For 40 years since the invention of the transistor, there was one inviolable rule about multiple epitaxial layers of different materials: heterostructures. Figuring out how to get beyond heterostructures has yielded improvements in semiconductor performance, however. And as the demand expands for semiconductors other than silicon, managing the growth of otherwise mismatched crystal layers is likely to lead to more advances in photonics and optoelectronics.
The original rule was all layers must be “lattice-matched,” i.e., have the same lattice constant or very nearly so (other chemical issues aside). It makes sense, intuitively, that solid layers grown otherwise will contain internal stresses, and these stresses will likely relax during the growth process, yielding an unacceptable density of dislocations. Or even worse, these stresses will relax later in operating devices under environmental conditions of high currents, voltages, or operating temperatures and result in device failure and unacceptably poor reliability.
Of course, the primary utility of semiconductor electronic and optical devices is based on high-quality, defect-free, single crystal semiconductor materials.
It is, for this reason, the early development of diode lasers was based on heterostructures of GaAs and AlGaAs. AlAs and GaAs have nearly identical lattice constants, and layers of the alloy AlGaAs are essentially lattice-matched to GaAs substrates at any composition. Even so, the lattice-match requirement was considered so important that there were experiments on adding small amounts of phosphorus to AlGaAs to precisely adjust the mismatch to zero.
As 1500-nanometer wavelength diode lasers became important for fiber optic telecommunications, new materials and substrates were required. This led to new InP substrates and different ternary and quaternary alloys such as InGaAs and InGaAsP. The lattice-match requirement remained the rule, however, and the art of crystal growth for all of these materials reflected the associated necessary precision.
Paradigm shifts don’t take place in isolation but generally emerge from multiple technological advancements. In the case of compound semiconductors, particularly for diode lasers, two major developments took place in the late 1970s. First, the methods for compound semiconductor epitaxial growth became much more sophisticated, controllable, and manufacturable. Interestingly, there were two fundamentally different methods — molecular beam epitaxy (MBE) and metalorganic chemical vapor deposition (MOCVD) — that appeared almost simultaneously. Both rapidly comprised the mainstream for growth technology, and both remain viable commercial processes many decades later. The key to both is the capability for large-area growth, very uniform layers, and sufficient precision and control for the growth of multiple very thin layers with abrupt interfaces.
The second development in the same time frame was the introduction of semiconductor quantum wells. Prior to this, a heterostructure layer was typically 100 nm or thicker. With better crystal growth, layers of less than 10 nm became practical. This allowed new physics concepts to be introduced to devices in the form of quantum confinement. Actually, the physics wasn’t so new, and the problem of a particle confined in a finite potential well is a mainstay of quantum mechanics classes. But there are few real applications of it as ubiquitous or commercially important as the semiconductor laser.
In addition to these developments, there were important potential applications that, at the time, were largely impractical owing to the lattice-match rule. For example, laser diodes with wavelengths in the range of 900 to 1100 nm simply were not available using conventional materials. This wavelength range includes 1.06 microns (useful as a diode replacement for Nd:YAG and 980 nm (important as a pump laser for Er-doped glass fibers). The same situation existed for integrated electronic devices, as well. For example, operating speeds for transistors are limited in part by carrier mobility in silicon integrated circuits. A higher mobility material, such as an alloy of Ge and Si, would be very useful if not for the large lattice mismatch between these elements.
It turns out that the lattice-match rule becomes less rigid for very thin layers.
Consider two materials, such as InAs and GaAs, which are similar in most respects and have the same zinc blended crystal lattice structure — with InAs having a larger lattice constant. If we try to grow a thick layer of alloy InGaAs on a GaAs substrate, the large mismatch results in a very high density of undesirable dislocations and a relaxed crystal. If we grow only a thin layer, it is possible for the strain associated with the mismatch to be accommodated elastically (i.e., coherently, without dislocations). The InAs lattice is distorted: compressed in the plane of growth by the smaller lattice of GaAs and elongated in the direction normal to the plane, so as to maintain a nearly constant volume — known as tetragonal distortion. The now strained, but not too strained, layer can be used effectively and reliably in an electronic or photonic device.
The relationship between the layer thickness and the amount of mismatch that can be elastically accommodated depends on material properties, such as elastic coefficients, and is called the critical thickness.
Consider a 980 nm strained-layer laser. We can make the laser by taking a conventional GaAs diode laser and making the active layer strained InGaAs. Since InAs have a smaller energy gap than GaAs, adding indium increases the emission wavelength of the laser but also increases the lattice mismatch, which, in turn, decreases the critical thickness. To reach the 980 nm wavelength, we balance indium composition and layer thickness to define a range of practical and stable, strained-layer 980 nm laser operation. We did exactly that in an early experiment on laser reliability using strained layers below, at, and above the critical thickness. The lasers below the critical thickness proved to be at least as reliable as comparable unstrained standard GaAs lasers, lasting many thousands of hours. At or above the critical thickness, however, the lasers die more quickly and exhibit generally unacceptable failure rates.
As a result of these experiments and similar work on GeSi alloys, strained layers have become a key component in the design toolset for both electronic and photonic devices. Strained thin layers of SiGe alloys are used in integrated circuits for higher mobility channels in metal oxide semiconductor (MOS) devices.
Strained layers are used for 910 nm, 980 nm, and 1060 nm diode pump lasers. 980 nm lasers are a critical component in erbium-doped fiber amplifier repeaters (EDFAs). This example is particularly interesting because the system requirements for EDFAs heavily depend on strained-layer lasers having both high power and extraordinary reliability. In telecommunication lasers at 1550 nm wavelengths, strain compensation from alternating layers of compressive and tensile strain can be used to access a wider range in the design space for multi-layer devices. Strained layers are also commonly used in the active regions of InGaN blue lasers and LEDs for white lighting.
There are other challenges associated with heterostructure materials that are just too different for elastic strain accommodation. The III-V nitrides, including GaN and InGaN, do not have a practical lattice-matched substrate and are usually grown on mismatched substrates such as sapphire, silicon carbide, or silicon. The result is unacceptable strain relaxation during growth with the generation of very large dislocation densities. Clever solutions have been adopted to force the dislocations to be confined in buffer layers well below the active region of the laser or LED. Interestingly, a blue nitride-based diode laser or LED will likely have both — a strained layer active structure grown above a strain-relaxed buffer layer structure.
Highly mismatched semiconductor materials can be exploited to form novel structures: semiconductor quantum dots. During the slow growth of a mismatched material, the growth process transitions quickly from two-dimensional (layer) growth to three-dimensional (island) growth. The result is nanometer-scale quantum dots of single crystal material on, or subsequently embedded in, otherwise mismatched material. These quantum dots have unique electronic and optical properties and are likely to play a role in silicon photonics — the integration of III-V photonic components into otherwise conventional silicon integrated circuits.
They say that rules are made to be broken, but for strained layers in electronic and photonic devices, perhaps the lattice-match rule is not broken but only strained a little, or at least carefully managed.
James J. Coleman received degrees in electrical engineering from the University of Illinois, Urbana. After working at Bell Laboratories and Rockwell International, he joined the faculty of Electrical and Computer Engineering at the University of Illinois, Urbana. After 31 years, he retired from Illinois as the Intel Alumni Endowed Chair Emeritus. In 2019, he joined the University of Texas at Arlington as Presidential Distinguished Professor of Photonics.
In the growth of quantum dots on an unmatched substrate, does the uniform deposition simply condense into quantum dots, like dew? Or does the substrate need to be patterned in some way/
What would be a recommended text for this subject?
Thank you for the excellent article and historical perspective!