Monolithic InP-on-silicon integration aims to combine optoelectronic devices with silicon-based electronics on a shared substrate. For industrial automation, communications infrastructure and other data-intensive systems, the underlying engineering challenge is significant: indium phosphide (InP) can support useful optical devices, but growing it directly on silicon introduces defects that can limit device performance. Research at Hong Kong University of Science and Technology (HKUST) examined two approaches: growing laser structures on V-grooved silicon and using quantum dots as dislocation filters in InP layers.
Why pursue monolithic InP-on-silicon integration?
Silicon provides a platform for complementary metal-oxide-semiconductor (CMOS) electronics, while InP-based materials can produce lasers operating near 1.55 micrometers. This wavelength lies in the C-band, approximately 1530–1565 nanometers, a low-loss region for silica optical fiber. Such sources are relevant to optical interconnects and dense wavelength division multiplexing (DWDM), which carries multiple optical channels through a fiber.
Growing the optical device directly on silicon could avoid some additional processing associated with wafer bonding and make use of larger-diameter, more economical substrates. The goal extends beyond laser diodes: a sufficiently low-defect InP-on-silicon process could also support integration of III–V photodetectors and high-frequency or high-speed transistors.
The obstacle is the lattice mismatch between InP and silicon, which is approximately 8%. This mismatch can generate threading dislocations, stacking faults, twins and anti-phase boundaries. These defects can propagate into device layers and interfere with the active regions of a laser. Researchers have explored methods including aspect-ratio trapping, selective-area growth, epitaxial lateral overgrowth, buffer layers, strained interlayers, thermal-cycle annealing and offcut substrates. The HKUST work investigated additional ways to manage the growth and defect propagation.
V-grooved silicon as a laser growth template
In one reported demonstration, InGaAs/InAlGaAs multi-quantum-well (MQW) laser diodes were grown directly on on-axis (001) silicon using metal-organic chemical vapor deposition (MOCVD). Unlike miscut substrates, on-axis silicon has the standard crystal orientation. V-grooves were etched into the silicon through a silicon dioxide mask, with parallel stripes at a 130-nanometer pitch.
Growth began with a thin GaAs wetting layer on the groove facets. The researchers reported that this layer helped limit nonuniform growth and large InP clusters that could interfere with the formation of a continuous film. InP was then deposited in stages: initial nucleation and buffer growth formed an array of nanowires, followed by higher-temperature growth that joined the material into a film about 1.5 micrometers thick. An 80-nanometer strained InGaAs layer was included as a dislocation filter.
The resulting InP surface had a reported root-mean-square roughness of 3.31 nanometers. The estimated threading-dislocation density was 2.4 × 108 per square centimeter—3.7 times lower than the reported value for InP grown on planar silicon. Stacking faults remained, at a measured density of 1.2 × 108 per square centimeter. The result illustrates both the potential and the limitation of substrate patterning: defect density was reduced, but not eliminated.
Laser performance and remaining trade-offs
The InP template supported an MQW laser structure with InGaAs/InAlGaAs active layers, separate-confinement waveguides and InP cladding. The material was processed into ridge-waveguide devices with different ridge widths. The ridge geometry included wider regions around the quantum wells, intended to keep current away from the sidewalls and reduce non-radiative surface recombination.
A 10-micrometer-wide, 0.5-millimeter-long device had a pulsed lasing threshold of 360 milliamperes, corresponding to a threshold current density of 7.2 kiloamperes per square centimeter. At a 1-ampere injection current, it produced 35 milliwatts per facet. Its main emission peak was at 1.46 micrometers, compared with 1.48 micrometers for similar devices on native InP.
The researchers attributed the wavelength shift to strain related to the difference in thermal expansion between InP and silicon. They also reported threshold currents around three times higher than for devices on InP, and internal quantum efficiency assessed at a 2.6-fold disadvantage. Defects reaching the quantum-well region were identified as a principal reason for the performance gap, although many defects were reported to terminate within the buffer layers. A longer device, measuring 20 micrometers by 2 millimeters, had a lower threshold current density of 3.3 kiloamperes per square centimeter.
Quantum-dot filters for dislocation control
A separate study examined InAs/InP quantum-dot (QD) layers as dislocation filters in InP grown on planar silicon. The concept is to use the strain field around the dots to bend or otherwise interact with dislocations as they propagate through the material. The researchers compared QD filter structures with a strained InGaAs interlayer and varied the QD structure, including the thickness of the low-temperature InP cap.
Plan-view transmission electron microscopy showed threading-dislocation densities of 1.2 × 109, 5.5 × 108 and 3.0 × 108 per square centimeter for the three samples. The sample with the lowest reported density also produced the strongest room-temperature photoluminescence from an added InAs QD layer. However, the samples showed different growth-related issues: one had notable stacking faults, while another contained extended InAs islands. The results underline that adding filter layers is not simply a matter of increasing their number; excess strain can itself create defects.
Practical implications for optoelectronic integration
These studies show how substrate patterning, buffer growth and strain-based filters can contribute to InP-on-silicon development, while also clarifying the remaining engineering work. For system designers and manufacturing decision-makers, the key consideration is not only whether a device can be fabricated, but whether defect density and optical performance are adequate for the intended application and repeatable integration process.
Direct growth offers a route toward combining silicon electronics with III–V optical devices, but the reported laser results still showed a performance gap compared with native InP. Further defect reduction, improved surface quality and optimization of quantum-dot growth were identified as areas for progress. In practice, monolithic integration will depend on balancing growth conditions and defect-control structures against the material quality required by the laser or other optoelectronic device.