn recipe was applied.
• Key Mechanism: The team leveraged the Spontaneous-Relaxed Enhanced (SRE) effect. By using a very thin Ge buffer, the lattice mismatch between the GeSn layer and the Si substrate is amplified, inducing higher tensile strain. This strain promotes the incorporation of Sn beyond the nominal target and facilitates plastic relaxation, allowing for higher Sn content without the need for a thick buffer.
• Characterization: Devices were characterized from 77 K to 300 K using:
  • XRD (2$\theta$-$\omega$ and Reciprocal Space Mapping) and SIMS for composition analysis.
  • TEM for structural interface verification.
  • I-V, C-V, and spectral response measurements to determine breakdown voltage, gain, and responsivity.

3. Key Contributions

• Ultra-Thin Buffer Strategy: Demonstrated that a 122 nm Ge buffer is sufficient to support high-Sn GeSn growth, solving the trade-off between lattice relaxation and electric field distribution in APDs.
• Enhanced Sn Incorporation: Achieved a final Sn content of 12.7% (significantly higher than the nominal 8% target) due to the strain effects induced by the thin buffer.
• Extended Detection Range: Successfully extended the cutoff wavelength of a monolithic GeSn-on-Si APD to 2.7 µm at 300 K, covering the full e-SWIR range.
• High Performance Metrics: Achieved high avalanche gains and responsivity at cryogenic temperatures, validating the SACM design for e-SWIR applications.

4. Key Results

• Material Quality:
  • XRD and SIMS confirmed three distinct Sn content regions within the absorber, peaking at 12.7%.
  • TEM revealed clear interfaces, though some bulk defects were noted, attributed to the high Sn content and thin buffer.
• Spectral Response:
  • Cutoff wavelength increased from 2.2 µm (77 K) to 2.7 µm (300 K).
  • This represents a significant improvement over previous GeSn-on-Si APDs (typically limited to ~2.14 µm).
• Avalanche Gain and Responsivity (at 77 K, 100 µW optical power):
  • At 1.55 µm: Gain up to 21, Responsivity up to 1.45 A/W.
  • At 2.0 µm: Gain up to 52, Responsivity up to 0.66 A/W.
  • Note: Responsivity and gain decreased as temperature increased due to phonon scattering reducing ionization efficiency.
• Dark Current:
  • Dark current density at gain $M=10$ was $7.3 \times 10^{-2} \text{ A/cm}^2$ (at 1.55 µm) and $2.6 \times 10^{-2} \text{ A/cm}^2$ (at 2 µm).
  • While higher than lattice-matched HgCdTe or Sb-based APDs, this is attributed to trap-assisted tunneling (TAT) from threading dislocations and the lack of surface passivation in the prototype.
• Primary Responsivity Limitation:
  • Primary responsivity (before gain) was lower than theoretical values (e.g., 0.066 A/W vs. theoretical 0.19 A/W at 2 µm).
  • This is attributed to partial depletion of the GeSn absorber caused by high background p-doping ($\sim 10^{17} \text{ cm}^{-3}$) and the resulting electric field drop, limiting carrier collection.

5. Significance and Future Outlook

• Feasibility Confirmation: This work proves that high-Sn-content GeSn APDs can be monolithically integrated on Si for e-SWIR detection, offering a CMOS-compatible alternative to expensive III-V or HgCdTe detectors.
• Path to MWIR: The SRE mechanism demonstrated here suggests that even higher Sn contents (up to 30%) are possible, potentially pushing detection into the Mid-Wave Infrared (MWIR, >3 µm).
• Optimization Strategy: The authors propose that slightly increasing the Ge buffer thickness (300–500 nm) could improve crystal quality and reduce background doping, thereby increasing primary responsivity. They suggest utilizing the high-dislocation region near the Ge/Si interface as the p+ charge layer to mitigate doping issues.
• Impact: These advancements are crucial for next-generation LiDAR systems requiring high sensitivity, eye-safe wavelengths, and cost-effective silicon-based manufacturing.