Silicon Photonics: A Field that Is Constantly Changing

Giacomo Losio, ProLabs

In this active, exciting area of development, new devices, techniques and studies are presented at every turn.

Electronics and photonics have been two worlds apart for many years: Electronics had an explosion that forever changed the way consumers live and businesses run; photonics became the de facto medium for the transport of information in the telecom and datacom industry. Although optical communication enabled the rise of the Internet and the steep price decline everyone enjoyed, it didn’t reach the level of diffusion and the pace of integration and performance experienced by the semiconductor business.

Semiconductor investments = photonics advances

In a nutshell, the semiconductor business “kept it simple,” while optics has always been a divided camp that failed to generate an economy of scale (Figure 1). The silicon industry built an ecosystem and focused investments in very precise directions: smaller, faster, cheaper. Following that approach, with “the cheaper” came the potential for mass application – and the cycle fueled itself. First, power consumption of chips decreased over time. Second, the cost per transistor went down by more than two orders of magnitude.

Figure 1. The difference between optical and semiconductor diffusion. Photo courtesy of ProLabs.

It’s no wonder that photonics is trying to jump on this train by leveraging all of the semiconductor industry’s investments. From the business point of view, silicon photonics is on a quest to leverage investment, infrastructure, tools and manufacturing processes of the complementary metal oxide semiconductor (CMOS) industry. This will put optics on a trajectory similar to that of integrated circuits (ICs) in terms of integration, manufacturability, scalability, power and cost.

Challenges for silicon photonic devices

Adding photonics to silicon is a challenge for the IC industry, which has to find ways to produce inexpensive and high-performing photonic devices out of silicon. On the other hand, adding silicon to photonics challenges the industry to introduce in the production processes the same concepts – standardization, economy of scale, integration and road map – that brought success to the microelectronics industry. Silicon is an optical material, and its absorption spectrum is actually transparent at optical wavelengths (1300-1500 nm) used by single-mode transmission.

While the refractive index of silicon is 3.5, that of silicon dioxide (SiO₂) is 1.45, which allows very narrow waveguides to be built. The most important problem to address is that silicon has an indirect bandgap, meaning that it is not well suited for lasers.

A silicon photonic circuit will resemble a conventional IC with a photon supply unit. Whereas a conventional IC has a DC electron supply and manipulates electrons with electrical input/outputs (I/Os), silicon photonic circuits will have a DC source of both electrons and photons (a laser), and will have both optical and electrical I/Os.

Many different structures have been proposed to manipulate light in silicon, including modulators, muxes, demuxes, couplers and almost all building blocks of photonic circuits. The three most important are modulation of light, detection of light and lasers.

Modulation of light

Silicon enables implementation of both electroabsorption modulators and Mach-Zehnder modulators, the two most used schemes in current transceivers. Silicon is a semiconductor, so by applying an electrical field, the physical properties of this material can be exploited to modulate the absorption of light. In fact, when an electrical field is present, the energy bands of silicon are distorted and electrons can tunnel across the bandgap, so the energy requirement is lower than in band-to-band absorption. The tunneling process is strongly favored if a photon with energy close to bandgap is absorbed, according to “The Physics of Modulators” by C. Buchal and M. Siegert (1998). This effect is more evident at wavelengths close to the bandgap.

An important physical phenomenon affecting the silicon refractive index is the free-carrier absorption. Optical properties of silicon are strongly modified by the injection of charge carriers into an undoped material or by removal of free carriers from a doped one.

This principle can be used to build an optical modulator in silicon – specifically, a Mach-Zehnder interferometer (MZI). If light coming from a coherent source is split into two beams and each follows a slightly different optical path, then recombining them produces an interference pattern caused by the phase change between both beams. The MZI starts with a waveguide and then splits into two symmetric branches. After a certain distance, the two branches become parallel; they join again in a straight waveguide.

If the MZI is exactly symmetrical and the optical path on both branches is exactly equal, the input light splits at the first Y-junction into two parallel branches and then recombines constructively into the final waveguide. If in one of the interferometer’s arms the light suffers a phase shift of 180°, at the end of the second Y-branch, the light coming from the two branches will recombine in phase opposition and will give rise to destructive interference with no light at the output. The phase shift in one arm is realized by applying a voltage across the waveguide. By designing the geometry, the electrode geometry and the applied voltage, a total phase shift of 180° can be obtained for a specific wavelength. The modulator can also be designed to apply half of the phase shift Δφ per each arm, with one arm applying a Δφ/2 and the other, −Δφ/2 (Figure 2).

Figure 2. Mach-Zehnder modulator operation: The electrical field changes the refractive index of the material to create destructive interference. Photo courtesy of ProLabs.

Detection of light

Because silicon is transparent at optical frequencies in the 1300- to 1500-nm region, it is not well suited for photodetectors. It has to be coupled with other materials, but it must still fit nicely into a CMOS production process for the silicon photonics to be successful. These are the specifications to look for:

• Compatibility with silicon-based materials: large wafer-scale technology; possibility of electronics integration (transimpedance amplifier, or TIA); low-cost integration schemes.

• Broadband detection (1.3-1.6 μm): high absorption coefficient.

• Low dark current: depends on the electrical configuration and the quality of the absorbing layer.

• High bandwidth (frequency operation >10 GHz): low carrier transit time; low RC constant, depending on the considered electrical (pin receiver, for example).

• High responsivity: Determine the configuration to achieve the best light interaction with absorbing layer.

• Compactness: strong absorption coefficient in the complete wavelength range.

In this context, as material compatibility must be ensured, the group IV silicon-germanium technology is important due to the energy band engineering that it allows, with the integration of silicon/silicon-germanium heterostructures into the existing silicon technology.

Silicon-germanium alloys can extend the absorption of light toward longer wavelengths. Pure germanium would allow the realization of high-speed, low-dark-current devices, but the manufacturing process would need to be adapted to grow germanium on silicon. Efficient optical coupling of light into the absorbing region is important to guarantee a high yield. Surface waveguide illuminations have been proposed. For integration with optical micro-waveguides in a silicon-compatible planar technology, the second option is preferred, while surface-illuminated devices have fewer optical alignment problems.

Photodetectors can be built also by integrating direct-bandgap materials from the III-V group (InP, GaAs, etc.) with silicon, but process complications such as a mismatch in lattice constant make this a less appealing solution.

Laser sources

The last and most complicated piece of the puzzle is the laser. As we mentioned, silicon is an indirect-bandgap semiconductor, so the probability for a radiative recombination is low, which means that the electron-hole radiative lifetime is long, on the order of milliseconds. If during this time both the electron and the hole encounter a defect or a trapping center, the carriers might recombine nonradiatively. Typical nonradiative recombination lifetimes in silicon are of the order of some nanoseconds. Silicon is a poor luminescent material because the efficient nonradiative recombinations deplete the excited carriers rapidly.

So far, in the commercial silicon photonics modules that have started being produced, discrete laser chips are attached to the surface of the chip, and light is coupled with different techniques. These early methods allowed devices to come to market but are far from optimal. In fact, they are expensive, require special manufacturing equipment for alignment and do not offer high scalability.

A second approach involves III-V materials, currently widely used for laser devices in conjunction with silicon. Three approaches are possible: • Flip-chip integration of optoelectronic components, a reliable method that allows testing of optoelectronic components in advance. It is a slow process that enables alignment accuracy, with a low density of integration.

• Heteroepitaxial growth of III-V on silicon, a collective process with a high density of integration; it causes a mismatch in lattice constant, thermal coefficients and contamination.

• Bonding of III-V epitaxial layers, a fast integration process with a high density of integration, resulting in high-quality epitaxial III-V layers.

The third approach has been recently explored using molecular bonding (van der Waals attraction between two surfaces) or adhesive bonding (using an adhesive as a bonding agent). Several startups are active in this area. The main attractions are the possibility of integration, the lack of alignment and the moderate final cost that can be achieved.

Another method, and the real “holy grail” of silicon photonics, will be the monolithic integration of source in the CMOS process, which will be – of course – the lowest-cost solution. One promising approach has been developed by Intel, according to “A continuous-wave Raman silicon laser” by H. Rong et al (2005). The company created a continuous-wave silicon Raman laser that overcomes previous limitations, allowing only pulsed modulation. This device exploits the Raman effect, a scattering process in which a lower-energy (longer-wavelength) photon is released. This phenomenon is widely used in optical amplification for long-haul systems.

Few recent works seem to question the belief that silicon is not suited to building a laser, suggesting new structures and physical mechanisms that could help achieve the scope, according to “Stimulated emission in a nanostructured silicon p-n junction diode using current injection,” by M.J. Chen et al (2004; doi: 10.1063/1.1687458), and “Optical gain in materials with indirect transitions,” by T. Trupke et al (2003; doi: 10.1063/1.1571223). All of these results are very encouraging since the proposed systems are p-n junctions, which have excellent electrical qualities. The main problem with the bulk of the silicon approaches is related to the presence of a gain large enough to overcome possible free-carrier losses, a topic that needs to be better explored.

Meet the author

Giacomo Losio is the head of technology at ProLabs in Cirencester, England; email: giacomo.losio@prolabs.com.