Formation Mechanism and Performance Impact of Twins in InP Single Crystals +

Indium phosphide (InP), as an important III-V compound semiconductor, holds an irreplaceable position in fiber-optic communications, 5G/6G radio-frequency front-ends, and high-speed integrated circuits due to its high electron mobility, direct bandgap, and excellent high-frequency characteristics. PAM-XIAMEN can grow twins-free InP substrate, for more wafer specifications please contact at [email protected].

Currently, InP single crystals are primarily grown by melt techniques such as the liquid-encapsulated Czochralski (LEC) method, the vertical gradient freeze (VGF) method, and the vertical Bridgman (VB) method. However, these processes readily introduce rotation twins – i.e., coherent interfaces formed by a 60° rotation of the crystal about the <111> axis. Twins are among the most common crystal defects in zinc-blende structure compound semiconductors. They not only degrade the structural integrity of the crystal but also cause a series of issues during subsequent processing and device fabrication. Therefore, systematically elucidating the nucleation dynamics of twins, characterization methods, their impact on material and device performance, and suppression strategies is of great theoretical and engineering significance.

1. Formation Mechanism and Kinetic Model of Twins in InP

1.1 Crystallographic Characteristics of Twins

In the zinc-blende structure, twins lie on {111} planes, and the lattices on either side of the twin boundary are mirror-symmetric. According to Wang et al., twins in InP nucleate primarily at two types of sites: one is at the solid‑liquid‑gas tri‑junction, termed “Twin1”; once formed, such twins propagate through the entire ingot. The other nucleates directly in the middle of the solid‑liquid interface, termed “Twin2”, forming local micro‑twins only a few hundred micrometers long. These two types of twins have distinctly different effects on material usability, but both originate from the same physical essence – dynamic undercooling.

Fig. 1 Schematic diagram of twin nucleation and growth

1.2 Undercooling‑Driven Kinetics of Twin Nucleation

By analyzing the microstructure at the tri‑junction region (near the solid‑liquid‑gas three‑phase contact line), Wang et al. proposed a kinetic explanation for twin nucleation. They argued that twins do not appear out of nowhere; their nucleation requires two preconditions: the presence of a facet and the undercooling in the facet region reaching a certain threshold.

Under normal growth conditions, the curved shape at the facet edge locally lowers the melting point, thereby naturally consuming part of the undercooling and reducing the driving force for twin nucleation. However, when the actual undercooling exceeds a critical value (estimated to be only about 0.2K), the “energy barrier” for forming a twin nucleus on the facet is greatly reduced. Once the undercooling further increases (e.g., to 1.5K), the nucleation probability of twins rises exponentially. Even if the absolute probability is very small (on the order of one in several billion), it is sufficient to cause frequent macroscopic twin formation during continuous growth.

Fig. 2 Schematic diagram of the twin nucleation model at the tri‑junction region: (a) Interfacial tension balance among the facet, transitional curved segment, and melt; (b) Formation of a twin nucleus during solid‑liquid interface migration

1.3 Influence of Temperature Gradient and Growth Rate

The magnitude of undercooling is mainly determined by two process parameters: the axial temperature gradient (G) and the growth rate (R). A high temperature gradient keeps the undercooling low in the facet region, whereas a low growth rate avoids rapid accumulation of undercooling. Conversely, if the temperature gradient is low (e.g., below 40K/cm) and the growth rate is high (e.g., exceeding 5mm/h), the undercooling in the facet region increases significantly, and twins become almost inevitable.

Comparative experiments by Wang et al. clearly demonstrate this rule: using a high temperature gradient of 150K/cm and a slow rate of 1mm/h in the LEC method reproducibly yields twin‑free crystals; reducing the temperature gradient to 40K/cm and increasing the rate to 5mm/h immediately induces twins. A similar phenomenon exists in the VGF method: when the temperature gradient is reduced from 5K/cm to 3K/cm, the twin density increases markedly. Therefore, maintaining a sufficiently large G/R ratio is the core strategy for suppressing twins.

Fig. 3 Relationship between the free energy barrier for twin nucleation and undercooling

2. Impact of Twins on Material Properties, Processing, and Devices

2.1 Degradation of Intrinsic Material Properties

• Structural integrity damage: Macroscopic through‑twins (Twin1) render the entire crystal completely worthless as a single‑crystalline substrate. As Herms et al. clearly pointed out: crystals coexisting with macroscopic twins are entirely unsuitable for use as substrate materials.
• Inhomogeneous optical anisotropy: The principal birefringence direction differs on either side of a twin boundary, causing an abrupt change in the polarization state of transmitted light. Herms et al. found via optical measurements that the principal axis direction inside a twin is rotated by approximately 45° relative to the twin boundary, leading to uncontrolled phase changes when light waves pass through the twinned region.
• Degraded mechanical properties: Twin boundaries are locations where dislocations readily accumulate and emit, easily initiating microcracks during thermal cycling or mechanical processing. Wang et al. observed local surface cleavage in the shoulder twin region of VGF‑VB ingots.

Fig. 4 Anisotropy orientation in the intersection region of twins and growth striations

Fig. 5 Schematic diagram of the rotation of optical anisotropy within twin lamellae

2.2 Impact of Twins on Subsequent Processing Steps

• Cutting and polishing: As regions of structural discontinuity, twin boundaries may exhibit different mechanical responses from the matrix during mechanical processing, thereby affecting the uniformity of cutting and chemical mechanical polishing and increasing the risk of surface defect formation.
• Selective etching: The reaction rate of twin boundaries in AB etchant differs from that of the matrix, resulting in distinct topographic steps after etching. This affects the uniform coating of photoresist and pattern alignment in subsequent lithography.
• Epitaxial growth: Epitaxial layers on twinned substrates replicate the substrate defects. For quantum well lasers or HEMT structures, twins at the interface introduce additional scattering centers, reducing carrier mobility. Hall measurement data from Zhu et al. show that in ingots grown by the VGF‑VB process, the average mobility in the tail section (approximately 1260cm²/V·s) is significantly lower than that in the head section (approximately 1640cm²/V·s), partly attributable to the accompanying twins and dislocations.

Fig. 6 Morphology comparison of the InP twin region after etching

Fig. 7 Mobility comparison between head wafer and tail wafer of InP

2.3 Impact on Device Fabrication and Yield

• Lithography and etching: Surface micro‑topography and optical anisotropy induced by twins interfere with the alignment signal of high‑resolution step‑and‑repeat lithography, causing overlay errors. During dry etching, the etch rate in twin regions may deviate by 5‑10%, leading to increased sidewall roughness of mesas.
• Yield loss: For local micro‑twins (Twin2), if their dimensions (typically 340μm long and 14μm wide) exceed the feature size of a single device, that device directly fails. In high‑volume production, such randomly distributed twins significantly reduce the number of functional chips per wafer.

2.4 Degradation of Final Device Performance

• Lasers and LEDs: As non‑radiative recombination centers, twins increase the threshold current and reduce the internal quantum efficiency. Although Hettick et al. reported a photoluminescence (PL) quantum yield of 10±2% for low‑temperature InP microcrystals, that value was measured without considering the long‑term effects of twins; for actual lasers, the additional losses due to twins raise the threshold current density by more than 30%.
• High‑frequency electronic devices (HEMT/HBT): Atomic misalignment at twin boundaries creates dangling bonds, which in turn introduce additional energy levels and alter the local carrier concentration. In the channel of a HEMT, such non‑uniform doping causes local fluctuations in transconductance and increases the device noise figure.
• Integrated photonic circuits: The polarization rotation effect induced by twins is particularly detrimental in photonic integrated circuits. When an optical waveguide crosses a twin region, uncontrolled conversion between different polarization modes occurs, disrupting the functionality of interferometers or polarization beam splitters. PL wavelength mapping by Zhu et al. shows that for a control VGF wafer containing twins, the peak wavelength span is as large as 10.90nm, whereas for an optimized VGF‑VB wafer it is only 1.40nm. For laser arrays in wavelength‑division multiplexing systems, a 10nm wavelength non‑uniformity means failure to meet the channel spacing requirements of communication standards.

Fig. 8 PL quantum yield of InP microcrystals

Fig. 9 Comparison of PL peak wavelength uniformity in InP tail wafers grown by different methods

3. Twin Suppression Strategies and Process Optimization for InP Crystal Growth

Based on the above analysis, the core of twin suppression lies in reducing the dynamic undercooling at the growth interface and minimizing thermal stress. Specific measures include:

• Increasing the axial temperature gradient: Use a gradient higher than 150K/cm in the LEC method, maintain 5‑7K/cm in VGF, and make the thermal field as axisymmetric as possible.
• Reducing the growth rate: For the constant‑diameter section, control the growth rate within 1‑2mm/h to avoid undercooling‑induced twin triggering.
• Optimizing the shoulder shape: Use a “flat shoulder” (shoulder angle close to 90°) or a gently varying shoulder expansion to avoid thermal stress concentration caused by abrupt diameter changes. Wang et al. prepared 170 mm diameter twin‑free InP single crystals using the flat‑shoulder method.
• VGF‑VB hybrid process: Introduce crucible rotation (2rpm) and constant‑speed lowering (2mm/h) during the constant‑diameter section to establish quasi‑steady‑state solidification conditions. Results from Zhu et al. show that the hybrid process reduces the average dislocation density in the tail wafer from 627cm⁻² to 172cm⁻² and narrows the X‑ray diffraction rocking curve full width at half maximum from 162arcsec to 101arcsec.
• Doping control: Avoid using impurities that lower the twin boundary energy, and choose an appropriate doping concentration to maintain a high barrier for twin nucleation.

Whether you need InP wafers for research or for industrial applications, please contact us email at [email protected] and [email protected].

 

References:

1. Wang, S., Sun, N., Shi, Y., Shao, H., Gu, Z., Li, X., … & Zhang, X. (2024). Dynamics analysis of twin formation for InP and preparation of 6 inch InP single crystals. CrystEngComm, 26(36), 4964-4974.
2. Herms, M., Stoehr, M., Mithun, D., Wagner, M., Frank-Rotsch, C., Juda, U., & Souptel, D. Characterization of Twin Structures and Growth Striations in Indium Phosphide and Gallium Arsenide Crystals. Available at SSRN 6755029.
3. Zhu, X., Li, B., Wei, H., & Ma, Y. VGF–VB Stage Switched Growth Process for 4.5 inch InP Single Crystals with Low Dislocation Density and High Uniformity. Available at SSRN 6161586.
4. Hettick, M., Li, H., Lien, D. H., Yeh, M., Yang, T. Y., Amani, M., … & Javey, A. (2020). Shape-controlled single-crystal growth of InP at low temperatures down to 220 C. Proceedings of the National Academy of Sciences, 117(2), 902-906.