Opto-electronics -> Quantum wells integrate optical devices

The performance of integrated devices is another consideration. III-V materials excel at implementing some optical functions; indeed, they sometimes are the only option. For other functions, III-V performance is lower than that of a discrete component made from a better-established material, as in the case of the ultralow optical loss of a silica-based waveguide.

That gives hybrid integration-combining discrete parts made of different materials-an advantage in the short to medium term. But as longer, serially integrated subsystems are fabricated, the elimination of device-to-device interfacing losses will start to favor monolithic approaches.

Contrasting the performance of various approaches on a function-for-function basis might not matter so much if integrated-device prices were an order of magnitude lower than prices for discretes. But because of low yields and packaging considerations, IC costs remain at a level that is unable to stimulate wholesale change. When the breakthrough in IC performance and pricing eventually come s, huge markets for new generations of optical peripherals likely will open, providing major opportunities for electronics design professionals.

Hybrid integration is an important development in optical-component integration. The designer is able to pick the best materials for each component and then plug the components together on a silicon substrate incorporating low-loss silica waveguides and passive alignment and coupling elements. That provides good performance across the spectrum of optical functions.

But hybrid integration is labor-intensive, and there are unanswered questions about whether its costs will remain too high for all-optical systems markets of the future and whether the interfacing losses and reflections between components will be too great to meet the required performance of future equipment.

The alternative, then, lies in monolithic integration using III-V ma terials. Here, photonic IC telecommunications devices such as lasers and detectors are fabricated in materials based on gallium arsenide or indium phosphide, depending on the wavelengths required. Many integration techniques are in early stages of use or development.

The traditional, and most widely used, monolithic integration approach makes use of regrowth. An epitaxial structure such as a gain region is first grown across a substrate. The region is then masked, and the device is etched in order to grow a further structure, such as a waveguide or bandgap, in a subsequent epitaxy stage.

Confining light

The biggest problems here are a mode mismatch at the butt interface joint-resulting in scattering and back-reflection losses-plus the currently low yields of regrowth. As yield effects compound at each processing stage, integrated devices with large numbers of bandgaps are difficult and prohibitively expensive to make.

That situation has led to efforts to exploit the advantages of quantum-well active structures. A quantum well is a very thin semiconductor layer-typically 10 nanometers or smaller-sandwiched between barriers of larger bandgap. Because of the bandgap difference, electrons and positively charged electron holes are trapped in the quantum well. The small size of the well causes electron and hole energy levels to become quantized, so that the bandgap energy of a quantum well is larger than that of an equivalent bulk-material layer.

A quantum well confines carriers efficiently, but it does not confine light effectively, the wavelength of light being much larger than the well thickness. Quantum-well waveguide devices are therefore typically grown with a separate guiding layer, known as a waveguide core. Because the carriers are confined in the quantum well and the light in the waveguide core, the structure is called a separate-confinement heterostructure.

The optical overlap between the waveguide mode and a quantum well is much smaller than t hat between the waveguide mode and the waveguide core. Thus, it becomes possible to remove the quantum wells from the waveguide core, leaving a passive waveguide with identical dimensions and virtually the same refractive index.

The scattering and reflection losses at the butt joint can therefore, in principle, be extremely small. However, if regrowth is then used to add a waveguide, the quality of the butt joint again depends on the precision to which the required layers can be grown.

Single-stage IC fabrication provides a solution. One method is to vary the width of quantum wells across the wafer during a single stage of epitaxy-a technique that exploits the discovery that III-V growth rates differ between patterned and flat surfaces.

In a process known as selective-area epitaxy or selective-area growth, the substrate is coated with a dielectric mask in which slots are opened using photolithography and etching. The growth rate-and hence the quantum-well width-in the opened areas d epends on the width of the opening and the patterning of the mask. The technique is quite limited in the bandgap changes that can be achieved (limiting the types of devices that can be made), and it cannot be used to pattern the bandgap across a wafer in two dimensions, making it hard or impossible to integrate components such as splitters, multiplexers and demultiplexers.

An alternative is disordering-commonly known as quantum-well intermixing (QWI)-which allows the properties of a quantum-well structure to be modified, typically allowing its energy bandgap to be increased, facilitating the monolithic integration of multiple optical functions. Of the several techniques developed, impurity-free techniques are preferable, since they eschew the introduction of electrically active dopants into a semiconductor waveguide-which could result in optical absorption.

With QWI, the active waveguide is in perfect alignment with its passive counterpart, a key benefit for an active/passive waveguide butt joint. That gives QWI an advantage over alternative monolithic integration techniques, such as regrowth, and over hybrid integration. Moreover, precision alignment means there is a near-zero reflection coefficient at the active/passive interface. That performance parameter will assume greater and greater significance as the industry progresses to ultrahigh-speed transmission.