Access Official Technical PDF

Share Summary:LinkedInEmail

InP-on-Si epitaxy could help bring III-V photonic and electronic materials onto silicon platforms, but direct growth is difficult. Differences in lattice spacing, polarity, and thermal expansion between silicon and III-V compounds can generate defects that undermine the quality of the resulting layers. A reported MOCVD growth approach combines patterned silicon, a GaAs buffer, and strain-compensated InGaAsP superlattices to address these challenges.

The approach is relevant to engineers and technical decision-makers evaluating monolithic integration: it demonstrates how substrate patterning and defect-filtering layers can be used together, while also showing the trade-offs involved in controlling strain and composition during epitaxy.

Why InP-on-Si epitaxy is challenging

InP and GaAs have substantial lattice mismatches with silicon. During direct heteroepitaxy, this mismatch—along with differences in crystal polarity and thermal expansion—can lead to anti-phase domains, stacking faults, twins, and misfit and threading dislocations. These defects can propagate through the grown material and affect the suitability of the layer for device fabrication.

Several strategies can help manage defects, including buffer layers, thermal cycling, selective-area growth, and strained-layer structures. The work described here combines a patterned silicon surface with a GaAs buffer and InGaAsP strain-compensated superlattices. The superlattices are intended to encourage dislocation movement and interaction, which can help defects annihilate or move away from regions of interest.

Patterned silicon and a GaAs foundation

The process began with an exact-oriented (001) 300 mm silicon wafer patterned with 90 nm-wide V-grooved trenches, separated by silicon dioxide stripes. The trenches were aligned along the [110] direction. Their V-shaped surfaces exposed {111} silicon planes, which were used to help avoid anti-phase domain formation during III-V nucleation.

A small undercut at the silicon/silicon dioxide interface provided an additional defect-management feature. This geometry supports defect necking, limiting the propagation of some stacking faults from the silicon into the overlying III-V material.

A 2 µm GaAs layer was then grown across the patterned wafer by MOCVD. The GaAs-on-Si layer served as the platform for subsequent InP growth. Characterization showed a threading dislocation density of approximately 7.8 × 107 cm-2 and an RMS surface roughness of 10.8 nm. The reported roughness was associated mainly with arsenic desorption during a high-temperature annealing step.

How strain-compensated InGaAsP superlattices work

After the GaAs-on-Si wafer was diced into smaller pieces, InP was grown on the GaAs using MOCVD. A thin, low-temperature InP nucleation layer helped accommodate the mismatch between InP and GaAs. The structure then incorporated four InGaAsP strain-compensated superlattices, separated by InP spacer layers and followed by an InP cap.

Each superlattice alternated compressively and tensilely strained InGaAsP layers. The opposing strains were designed to balance one another, limiting the net strain while retaining the potential for the layers to interact with threading dislocations. Because InGaAsP is a quaternary material, its composition can be adjusted to control strain and material properties separately.

The superlattices used progressively larger absolute strain levels: approximately 0.5%, 1%, and 1.5% across the sequence. Layer thicknesses were kept below the estimated critical thickness for each strain to reduce the risk of relaxation. The structure was annealed under phosphine flow at the end of growth to promote surface smoothing.

What the characterization revealed

X-ray diffraction, atomic force microscopy (AFM), transmission electron microscopy (TEM), and electron channeling contrast imaging (ECCI) were used to assess the epitaxial layers. In X-ray diffraction, the superlattices appeared as shoulders on either side of the InP peak, consistent with the presence of the strained layers.

The final InP structure had a reported threading dislocation density of approximately 5 × 108 cm-2 and an RMS roughness of 1.2 nm. The smooth surface is a notable improvement over the measured GaAs-on-Si starting layer. ECCI and TEM observations also indicated that the superlattices acted as defect filters, although the work presents a growth approach rather than a complete device demonstration.

Interface quality varied across the stack. TEM showed sharper interfaces in the first two, lower-strain superlattices than in the third and fourth. The higher-strain layers required changes to multiple precursor flows, making precise control of both composition and thickness more demanding. This highlights a practical process challenge: strain design and growth repeatability must be considered together.

Implications for process development

For teams developing III-V materials on silicon, the results illustrate why defect reduction is unlikely to depend on a single layer or process step. Pattern geometry, nucleation conditions, buffer-layer quality, superlattice design, and annealing all contribute to the final material structure.

  • Use substrate patterning deliberately: V-grooves and interface undercuts can help address polarity-related defects and restrict fault propagation.
  • Treat strain as a design variable: Alternating tensile and compressive layers can provide a defect-filtering mechanism while balancing accumulated strain.
  • Monitor composition and interfaces: Higher-strain quaternary layers require careful precursor-flow control, and interface quality may vary through the superlattice stack.
  • Evaluate more than defect density: Surface roughness, structural quality, and uniformity also matter when assessing whether an epitaxial platform is suitable for later device processing.

The reported work supports further development of MOCVD-grown InP virtual substrates on silicon, with potential relevance to monolithic photonic integration. Its practical message is that strain-compensated InGaAsP superlattices can contribute to defect filtering and surface smoothing, but optimizing their strain, period count, thickness, and growth conditions remains important. Combined with patterned substrates and a carefully grown buffer, this approach offers a process framework for improving InP-on-Si material quality.

Access Official Technical PDF

About The Author