Chapter 1: History and Current Status Free

It is a common consensus that the first proposal and preliminary demonstration of silicon photonics were reported by Soref et al. in the mid-1980s (Soref and Lnrenzo, 1986). In the introduction of the work, Soref was a visionary to provide two perspectives: (1) the fabrication of silicon photonics is compatible with that in microelectronics; (2) silicon photonics offers a possibility of monolithic integration of photonics and electronics. More specifically, the work reported edge coupled waveguides and an optical power divider at the telecom wavelengths. Optical modulation and switching based on different mechanisms were discussed, including the injection of free carriers, which is the fundamental effect in silicon modulators and has been widely used in today's research and commercial products. The group provided a more complete review on the advances of group IV components in 1993 (Soref, 1993), with hybrid and monolithic integration implementations on silicon chips. In the review paper, a concept of monolithic integration of opto-electronic components was proposed to offer essential and important functions, such as light emission, modulation, switching, amplification, and detection. Figure 1.1 is a schematic illustration of such a concept with some functions that have been realized, while others remain to be exploited. This proposal essentially paved the road for silicon photonics development for at least 30 years.

However, after Soref's pioneering work, little progress on silicon photonics research was made until 2005 when a few research groups demonstrated silicon modulators (Liao et al., 2005; and Xu et al., 2005) based on the carrier refraction effect (Soref and Bennett, 1987). Part of the reason for this is due to the fact that silicon waveguide dimensions are typically of the order of a few hundred nanometers as determined by the high index contrast of the silicon core and its surroundings, imposing significant challenges to the fabrication process. The nanofabrication technologies kept advancing rapidly in microelectronics, and the linewidth reached ∼100 nm in the early 2000s (Deleonibus, 2006). The fabrication challenges for silicon photonics were overcome by then. Thereafter, there has been a boom of silicon photonics research since the mid-2000s, first driven by the intense interest from industry, such as Intel, IBM, Luxtera, Kotura, etc.

In general, any information transfer system requires three basic components: a transmitter, a link, and a receiver. A silicon photonic transmitter typically employs a light source and a modulator, and the direct modulation of a laser source can be seen in some applications. A link usually takes multiplexed signals as the input, and there may be a switching process through the link in a network. Finally, signals are de-multiplexed at the receiver end, with each detected by a photo-detector. Here, we briefly describe the development history of silicon photonic components, including passive devices, modulators, detectors, and silicon lasers.

To couple light into and out of a silicon chip, a coupler between a fiber and a chip is indispensable, especially considering that the typical mode field area of the silicon waveguide is at least two orders of magnitude smaller than that of the fiber. An inverse taper structure (Almeida et al., 2003) was proposed in 2003 to convert the mode size of light and match that between the silicon waveguide and the fiber. Another approach to achieve effective coupling was based on grating couplers (Taillaert et al., 2004), which receive/emit light from almost vertically positioned fibers with a few degrees of title angles. The area of a grating coupler is larger than that of the fiber, thus ensuring high coupling efficiency. The grating waveguide is then tapered to submicrometer to match with the silicon waveguide. An advantage of such a vertical coupling scheme is the easy testing of multiple devices on the wafer by simply shifting the chip on a translation stage at the cost of limited bandwidths of the grating couplers. Once the coupling issue is resolved, devices on a chip can be characterized.

There are a large number of passive silicon devices, including but not limited to, power splitters/combiners, bends, couplers, crossings, filters, wavelength (de)multiplexers, polarization beam splitters (PBSs), polarization splitter and rotators, mode converters, and (de)multiplexers, and thermal tuning devices based on phase shifters. A review of the passive silicon photonic devices can be found in Su et al. (2020). Significant progress has been made in the passive devices category, with great interest placed on the mode manipulation in recent years, as this is perhaps the last available physical dimension to exploit.

The mid-2000s witnessed many technology breakthroughs in key silicon photonic devices, with silicon modulator as a typical example. In 2005, Intel demonstrated a silicon Mach–Zehnder modulator (Liao et al., 2005) based on the free-carrier plasma dispersion effect, where the two phase shifters are metal–oxide–semiconductor (MOS) capacitors embedded in silicon waveguides. By applying a gate voltage, charges can accumulate to modify the refractive index of the waveguide and, therefore, the optical phase. At the same time, Cornell reported a silicon microring modulator (Xu et al., 2005) by using carrier injection through a PN junction along the microring. Both modulators were based on the carrier refraction effect (Soref and Bennett, 1987) and achieved a 10-Gb/s modulation rate. Since then, there have been increased research interest and record-breaking results on this topic (Green et al., 2007; Dong et al., 2012; Thomson et al., 2012; Xiao et al., 2017; and Sun et al., 2019), with the data rate exceeding 100 Gb/s.

In 2005, Intel reported two consecutive results on Raman laser on silicon operated in pulsed and continuous wave (CW) modes (Rong et al., 2005a, 2005b), respectively. Lasers on silicon have been a holy grail in silicon photonics research. This subject perhaps imposes the most significant challenge in realizing monolithic silicon photonic chips. An extensive review was provided by Zhou et al. (2015). These exciting results greatly facilitated the advances of silicon photonics research and development toward commercialization.

On the waveguide photodetectors, Ge has been a preferred material of choice (Marris-Morini et al., 2018) mainly for two reasons: (1) Ge is compatible with the complementary metal–oxide–semiconductor (CMOS) fabrication process; (2) Ge shows a direct bandgap of 0.8 eV, falling into the near infrared regime for light absorption. To grow Ge on silicon, the lattice mismatch between the two materials should be taken care of. Typically, a two-step fabrication approach can be adopted, with a thin layer Ge grown at a low temperature to minimize the strain, followed by a thick layer G grown at a high temperature for photodetection. The development of Ge photodetectors has continued in the past two decades, and the performances kept improving in terms of speed, responsivity, and dark current. Today, Ge photodetectors on silicon are readily available in process development kits (PDKs) with >35 GHz bandwidth to support 56 Gbaud rate.

Passive silicon photonic devices play many roles in photonic circuits to manipulate lights, e.g., coupling/splitting of power, wavelength filtering, polarization splitting/combining, and mode conversion/multiplexing.

Generally, a conventional approach to categorizing photonic devices is based on different physical dimensions (Su et al., 2020), which typically include wavelength, polarization, and mode. Therefore, passive silicon photonic devices can be divided into fixed and tunable filters, polarization beam splitters and rotators, as well as mode converters and multiplexers. While we limit the discussion to silicon waveguide based devices, similar structures and principles can be extended to a wide range of photonic devices made of silicon-family materials (i.e., Si₃N₄, silica, etc.) or other group IV materials.

A wavelength filter can select/block one or multiple signal channels in the wavelength domain. Silicon photonic wavelength filters are essential components in wavelength division multiplexing (WDM) systems. A variety of silicon wavelength filters have been demonstrated using resonance or interference structures, such as resonators, interferometers, diffraction gratings, and arrayed waveguide gratings (AWGs). An ideal bandpass filter possesses a flat-top transmission band that filters the desired spectral component of incoming signals. Here, we review the current status of fixed and tunable filters with the following representative structures in Fig. 1.2.

A microring resonator (MRR) based add-drop filter consists of a waveguide ring resonator side coupled to two bus waveguides with four ports [Fig. 1.2(a)]. When multiple signals of different wavelengths enter the input waveguide, only the signal(s) on resonance can couple into the MRR and output at the drop port, while other off-resonance signals pass through. Simultaneously, one can inject a signal on resonance from the add port to let it combine with the pass signals. MRR-based add-drop filters are easy to design and fabricate, while the downsides of the device are the limited free spectral range (FSR) of ∼10 nm, and its Lorentzian transmission function with a limited suppression ratio out of band. Multiple MRRs can be synthesized to improve the filter response. A vertically triple-coupled MRR add-drop filter shows a flat-top passband with an FSR of 25.8 nm (Yanagase et al., 2002). A third-order coupled resonator-based filter can achieve a large flat 3-dB bandwidth of 412 GHz, an FSR of ∼18 nm, and >18-dB out-of-band rejection ratio (Li et al., 2009). A filter based on quadruple series-coupled racetrack resonators demonstrates a flat-top passband with a ripple of 0.2 dB, an FSR of 37.52 nm, and a suppression ratio of 37.2 dB (Boeck et al., 2013). A tunable filter with cascaded tunable MRRs enables a bandwidth of 1–2 GHz and a high out-of-band suppression ratio (ER) of 50 dB (Dong et al., 2010a). By cascading 40 MRRs, a tunable filter achieves a flat-top bandwidth range of 10–20 GHz (Morton et al., 2014).

Mach-Zehnder interferometer (MZI) is a typical interferometer filter. It consists of two directional couplers connected to two waveguides with different arm lengths; thus, different wavelengths may output at different ports with constructive/destructive interference as shown in Fig. 1.2(b). This type of interferometer filter is a finite impulse response (FIR) filter. Filters based on asymmetrical MZI configurations can have much larger FSRs. To achieve flat-top passbands, multiple MZIs can be cascaded for WDM applications (Horst et al., 2013). Practical applications often need athermal operation, and such an athermal flat-top two-stage silicon MZI filter can be realized by using a strip waveguide and a hybrid strip-slot waveguide to achieve a 1-dB bandwidth of 688 GHz and an FSR of 14.7 nm (Deng et al., 2016).

Diffraction grating is based on periodic dielectric perturbations; it can select a certain wavelength if the Bragg condition is satisfied, i.e., 2π phase shift between the diffraction of two adjacent grating structures [Fig. 1.2(c)] with an index modulation period of $λn2$ for a wavelength $λn$ of interest. Such a filter is an FIR filter. Various grating-based devices have been demonstrated, including gratings, anti-symmetric gratings, and subwavelength grating (SWG)-based contra-directional couplers (contra-DCs) (Liu et al., 2017; Charron et al., 2018; Xiao et al., 2018; and Yun et al., 2019). The contra-DCs show the flat-top response, nearly infinite FSRs, and large bandwidths. Toward practical application, a four-channel WDM filter based on the asymmetric contra-DC has been demonstrated with a channel spacing of 2–3 nm and an insertion loss of 0.23–0.58 dB (Zhu et al., 2019).

The greatest success in optical filters may be the invention and commercialization of arrayed waveguide grating (AWG), which is the most widely used integrated (de)-multiplexing device for multiple wavelength channels. The fundamental mechanism of the AWG filter is based on the interference principle by employing multiple parallel MZIs to (de)multiplex multiple wavelength channels (Dragone et al., 1997), as illustrated in Fig. 1.2(d). The AWG wavelength filter is, thus, an FIR filter. Usually, there is a compromise between the filter performances and the device footprint determined by the number of channels and their spacing, and device layout parameters such as waveguide bending radii. For AWGs built on silicon waveguides, good performances with small footprints can be obtained with a relatively large channel spacing of 400 GHz or above (Pathak et al., 2013; and Chen et al., 2016). With smaller channel spacings of 200 GHz or lower (Cheben et al., 2007; and Cheung et al., 2012, 2014), the device performances such as the crosstalk degrade and the device footprint increases. As one example, a bidirectional AWG with an MZI interleaver achieves 18 WDM channels with a 200-GHz spacing on a footprint of 520 × 190 µm² (Chen et al., 2015).

Silicon material possesses a high TO coefficient of 1.86 × 10⁻⁴ K⁻¹ and large heat conductivity (∼149 W m⁻¹ K⁻¹), which are suitable for highly efficient thermally tunable devices. Filters are indispensable in many applications, particularly in reconfigurable optical add-drop multiplexers, where wavelength and bandwidth tenability is of particular interest. Based on the aforementioned fixed filters, tunable filters employing similar configurations can be realized to achieve wavelength or bandwidth tenability. These devices mainly include MRRs (Chen et al., 2007; and Yao and Wu, 2009), cascaded MRRs (Ong et al., 2013; Dai et al., 2016; Wang et al., 2018; and Poulopoulos et al., 2019), MZIs (Deng et al., 2018), MZI-MRRs (Ding et al., 2011; and Orlandi et al., 2014), and cascaded grating structures (St-Yves et al., 2015; and Jiang et al., 2018). Table 1.1 lists some typical tunable filters with both bandwidth and wavelength tenability.

For tunable filters, one important consideration is the energy efficiencies of the tunable elements. A silicon resonator with a silica upper cladding shows a typical tuning efficiency of ∼0.25 nm mW⁻¹ (Gan et al., 2007), which is related to the thermal conductivity of silica of 1.44 W m⁻¹ K⁻¹ (Yamane et al., 2002). Researchers have proposed different approaches to achieving high tuning efficiencies, e.g., placing the microheater directly on the resonant cavities, integrating graphene microheaters on silicon filters, and using suspended waveguide structures. Table 1.2 summarizes the properties of typical energy-efficient silicon tunable filters.

Silicon waveguide devices usually show large birefringence due to the asymmetric dimension of the waveguide cross section. Thus, different polarization modes need to be separated or converted through the use of polarization handling devices. These devices typically include polarization beam splitters (PBSs), polarization rotators (PRs), and polarization splitter-rotators (PSRs). A polarization diversity scheme (Barwicz et al., 2007) employs these devices to realize polarization insensitive photonic circuits. In the following, we review these three types of polarization handling devices.

A PBS can split the two orthogonal polarizations of TE and TM modes in a silicon waveguide. A general method to split two polarizations is based on the cross coupling in two side-coupled waveguides. One can design the structural parameters of the two side-coupled waveguides such that the effective refractive indices are equal for one polarization but different for the other polarization. In the first case, the phase matching condition is satisfied and cross coupling from one waveguide to the other happens. In the latter case, there is no cross coupling, so the light goes straight through. By these means, the input two orthogonal polarizations go to two different output ports separately, realizing the polarization beam splitting function. The same PBS device can achieve polarization combining if orthogonal signals are injected from the two output ports, respectively, and they combine at the input port, since the PBS is a linear reciprocal device. The simplest structure of a PBS is an asymmetrical directional coupler (DC) by leveraging the strong birefringence of the two waveguides. This structure can be extended to many variations, such as partially etched DC and bent DC. In addition, grating-assisted contra-DC structure uses contra-directional coupling, so it does not require stringent coupling length conditions. Subwavelength structures allow complete control of the phase, amplitude, and polarization of light beams, enabling flexible designs of PBSs. Table 1.3 shows the results of some reported silicon PBSs with their schematic illustrations. Asymmetrical DC is a commonly used structure, and there are many variations from this structure with improved performances.

The polarization diversity circuits may employ PSRs to perform polarization splitting and rotating simultaneously, thus shrinking the footprint of the devices. For a waveguide PSR, an asymmetrical dimension of the waveguide cross section is required to break the vertical symmetry. Generally, the operation principle of PSRs is based on mode coupling (Dai and Bowers, 2011). A PSR usually breaks the vertical and horizontal symmetries in its structure; thus, the cross-polarization coupling effect happens if two orthogonal modes exhibit equal effective refractive indices and optical paths. As a result, one polarization can be effectively converted into the other one. Table 1.4 shows the results of some reported silicon PSRs. All these PSRs are based on a DC structure, with many variations to engineer the effective refractive indices and improve the performances such as footprint, crosstalk, bandwidth, and fabrication tolerance.

Mode conversion devices are indispensable in mode division multiplexed systems. A mode converter converts a spatial mode to a desired one, typically from a fundamental mode to a higher order mode, and vice versa. Significant efforts have been devoted to realizing such mode converters in past years. There are many approaches to realizing mode conversion, which can be mainly divided into four categories as shown in Fig. 1.3.

• Phase matching: A fundamental mode in an access waveguide has the same effective refractive index as that of a higher order mode in a bus waveguide. Phase matching and cross coupling are achieved between the different two modes, resulting in high-efficiency mode conversion [Fig. 1.3(a)]. Mode conversion devices based on adiabatic directional couplers (ADCs) are scalable in the number of mode channels without limitation in theory. However, they require precise control of the coupling length, and the coupling strength or the gap distance between two waveguides. These devices typically show long conversion lengths. In addition, the stringent requirement on the waveguide widths and the gap distance impose significant challenges to the fabrication accuracy, thus limiting the achievable largest number count of the mode channels. Other ADC based mode converters may introduce grating couplers and contra-directional grating-assisted couplers to enlarge the design space. Since they are resonant devices with relatively low coupling coefficients, these mode converters are typically tens of micrometers in length with limited bandwidths of several nanometers.

• Beam shaping: A beam splitter separates an input beam into two branches. The beam in the upper branch accumulates an excess phase compared to that in the lower branch through the device design. Thus, an output mode with a desired intensity and phase profile can be obtained at the joint of the two branches. As the spatial splitter is insensitive to the wavelength, such a beam shaping based convertor in inherently broadband [Fig. 1.3(b)]. The beam shaping technique is a straightforward method for mode conversion. Mode-order conversion between the four modes has been demonstrated by manipulating the mode evolution in the multimode branching waveguides (Lee and Shin, 2003). The length of the device is only several millimeters.

• Constructive interference: By launching an input single mode light to a multi-mode device, it excites multiple high-order modes that interfere with one another. Through a careful design of the device structure and parameters, all high-order modes can evolve and converge to a specific mode at the output, while others vanish. Such a mode converter is advantageous in achieving compact footprint, as multiple coherent interferences can take place within short propagation distances [Fig. 1.3(c)]. However, similar to the mode converters based on the branching waveguides, the mode channel count is limited by the number of output ports of the multi-mode interferometer (MMI) coupler. Although these devices are the most compact reported to date, the high-resolution structures require high-accuracy fabrication process, which makes them less attractive in practical applications.

• Metasurface phase perturbation: Given an input mode and a desired output mode, one can design a refractive-index perturbation profile based on the coupled mode theory (Wang et al., 2019) to realize the mode conversion. The excess phase gradient can be introduced along the propagation direction by metal or dielectric meta-surface structure [Fig. 1.3(d)].

The gradient phase can be introduced by metasurfaces or materials to enable mode conversion. Gradient metasurface structures consisting of arrays of plasmonic or dielectric nanoantennas at subwavelength intervals have shown effective waveguide mode conversions over short lengths of ∼10 µm on diverse platforms of silicon, silicon nitride, and lithium niobate (Li et al., 2017). Alternatively, only phase perturbation employing dielectric material can significantly facilitate the design cycle, simplify the fabrication process, and lower the cost. As an example, researchers proposed and experimentally demonstrated TE₀-to-TE₁ mode converter with SOI waveguides with high mode purity and relatively low insertion loss (Ohana et al., 2016). TE₀-to-TE₁ and TE₀-to-TE₂ mode converters have been proposed and experimentally demonstrated using tilted subwavelength dielectric nanoslots, leading to shorter conversion lengths (Wang et al., 2019).

In general, employing adiabaticity and SWGs in the design of mode converters is effective to broaden the operation bandwidths of the mode conversion device, while introducing metasurfaces is promising to shorten the device lengths and reduce the insertion losses. Design tradeoffs are often needed, which depend on the application scenarios. More details of the mode conversion devices and their performances can be found in Su et al. (2020). Table 1.5 summarizes some mode conversion devices we have discussed.

Passive silicon photonic devices are essential in the silicon photonics family. Compared to the active devices, the passive ones have many types with diverse requirements and performance metrics. To implement functional passive silicon devices, researchers and developers typically start from materials of known properties, go through a design process, then send the designed layout for fabrication, and finally test and package the chips. More specifically, this cycle may involve design capture, circuit simulation, layout, global chip design, verification, fabrication, testing, and packaging, with possibly multiple iterations (Bogaerts and Chrostowski, 2018).

This book has seven chapters that discuss this development cycle in greater detail. Chapter 2 introduces the materials and waveguide properties in the silicon family, including silicon, silicon nitride, and silica. Chapter 3 provides the design principles of passive silicon photonic devices. Chapter 4 shows the fabrication process in a research environment mainly using e-beam lithography and dry etching for patterning. Chapter 5 provides the details of the equipment and materials used in clean rooms based on a running fabrication lab in a university. Chapter 6 discusses the testing facility and methods for the characterization of the passive silicon photonic devices. Chapter 7 investigates the integration approaches toward large-scale, multi-element, and multi-channel silicon chips with several examples. Finally, we brainstorm the future technology development trend and emerging applications and offer a perspective of passive silicon photonics at the end of the book.