The central result is a laboratory demonstration of an avalanche photodiode, a light detector designed to amplify a photon-generated electrical signal, that combines direct growth of germanium-tin, or GeSn, on silicon with a reported infrared cutoff at 2.6 microns. In a GeSn-on-silicon device, external quantum efficiency, or EQE, reached 163% at 72 volts under 165 nanowatts of optical power. The paper treats EQE values above 100% as evidence of avalanche multiplication. The authors present the architecture as a possible silicon-compatible route to infrared detectors beyond the conventional silicon and germanium wavelength range.
A detector built around two jobs
The proposed architecture separates absorption from multiplication. GeSn provides the infrared-absorbing layer, while a lateral thin-junction design puts the multiplication region in silicon. The study proposed, demonstrated and characterized this direct GeSn-on-Si arrangement alongside Si reference APDs. That intended division of labor is central to the result: the device is designed to combine GeSn's longer-wavelength response with a silicon avalanche structure.
From wafer to measured signal
GeSn was grown directly on 4-inch Si (100) wafers using low-pressure chemical vapor deposition. The process used ultra-pure hydrogen and a 10% monogermane/tin-tetrachloride feed. The reported optimum was 340 degrees C for 20 minutes, at 50 Torr and a Ge/Sn ratio of about 1,500. Structural checks using X-ray diffraction and transmission electron microscopy found a continuous but irregular film 50 to 80 nanometres thick, with tin varying from 3 to 6 atomic percent. Some areas were crystalline and faceted, while others had dense dislocations, and the film was largely strain-relaxed.
Devices were etched into circular GeSn mesas, or raised device areas, with diameters from 25 to 200 microns using inductively coupled plasma reactive-ion etching. Silicon regions were doped by ion implantation, and annealing for the GeSn-on-Si devices stayed below 450 degrees C. In 100-micron devices tested at 0.5 volts, both materials showed high rectification, with the reported ratio higher for GeSn-on-Si than for the Si reference. Rectification means the junction passed current much more readily in one direction than the other.
Testing combined current-voltage and photocurrent measurements with Fourier-transform infrared spectroscopy for spectral responsivity. Low-temperature EQE and laser tests were performed at 78 K using wavelengths of 1.3, 1.55 and 2.3 microns.
Longer wavelengths, with a trade-off
The electrical measurements showed a trade-off. At 78 K, dark current, the current that flows without light, in the 100-micron Si reference stayed below 40 pA through 20 volts of reverse bias and broke down at 41 volts. The GeSn-on-Si device had 1.12 nA of dark current at 40 volts and approached breakdown at 72 volts. For the GeSn-on-Si APDs, the study reported that breakdown voltage was independent of device diameter.
The optical measurements separated the room-temperature cutoff from the colder laser response. At room temperature and 1 volt, FTIR intensity declined monotonically with wavelength until a 2.6-micron cutoff. At 78 K, laser tests produced significant photocurrent at 1.3, 1.55 and 2.3 microns. The reported 2.3-micron response therefore came from the 78 K test condition, while the cutoff measurement was made at room temperature.
EQE was calculated from wavelength-specific responsivity, photon energy and electron charge. In ordinary terms, it compares the detector's electrical response with the incoming light; the paper treated a value above 100% as evidence of avalanche multiplication. In the Si APD, EQE exceeded 100% above 30 volts and 1,000% at 37 volts. In the GeSn-on-Si APD, it exceeded 1% above 60 volts and reached 163% at 72 volts with 165 nanowatts of optical power; EQE above 100% was also reported at 1.3 microns. These avalanche-related EQE measurements were made at 78 K, so the study does not establish avalanche EQE at room temperature.
What the measurements leave open
The same data point to the device's main constraint. At 1.55 microns, the active GeSn layer was only 50 to 80 nanometres thick, and EQE varied with optical power. The authors linked a reduction at higher power to saturation. A slow light-off decay in the time response was identified as a low-bandwidth limitation for single-photon counting. The authors argue that optical absorption in the thin GeSn layer, rather than avalanche multiplication, is the main present performance constraint, and propose thicker GeSn, higher tin content and architecture optimization as ways to improve future performance.
The manuscript is still a preprint, and its front matter identifies it as arXiv version 2 dated 1 September 2026. It also does not report the replication count, yield or device-to-device dispersion, so the consistency of the result remains unresolved. The findings support the demonstrated architecture and the authors' proposed next steps, while leaving its reproducibility to be tested.
Paper data and sources
Original title: Extending Silicon Avalanche Photodetection Beyond 2 ~μ$m by Direct GeSn Integration
Authors: M. R. M. Atalla, J. Bélec, E. Rahier et al.
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-27
DOI: Not available
Original paper · Full text