Quantum–well intermixing enables multiple functions on a chip

By YeeLoy Lam

For the optical networking industry to repeat the success of the electronics industry, photonic design rules must be standardized and scalable device–fabrication processes must be implemented. Ion–implantation QWI helps meet these goals by providing bandgap engineering in two dimensions on InP substrates.

The limitations inherent in photonic components for optical communications are best overcome by components based on photonic integrated circuits (PIC). These circuits have increasing degrees of active and passive optical functions integrated within a small package, perhaps the size of a coin. They possess advantages such as superior optical performance, compact size, and enhanced package robustness. These attributes could translate into smaller networking racks that consume much less power without the coils of fibers within the network box. But is the cost low enough?

The good news is that PICs can be mass–manufactured using the same wafer–scale manufacturing processes that keep costs low in the semiconductor industry. For example, a 40–channel demultiplexing arrayed–waveguide filter—monolithically integrated with an array of photodiodes on a single piece of indium phosphide (InP) substrate—can save numerous discrete parts assembled using traditional manual methods. Such a PIC will have lower optical losses, at least two orders of magnitude reduction in size, and a drastic reduction in manual labor associated with discrete photonic–component assembling.

Photonic integrated circuits with many functions would be a boon to network equipment designers because they could be designed as integrated solutions to network problems, much like off–the–shelf electronic chip sets. Furthermore, in terms of manufacturability, 100 PICs can be fabricated on a typical 4–in. InP wafer. If 100 wafers can be produced in a week, then 10,000 individual PICs can be produced per week at an InP–wafer fabrication plant.

There are engineering challenges that must be overcome so that PICs can combine electronic control functions, passive waveguides, and active photonic sources at micron dimensions. Semiconductor fabrication processes such as bandgap engineering are critical to realizing the potential benefits of PICs.

FABRICATING PICs
Negatively charged electrons occupy discrete energy states in a semiconductor, and the energy difference between the bands of valence states and conduction states is known as the bandgap. To create different photonic functions on a single semiconductor substrate, localized epitaxial sections with differing energy bandgaps must be formed.

The absorption of photon energy by the electrons in certain direct bandgap semiconductors will cause the electrons to gain energy and jump across the bandgap. Direct bandgap semiconductors such as InP are efficient in electron–photon interactions and form ideal materials for constructing optoelectronic devices. The bandgap energy of these devices can be engineered such that they are close to photon energies at the desired wavelengths. Differences in these bandgap energies are necessary for different devices to ensure their optimized operation.

From a material perspective, the original energy bandgap of a semiconductor wafer is predetermined by its material composition. Further engineering of the bandgap is possible by growing quantum–well structures by uniform epitaxial crystalline growth over the entire wafer. Since the quantum–well structure that is grown is consistent, the effective bandgap achieved is also uniform across the wafer. But this uniformity poses difficulty when two or more different devices need to be monolithically integrated. How would you engineer different bandgaps for different regions of a wafer?

For example, in the integrated laser modulator (ILM), a distributed–feedback (DFB) laser diode is integrated with an electroabsorption modulator (EAM) to overcome higher chirp behavior at high transmission rates. The DFB laser and EAM require different energy bandgaps to optimize the ILM performance (see Fig. 1).

PIC FABRICATION PROCESSES
Conventionally, the process for fabricating PICs with different bandgaps consists of three steps: patterning, etching out the unwanted epitaxial layers, and subsequent epitaxial regrowth over the entire wafer. This process is repeated for each additional bandgap region to be added.

The regrowth step demands precision in controlling the thickness of the regrown epitaxial layers to ensure high quality and therefore precisely align the optical waveguides. Even with perfect alignment, the bulk coupling of the waveguides still gives an undesirably high reflection of 10–3. Such repeated regrowths results in a sharp reduction of process yield.

An alternative approach to fabricating PICs is selective–area epitaxy (SAE), in which predesigned openings are patterned by photolithography on the wafer in the dielectric–masking layer and then the growth steps are performed. Quantum–well structures can be grown using this technique since the effective bandgap of the structure can be tailored by slightly varying the layout of the masking layer.

This method eliminates the butt–joint problem of the regrowth technique but requires a nonfunctional, intermediate mode–conversion section., which is typically longer than the functioning counterpart, taking up valuable chip space and contributing to the propagation loss commensurate with its length. In this technique bandgap engineering is restricted only to the orientation along the direction of the opening, somewhat limiting the two–dimensional (2–D) design freedom usually associated with circuit layout principles (see Fig. 2).

FIGURE 2. In the regrowth technique, an active and cladding layer are grown (top left), then a region is etched away (middle left), and passive waveguide and cladding layers are grown (bottom left). Potential problems in the butt joint cause optical feedback; layer mismanagement causes optical loss. In selective–area epitaxy, a substrate is patterned with a dielectric mask with two opening widths and a coupling section, as seen from the top (top right); after epitaxial growth on the substrate, and even though there is no disruptive butt joint, a side view shows that the coupling section causes some absorptive optical loss (bottom right). Quantum–well intermixing (QWI) is a post–growth–based technique that allows the energy bandgap of a grown quantum well to be modified without any use of epitaxial regrowth. This is achieved by slightly changing the composition of the interface between the quantum well and the barrier layers through a process of controlled lattice disordering in an existing QW structure. Compared to conventional regrowth and SAE, process cycle times are much shorter, and processing is simpler (see Fig. 3).

The ion–implantation QWI method makes use of a glass film etched with a 3–D profile as a mask. Neutral impurities such as phosphorus ions are implanted into the semiconductor material through the glass film. The varying thickness of the film controls the amount of phosphorus ions implanted in the region above a quantum well.

The wafer is sent through an annealing process. At elevated temperatures, the generated vacancies diffuse from their high–concentration region into low–concentration regions further down into the wafer structure. The vacancies movement intermixes the different atoms in the QW structure. The effect of intermixing creates a graded QW structure, thereby increasing its bandgap energy profile at the intermixed region.

Because bandgaps can be fine tuned at any selected region of the QW structure, QWI is able to avoid optical losses at the butt joints of an active region with a passive region. In effect, seamless wafer–scale light connections can be created (see table).

2–D BANDGAP ENGINEERING
There are several significant advantages of QWI fabrication. An important advancement to the art is the use of a glass barrier film comprising patterns of different profile heights to create different bandgaps in the corresponding section of the wafer or die. This one–step technique simplifies the PIC fabrication process, greatly improving the prospect of high–yield and low–cost PIC products. Furthermore, the area of intermixing can be highly precise–up to 2–μm spatial resolutions–and the level of bandgap engineering control can be finely tuned within a cell, which opens up excellent design possibilities in PICs. Different active and passive regions can be fabricated in various geometric configurations on a single wafer, creating densely packed functions in a single PIC.

An array of lasers, for example, can be arranged in a parallel cell configuration to create a multiple–wavelength laser. Alternatively, different active and passive cell sections can be arranged in a serial fashion to create high–power lasers, tunable distributed–Bragg–reflector lasers, superluminescent LEDs, or optical channel monitors (see Fig. 4).

FIGURE 4. By engineering various active and passive regions with QWI, multiple types of PICs can be fabricated, including a multiwavelength laser array (a); a high–power laser in which the windows have higher bandgap energies than the active region (b); an integrated EAM laser in which the DFB section has a bandgap energy close to the photonic range, while the EAM has higher bandgap energy at zero bias (c); an optical channel monitor in which the AWG filter is transparent, while the broadband photodetector array converts light to electricity (d); and a tunable DBR laser with active gain, phase control, and filter sections (e).

We have used the ion–implantation QWI technique to fabricate a 39–wavelength laser within a 6–mm cell area on an InP wafer. Various levels of ion implantations in 2–μm cell resolutions were performed in a single application via a glass film on an InP wafer with an embedded quantum–well layer. The final annealing step changes the quantum–well bandgap energies in smooth steps. All the wavelengths demonstrated similar threshold current densities.

PICs FOR DWDM NETWORKS
Having established that ion–implantation QWI is able to perform bandgap engineering in two dimensions, it is a small step to design and fabricate thin, transparent light circuits to connect different active devices to form a new generation of PICs. Quantum–well intermixing fabrication may allow a PIC designer to lay out optical functions in a 2–D manner much like an integrated–circuit designer in the electronic domain. This potential may allow a new approach to the design and manufacture of photonic components.

To enable this approach, several things need to happen simultaneously. First, a set of common photonic devices such as laser sources, semiconductor optical amplifiers, demultiplexing filters, EAMs, photodetectors, passive waveguides, and couplers should be characterized and standardized as a set of photonic–device design libraries. These device designs must be modular, with well–defined interfaces that can be combined in various circuit designs to perform an array of functions.

Furthermore, a photonic foundry firm should be able to take such a design and manufacture it in volume. This model essentially aims to give great productivity power to the PIC designer. It frees the designer from low–level implementation issues to focus on solution designs through a set of photonic–device design rules.

This model ultimately points the way to an outsourced photonic supply chain in which it is feasible for firms to segregate design, manufacture, and distribution activities, reducing the need for vertically integrated firms. With such segregation, outsourcing is possible, yielding benefits from economies of specialization in design firms and economies of scale in foundry firms.

REFERENCE
J. Marsh, Compound Semiconductors 7(8) (September 2001).

YeeLoy Lam is chief technology officer of DenseLight Semiconductors, 6 Changi North Street 2, Singapore S498831. He can be reached at [email protected].