Introduction to Integrated Optics

This is a continuation from the previous tutorial - dispersive prisms and gratings.

 

1. Introduction

The field of integrated optics is concerned with the theory, fabrication, and applications of guided wave optical devices. In these structures, light is guided along the surface region of a wafer by being confined in dielectric waveguides at or near the wafer surface; the light is confined to a cross-sectional region having a typical dimension of several wavelengths.

Guided wave devices that perform passive operations analogous to classical optics (e. g., beam splitting) can be formed using microelectronic-based fabrication techniques. By fabricating devices in active materials such as ferroelectrics, modulators and switches based on the classical electro-optic effect can be formed.

By fabricating the devices in compound semiconductors, passive and active guided wave devices can be monolithically combined with lasers, detectors, and optical amplifiers. The combination of both passive and active devices in a multicomponent circuit is referred to as an integrated optic circuit (IOC) or a photonic integrated circuit (PIC).

In semiconductor materials, purely electronic devices can be integrated as well to form what is often referred to as an optoelectronic integrated circuit (OEIC).

Progress in the field of integrated optics has been rapid since its inception in 1970. Much of this progress is due to the availability of high-quality materials, microelectronic processing equipment and techniques, and the overall rapid advancement and deployment of fiber optic systems.

The interest in integrated optics is due to its numerous advantages over other optical technologies. These include large electrical bandwidth, low power consumption, small size and weight, and improved reliability.

Integrated optics devices also interface efficiently with optical fibers, and can reduce cost in complex circuits by eliminating the need for separate, individual packaging of each circuit element.

The applications for integrated optics are widespread. Generally these applications involve interfacing with single-mode fiber optic systems. Primary uses are in digital and analog communications, sensors (especially fiber optic gyroscopes), signal processing, and instrumentation.

To a lesser extent, IO devices are being explored in nonfiber systems for laser beam control and optical signal processing and computing. IOCs are viewed in the marketplace as a key enabling technology for high-speed digital telecommunications, CATV signal distribution, and gyros.

This tutorial reviews the IO field, beginning with a brief review of IO device physics and fabrication techniques. A phenomenological description of IO circuit elements, both passive and active, is given, followed by a discussion of IO applications and system demonstrations.

 

2. Device Physics

Optical Waveguides

Central to integrated optics is the concept of guiding light in dielectric waveguide structures with dimensions comparable to the wavelength of the guided light. In this section we present only a brief survey of the relevant physics and analysis techniques used to study their properties.

A dielectric waveguide confines light to the core of the waveguide by somehow reflecting power back towards the waveguide core that would otherwise diffract or propagate away.

While any means of reflection can accomplish this end (for example, glancing-incidence partial reflections from interfaces between different media can serve as the basis for leaky waveguides), the most common technique employs a 100 percent total internal reflection from the boundary of a high-index core and a lower-index cladding material.

As light propagates down the axis of such a structure, the waveguide cross section can also be viewed as a lens-like phase plate that provides a larger retardation in the core region. Propagation down the guide then resembles a continuous refocusing of light that would otherwise diffract away.

The pedagogical structure used to illustrate this phenomenon is the symmetric slab waveguide, composed of three layers of homogeneous dielectrics as shown in Fig. 1.

 

 

It is well known that propagation in slab structures can be analyzed using either a ray-optics approach or through the use of interface boundary conditions applied to the simple solutions of Maxwell’s equations in each homogeneous layer of the structure.

In the ray-optics description, the rays represent the phase fronts of two intersecting plane-waves propagating in the waveguide core region. Since the steady-state field has a well-defined phase at each point, a finite set of discrete modes arises from the self-consistency condition that, after propagation and two reflections from the core-cladding boundaries, the phase front must rejoin itself with an integral multiple of a \(2\pi\) phase shift.

For a given core thickness, there will be a limited discrete number of propagation angles in the core that satisfy this criterion, with the lower bound on the angle given by the critical angle for total internal reflection, \(\theta_\text{crit}=\sin^{-1}(n_0/n_1)\).

In general, a thicker and higher-index waveguide core will admit a larger number of confined solutions or bound modes. Figure 1 shows both the fundamental even mode and the first higher-order odd mode. If the dimensions are small enough, only one bound mode for each polarization state will exist and the guide is termed a single-mode waveguide.

Care must be exercised to include the angle-dependent phase shift upon total internal reflection, referred to as the Goos-Hanchen shift, that can be viewed as a displaced effective reflection plane.

The quantity \(\beta=(2\pi{n_1}/\lambda)\cdot\sin\theta\), referred to as the propagation constant, is the \(z\) projection of the wave-vector and thus governs the phase evolution of the field along the guide.

In addition to the discrete set of bound modes, plane waves can also enter from one side and pass vertically through such a structure, and form a continuous set of radiation modes. In an asymmetrical structure, some of the radiation modes may be propagating on one side of the guide but evanescent on the other. From a mathematical point of view, the set of all bound and radiation modes forms a complete set for expansion of any electromagnetic field in the structure. Analysis techniques will subsequently be discussed in more detail.

The slab waveguide in Fig. 1 employed total internal reflection from an abrupt index discontinuity for confinement. Some fabrication techniques for waveguides, particularly in glasses or electro-optic materials such as LiNbO₃, achieve the high-index core by impurity diffusion or implantation, leading to a graded index profile. Here a field solution will usually be required to properly describe the modes and the ray paths become curved, but total internal reflection is still responsible for confinement.

Most useful integrated optics devices require waveguide confinement in a stripe or channel geometry. While recognizing that the vector nature of the electromagnetic field makes the rigorous analysis of a particular structure quite cumbersome, the reader can appreciate that the same phenomenon of confinement by reflection will be operative in two dimensions as well. Figure 2 shows the cross sections of the most common stripe or channel waveguide types used in integrated optics.

 

 

Common to the cross section for all these structures is a region on the waveguide axis containing more high-index material than the surrounding cladding areas. The diffused waveguide may require a full two-dimensional analysis, but a common technique for the approximate analysis of high-aspect-ratio channel guides such as (a), (c), and (d), is the effective index method.

In this technique, a slab waveguide analysis is applied sequentially to the two dimensions. First, three separate vertical problems are solved to obtain the modal phase index \(n_\text{mode}\equiv\beta\cdot\lambda/2\pi\) for each lateral region as if it were an infinite slab. These indices are then used as input to a final ‘‘effective’’ slab waveguide problem in the lateral dimension. Since the properties of multilayer slab waveguides play an important role in waveguide analysis, a more comprehensive general formulation is outlined below. This task is more tractable using the field solutions of Maxwell’s equations than the ray-optics approach.

 

 

A general multilayer slab is shown in Fig. 3. Since Maxwell’s equations are separable, we need only consider a two-dimensional problem in the \(y\) direction perpendicular to the layers, and a propagation direction \(z\).

The concept of a mode in such a structure is quantified in physical terms as a solution to Maxwell’s equations whose sole dependence on the coordinate in the propagation direction \(z\) is given by \(e^{i\beta{z}}\). This translates to a requirement that the shape of the field distribution in the \(y\) direction, perpendicular to layers, remain unchanged with propagation.

If we generalize to leaky structures or materials exhibiting loss or gain, \(\beta\) may be complex, allowing for a scaling of the mode amplitude with propagation, but the relative mode profile in the perpendicular \(y\) direction still remains constant. These latter solutions are not normalizable or ‘‘proper’’ in the sense of mathematical completeness, but are very useful in understanding propagation behavior in such structures.

Since the field in each homogeneous layer \(m\) is well known to be \(e^{\pm{i}\vec{k}_m\cdot\vec{r}}\), with \(|\vec{k}_m|=2\pi{n}_m/\lambda\) for the (generally complex) index of refraction \(n_m\), the general solution to the field amplitude in each layer \(m\) is

\[\tag{1}\mathcal{E}_m=[a_me^{iq_my}+b_me^{-iq_my}]e^{i\beta{z}}\]

where \(q_m\equiv[(2\pi{n}_m/\lambda)^2-\beta^2]^{1/2}\).

Inspection of the vector Maxwell’s equations reveals that the general vector solution in the multilayer slab can be broken down into the superposition of a TE (transverse electric) and a TM (transverse magnetic) solution. The TE (TM) solution is characterized by having only one component of the electric (magnetic) field that points in the \(x\) direction, parallel to the layers and perpendicular to the propagation direction \(z\). The mode field amplitude \(\mathcal{E}_m\) in Eq. (1) refers to the \(E_X\) or the \(H_X\) field for the TE and TM case, respectively.

In a very simple exercise, for each of these cases one can successively match boundary conditions for continuous tangential \(\vec{E}\) and \(\vec{H}\) across the interfaces to provide the coefficients \(a_{m+1}\) and \(b_{m+1}\) in each layer \(m+1\) based upon the value of the coefficients in the preceding layer \(m\),

\[\tag{2}\begin{bmatrix}a_{m+1}\\b_{m+1}\end{bmatrix}=\begin{bmatrix}\frac{1}{2}\left(1+\frac{q_m\gamma_m}{q_{m+1}\gamma_{m+1}}\right)e^{-i(q_{m+1}-q_m)y_m}&\frac{1}{2}\left(1-\frac{q_m\gamma_m}{q_{m+1}\gamma_{m+1}}\right)e^{-i(q_{m+1}+q_m)y_m}\\\frac{1}{2}\left(1-\frac{q_m\gamma_m}{q_{m+1}\gamma_{m+1}}\right)e^{i(q_{m+1}+q_m)y_m}&\frac{1}{2}\left(1+\frac{q_m\gamma_m}{q_{m+1}\gamma_{m+1}}\right)e^{i(q_{m+1}-q_m)y_m}\end{bmatrix}\cdot\begin{bmatrix}a_m\\b_m\end{bmatrix}\]

where \(y_m\) are the coordinates of the interfaces between layers \(m\) and \(m+1\), and \(\gamma_m\equiv1\) for TE modes and \(\gamma_m\equiv{n}_m^{-2}\) for TM modes. The wave is assumed evanescently decaying or outward leaking on one initial side of the arbitrary stack of complex-index layers, i. e., \(b_0=0\) on the uppermost layer. When the lowermost ‘‘cladding’’ layer \(m=p\) is reached, one again demands that only the coefficient \(b_p\) of the evanescently decaying, or possibly the outward leaking, component be nonzero, which recursively provides the eigenvalue equation \(a_p(\beta)=0\) for the eigenvalues \(\beta_j\). Arbitrarily letting \(a_0=1\), this can be written explicitly as

\[\tag{3}a_p(\beta)=\begin{bmatrix}1&0\end{bmatrix}\cdot\left[\prod_{m=p-1}^{m=0}\mathbf{M}_m(\beta)\right]\cdot\begin{bmatrix}1\\0\end{bmatrix}=0\]

where \(\mathbf{M}_m(\beta)\) is the matrix appearing in Eq. (2).

In practice this is solved numerically in the form of two equations (for the real and imaginary parts of \(a_p\)) in two unknowns (the real and imaginary parts of \(\beta\)). Once the complex solutions \(\beta_j\) are obtained using standard root-finding routines, the spatial profiles are easily calculated for each mode \(j\) by actually evaluating the coefficients for the solutions using the relations above with \(a_0=1\), for example.

Application of Eq. (3) to the simple symmetric slab of Fig. 1 with thickness \(t\), and real core index \(n_1\) and cladding index \(n_0\) can be reduced with some trigonometric half-angle identities to a simple set of equations with intuitive solutions by graphical construction.

Defining new independent variables \(r\equiv(t/2)[(2\pi{n_1}/\lambda)^2-\beta^2]^{1/2}\) and \(s\equiv(t/2)[\beta^2-(2\pi{n_0}/\lambda)^2]^{1/2}\), one must simultaneously solve for positive \(r\) and \(s\) the equation

\[\tag{4}r^2+s^2=(\pi{t}/\lambda)^2(n_1^2-n_0^2)\]

and either one of the following equations:

\[\tag{5}s=\frac{\gamma_1}{\gamma_0}\cdot{r}\cdot\begin{cases}\tan(r)\quad\quad(\text{even modes})\\-\cot(r)\;\quad(\text{odd modes})\end{cases}\]

where again \(\gamma_m\equiv1\) for TE modes and \(\gamma_m\equiv{n}_m^{-2}\) for TM modes.

By plotting the circles described by Eq. (4) and the functions in Eq. (5) in the \((r,s)\) plane, intersections provide the solutions \((r_j,s_j)\) for mode \(j\), yielding \(\beta_j\) from the definition of either \(r\) or \(s\).

This construction is shown in Fig. 4 for TE modes, where Eq. (4) has been parametrized with \(u\equiv(\pi{t}/\lambda)(n_1^2-n_0^2)^{1/2}\). Due to the presence of \(\gamma_m\) in Eq. (5), the TE and TM modes will have different propagation constants, leading to waveguide birefringence.

 

 

It is easy to see from the zero-crossing of Eq. (5) at \(r=\pi/2\) that the cutoff of the first odd mode occurs when the thickness reaches a value

\[\tag{6}t_\text{cutoff}=\left(\frac{\lambda}{2}\right)(n_1^2-n_0^2)^{-1/2}\]

Another important feature of the symmetric slab is the fact that neither the TE nor TM fundamental (even) mode is ever cutoff. This is not true for asymmetric guides. More complicated structures are easily handled using Eq. (2), and one can also approximate graded-index profiles using the multilayer slab. However, analytical solutions also exist for a number of interesting graded-index profiles, including parabolic, exponential, and \(\cos{h}^{-2}\).

Once the modes of a waveguide are known, there are many physical phenomena of interest that can be easily calculated. Quite often the propagation constant is required to evaluate the phase evolution of guided-wave components. In other cases, the designer may need to know the fraction of the propagating energy that lies within a certain layer of the waveguide system.

Perhaps the most important application is in evaluating waveguide coupling phenomena that describe how the light distribution evolves spatially as light propagates down a waveguide system. For example, it is easy to show from the mathematical completeness of the waveguide modes that the energy efficiency \(\eta\) of the coupling into a mode \(\mathcal{E}_m\) of a waveguide from a field \(\mathcal{E}_\text{inj}\) injected at the input facet of the waveguide is given by \(\eta=|\int\mathcal{E}_m^*(y)\mathcal{E}_\text{inj}(y)dy|^2\). Here the fields have been assumed normalized such that \(\int_{-\infty}^{\infty}|\mathcal{E}_m|^2dy=1\).

Coupled-mode theory is one of the most important design tools for the guided-wave device designer. This formalism allows for the calculation of the coupling between parallel waveguides as in a directional coupler. It also allows for the evaluation of coupling between different modes of a waveguide when a longitudinal perturbation along the propagation direction ruins the exact mode orthogonality.

An important example of the latter is when the perturbation is periodic as in a corrugated-waveguide grating. Waveguide gratings are used to form wavelength selective coupling as in Bragg reflectors, distributed feedback lasers, and other grating-coupled devices.

In some instances, evaluating the performance of devices where radiation plays a significant role may be tedious using a modal analysis, and techniques such as the beam propagation method (BPM) are used to actually launch waves through the structure to evaluate radiation losses in waveguide bends, branches, or complicated splitters.

 

Index of Refraction and Active Index-changing Mechanisms

Index of Refraction

Waveguide analysis and design requires precise knowledge of the material index of refraction. One of the most common electro-optic materials is LiNbO₃, a uniaxial birefringent crystal whose index can be characterized by providing the wavelength-dependent ordinary and extraordinary indices \(n_o\) and \(n_e\). They are given by

\[\tag{7}n_{o,e}^2=A_{o,e}+\frac{B_{o,e}}{D_{o,e}-\lambda^2}+C_{o,e}\lambda^2\]

where

\[\begin{matrix}A_o=4.9048&B_o=-0.11768&C_o=-0.027169&D_o=0.04750\\A_e=4.5820&B_e=-0.099169&C_e=-0.021950&D_e=0.044432\end{matrix}\]

Glasses are also common substrate materials, but compositions and indices are too varied to list here; indices usually lie in the range of 1.44 to 1.65, and are mildly dispersive. Commonly deposited dielectrics are SiO₂ and Si₃N₄ with indices of ~ 1.44 and ~ 2.0 at 1.55 μm.

InP and GaAs are by far the most common substrates for IO devices in semiconductors. The usual epitaxial material on GaAs substrates is Al_xGa_1-xAs, which is nearly lattice-matched for all values of \(x\), with GaAs at \(x=0\) providing the narrowest bandgap. For photon energies below the absorption edge, the index of refraction of this material system is given by

\[\tag{8}n_{\text{AlGaAs}}(E,x)=\left[1+\gamma(E_f^4-E_\Gamma^4)+2\gamma(E_f^2-E_\Gamma^2)E^2+2\gamma{E}^4\cdot\ln\left(\frac{E_f^2-E^2}{E_\Gamma^2-E^2}\right)\right]^{1/2}\]

where \(E=1.2398/\lambda\) is the incident photon energy,

\[\tag{9}\gamma=\frac{E_d}{4E_0^3(E_0^2-E_\Gamma^2)}\quad\text{and}\quad{E_f}=(2E_0^2-E_\Gamma^2)^{1/2}\]

where

\[\begin{align}E_0(x)&=3.65+0.871x+0.179x^2\\E_d(x)&=36.1-2.45x\\E_\Gamma(x)&=1.424+1.266x+0.26x^2\end{align}\]

For devices fabricated on InP substrates, common in telecommunications applications for devices in the 1.3 μm and 1.55 μm bands, the most common epitaxial material is a quaternary alloy composition \(\text{In}_x\text{Ga}_{1-x}\text{As}_y\text{P}_{1-y}\). In this case, the material is only lattice matched for the specific combination \(y=2.917x\), and this lattice-matched alloy can be characterized by its photoluminescence wavelength \(\lambda_{\text{PL}}\) under low-intensity optical excitation. The index of this quaternary allow is given by

\[\tag{10}n_Q(E,E_\text{PL})=\left(1+\frac{A_1}{1-\left(\frac{E}{E_\text{PL}+E_1}\right)^2}+\frac{A_2}{1-\left(\frac{E}{E_\text{PL}+E_2}\right)^2}\right)^{1/2}\]

where \(E=1.2398/\lambda\) and \(E_\text{PL}=1.2398/\lambda_\text{PL}\) are, respectively, the incident photon energy and photoluminescence peak photon energy for \(\lambda\) in μm and \(A_1(E_\text{PL})\), \(A_2(E_\text{PL})\), \(E_1\), \(E_2\) are fitted parameters given by 

\[\tag{11}\begin{align}A_1&=13.3510-5.4554\cdot{E}_\text{PL}+1.2332\cdot{E}_\text{PL}^2\\A_2&=0.7140-0.3606\cdot{E}_\text{PL}\\E_1&=2.5048\text{ eV}\quad\text{and}\quad{E_2}=0.1638\text{ eV}\end{align}\]

For application to the binary InP, the value of the photoluminescence peak should be taken as \(\lamba_\text{PL}=0.939 μm.

Many integrated optics devices rely on active phenomena such as the electro-optic effect to alter the real or imaginary index of refraction. This index change is used to achieve a different device state, such as the tuning of a filter, the switching action of a waveguide switch, or the induced absorption of an electroabsorption modulator. A brief survey is provided here of the most commonly exploited index-changing phenomena.

 

Linear Electro-Optic Effect

The linear electro-optic or Pockels effect refers to the change in the optical dielectric permittivity experienced in noncentrosymmetric ordered materials that is linear with applied quasi-static electric field. This relation is commonly expressed using the dielectric impermeability \((1/n^2)_{ij}\equiv\epsilon_0\partial{E_i}/\partial{D_j}\) appearing in the index ellipsoid equation for propagation in anisotropic crystals.

Symmetry arguments allow \((1/n^2)_{ij}\) to be contracted to a single subscript \((1/n^2)_i\) for \(i=1, . . . ,6\). In the principal axes coordinate system, the impermeability is diagonalized and \((1/n^2)_i=0\) for \(i=4,5,6\) in the absence of an applied electric field, with the value of\((1/n^2)_i\) providing the inverse square of the index for optical fields polarized along each axis \(i=1,2,3\).

For an electric field expressed in the principal axes coordinate system, the changes in the coefficients are evaluated using the \(6\times3\) electro-optic tensor \(\mathbf{r}\)

\[\tag{12}\Delta\left(\frac{1}{n^2}\right)_i=\sum_{j=1}^3r_{ij}E_j\qquad{i=1,...,6}\]

With an applied field, the equation for the index ellipsoid in general must be rediagonalized to again yield \((1/n^2)_i=0\) for \(i=4,5,6\). This provides a new set of principal axes and the coefficients in the new index ellipsoid equation provide the altered value of the refractive index along each new principal axis.

For a particular crystal, symmetry also limits the number of nonzero \(r_{ij}\) that are possible. In the cubic zinc-blend III-V compounds there are only three equal nonzero components \(r_{63}=r_{52}=r_{41}\) and the analysis is relatively easy.

As an example, consider a static field \(\mathbf{E}\) applied along the (001) direction, surface normal to the wafers commonly used for epitaxial growth. The rediagonalized principal axes in the presence of the field become the (001) direction (z-axis), the (011) direction (x-axis), and the (\(01\bar{1}\)) direction (y-axis); the latter two directions are the cleavage planes and are thus common directions for propagation.

The respective index values become

\[\tag{13}\begin{align}n_x&=n_0-\frac{1}{2}n_0^3r_{41}\mathbf{E}\\n_y&=n_0+\frac{1}{2}n_0^3r_{41}\mathbf{E}\\n_z&=n_0\end{align}\]

For a slab guide on a (001) substrate and propagation in the (011) direction, the applied field would produce a phase retardation for TE-polarized light of \(\Delta\phi=(\pi/\lambda)n_0^3r_{41}\mathbf{E}\cdot{L}\) after a propagation length \(L\). With values of \(r_{41}\sim1.4\times10^{-10}\) cm/V, micron-scale waveguide structures in GaAs or InP lead to retardations in the range of 10\(^\circ\)/V\(\cdot\)mm. This TE retardation could be used as a phase modulator, or in a Mach-Zehnder interferometer to provide intensity modulation. For fields applied in other directions such as the (011), the principal axes are rotated away from the (011) and (\(01\bar{1}\)) directions. Propagation along a cleavage direction can then alter the polarization state, a phenomenon that also has device implications as will be discussed in more detail later.

In the case of LiNbO₃ and LiTaO₃, two of the most common integrated optic materials for electro-optic devices, the dielectric tensor is more complex and the materials are also birefringent in the absence of an applied field. There are eight nonzero components to the electro-optic tensor, \(r_{22}=-r_{12}=-r_{61}\), \(r_{51}=r_{42}\), \(r_{13}=r_{23}\), and \(r_{33}\). For LiNbO₃, the largest coefficient is \(r_{33}\sim30.8\times10^{-10}\) cm/V. Both retardation and polarization changes are readily achieved.

The electro-optic effect in these materials is associated with field-induced changes in the positions of the constituent atoms in the crystal, and the resulting change in the crystal polarizability. The absorption induced by the conventional electro-optic effect is thus negligible.

 

Carrier Effects

In semiconductors, other powerful index-changing mechanisms are available related to the interaction of the optical field with the free electrons or holes. The simplest of these is the plasma contribution resulting from the polarizability of the mobile carriers.

According to the simple Drude model, this is given for each carrier species by \(\Delta{n}\approx-N\cdot{e^2}\lambda^2/(8\pi^2\epsilon_0nc^2m^*)\) in MKS units, where \(N\) and \(m^*\) are the carrier concentration and effective mass, \(e\) is the electronic charge, and \(\epsilon_0\) is the free-space permittivity.

This can produce index changes approaching \(\Delta{n}\sim-0.01\) at \(10^{18}/\text{ cm}^3\) electron/hole doping of the semiconductor, and can be exploited for waveguide design.

Near the bandgap of the semiconductor, there are additional strong index changes with variations in the carrier density that arise from the associated dramatic changes in the optical loss or gain. Since these correspond to changes in the imaginary index, the Kramers-Kronig relation dictates that changes also occur in the real index. These effects are comparable in magnitude and are of the same sign as the free-carrier index contribution, and dramatically impact the performance of semiconductor devices and PICs that employ gain media.

In addition to the effects described here arising from changing carrier populations, the electronic transitions that generate the free carriers can be modified by an applied electric field. For optical frequencies close to these transitions, this can give rise both to electroabsorption and to an enhanced electro-optic effect, which shall be termed electrorefraction, due to the Stark effect on the carrier-generating transitions.

In bulk material , the induced absorption is termed the Franz-Keldysh effect,and can be viewed as a tunneling effect. For an electron in the valence band with insufficient energy to complete a transition to the conduction band, the field can be viewed as adding a potential to the bands that effectively tilts them in space as shown in Fig. 5.

 

 

If the excited carrier also traversed a short distance down-field from its initial location, it would have sufficient energy to complete the transitions. This distance depends on the tilt, and thus the strength of the applied field. Since carriers cannot be precisely localized according to the Heisenberg uncertainty principle, there is a finite amplitude for completing the transition that is an increasing function of electric field. For fields on the order of \(10^5\) V/cm, absorption values of \(\sim100\text{ cm}^{-1}\) can be readily achieved in material that is quite transparent at zero field. According to the Kramers-Kronig relations, in addition to the absorption described above, this will also induce a change in the real index that will be positive below the band edge.

In the case of quantum-wells, carrier-induced effects can be enhanced due to the confinement in the wells. Excitonic effects, resulting from the Coulombic attraction between electrons and holes, produce sharp features in the absorption spectrum near the band gap that can be dramatically altered by an applied electric field. This quantum-confined Stark effect (QCSE) can enhance both electroabsorptive and electrorefractive effects.

This suggests that more compact, lower-drive voltage devices are possible when quantum wells are employed, a fact that has been confirmed experimentally. However, in both the bulk and especially the quantum-well case, care must be taken to operate at an optical frequency where strong electroabsorptive or electrorefractive effects are operative but the zero-field background absorption is not prohibitively high. Another issue that impacts the design of devices based on electroabsorption is the requirement for removal of the photogenerated carriers to prevent saturation.

 

Thermal Effects

In virtually all materials, the optical path length of a given section of waveguide will increase with temperature. This is the combination of both the physical expansion of the material and the change of the index of refraction with temperature. While both are significant, in most integrated-optic materials the latter effect is dominant.

In SiO₂ on Si, for example, this mechanism provides a useful means of index change, and numbers on the order of \(\Delta{n}\sim10^{-5}/^\circ\text{C}\) are observed. This effect has been used to form a variety of thermo-optic switches and filters, but a significant disadvantage for many applications is the inherent slow speed and high power dissipation. In semiconductors, this index change is more than an order of magnitude larger, and leads to frequency shifts in filter devices and single-longitudinal-mode lasers of \(\Delta{f}\sim10\text{ GHz}/^\circ\text{C}\).

 

Nonlinear Effects

Another class of index changes results from the nonlinar optical effects caused by the incident optical fields themselves. Two phenomena will be mentioned here. The first is closely related to the electrooptic effect discussed earlier, where the field giving rise to the index change is no longer ‘‘quasi-static’’ but is in fact the optical field itself. The response of the medium is in general not the same at optical frequencies, but the same symmetry arguments and contracted tensor approach are employed. The polarization resulting from the incident field at \(\omega\) multiplied by the index oscillating at \(\omega\) generates second harmonic components at \(2\omega\).

This frequency doubling can occur in waveguides, but great care must be taken to achieve phase matching where each locally generated frequency-doubled field propagates in such a way as to add coherently to frequency-doubled fields generated farther along the guide.

This requires either that the dispersive properties of the materials and guide geometry allow \(n(\omega)=n(2\omega)\), or that the frequency-doubled light radiates away from the guide at an angle to allow phase-matching of the \(z\)-component of the wave-vector, or that a periodic domain reversal be introduced into the crystal to allow phase matching.

This latter approach, often referred to as quasi phase matching, has generated considerable interest recently. In this approach, the optic axis of the crystal is periodically reversed with a period equal to the difference in the wave vectors of the fundamental and second harmonic. To date, the most promising results are in LiTaO₃, LiNbO₃, and KTP.

In LiNbO₃, periodic domain reversal has been obtained by the application of 100 μsec pulsed electric fields of 24 kV/mm using a 2.8-μm-period segmented electrode that is subsequently removed. The domain reversal in LiTaO₃ can be obtained on a few-micron scale by proton exchange or electron bombardment. KTP has a higher nonlinear coefficient, but the material is not as well developed. Lower efficiencies have been obtained.

A second application of nonlinear optics to integrated structures involves a higher order of nonlinearity referred to four-wave mixing, or in some situations as self-phase modulation. The change in index in these cases arises from the product of two optical fields. If all fields are the same frequency, this is termed degenerate, and if only one field is present, it becomes self-phase modulation with the index change driven by the intensity of the incident wave. This nonlinearity has been of interest in research aimed at creating all-optical logic devices. Here the intensity of either the input signal or a separate gating signal can determine the output port of a Mach-Zehnder or directional coupler switch, for example.

 

3. Integrated Optics Materials and Fabrication Technology

Ion-exchanged Glass Waveguides

Passive integrated optic devices can be fabricated in certain glass substrates using the ion-exchange technique. In this fabrication process, a sodium-rich glass substrate is placed in a mixture of molten nitrate salts containing alkali cations, such as Cs⁺, Rb⁺, Li⁺, K⁺, Ag⁺, and Tl.

During this process, sodium ions at the surface of the host glass are replaced with the alkali cations, resulting in a local increase in the refractive index. Channel waveguides are realized by selectively masking the glass surface. The index change and the dif fusion depth are a function of host material composition, the specific alkali cation being used, the temperature , and the diffusion time. The exchange process can be substantially enhanced by applying an electric field across the wafer while the substrate is immersed in the salt bath.

Multimode devices are typically fabricated using thallium ion exchange in borosilicate glasses. The high polarizability of the thallium ions results in a large index change (> 0.1) while providing low propagation losses (0.1 dB/cm).

However, thallium-sodium ion exchange has two significant drawbacks. Thallium is extremely toxic, and it also has a large ionic radius compared to sodium (1.49 Å compared to 0.95 Å), resulting in low diffusion coefficients and low mobility. It is therefore necessary to process the waveguides at high bath temperatures approaching the glass transition temperature of the host material (500\(^\circ\)C) for long diffusion times (10 h) with high applied electric fields (> 100V/cm) to achieve deep multimode waveguides that efficiently couple to commercially available multimode fiber (50 to 100 micron core size). Finding suitable masking materials is a challenge.

Single-mode devices are typically realized using either Ag⁺-Na⁺, K⁺-Na⁺, or Cs⁺-K⁺ exchange. The first two processes have been extensively studied and are well understood; however, they each appear to have drawbacks. Ag⁺-Na⁺ exchanged waveguides are somewhat lossy (0.5 dB/cm) due to a tendency for silver reduction in the host material. K⁺-Na⁺ exchanged waveguides are typically highly stressed and prone to surface scattering that increases the propagation loss. Although not as extensively studied, Cs⁺-K⁺ exchanged waveguides show great promise. These waveguides are nearly stress-free, low-loss (< 0.1 dB/cm), reproducible, and can be buried using a two-step exchange process. The two-step process further reduces the propagation loss and results in efficient fiber-waveguide coupling (< 0.1 dB loss per interface).

 

Thin Film Oxides

In recent years there has been substantial interest in IO devices fabricated in thin film dielectrics on silicon substrates. This is due in part to the excellent surface quality, large-area wafers, and mechanical integrity of silicon itself.

However, this interest also stems in some cases from the availability of mature silicon processing technology developed by the electronic integrated circuit industry. IO technology on silicon substrates is usually carried out in SiO₂, and there are two generic approaches to the Si/SiO₂ fabrication that have proven capable of producing very high performance IO devices.

IO devices using both approaches are characterized by waveguides that are extremely low-loss and are easily matched in mode characteristics to optical fibers used for transmission, thereby providing very efficient coupling.

The first approach borrows more from the technology of optical fiber manufacture than it does from the Si electronics industry. Using a technique known as flame hydrolysis (FHD), a ‘‘soot’’ of SiO₂ is deposited on a Si wafer to a depth of 50–60 μm, followed by a thinner layer of a SiO₂/GeO₂ mix to form what will become the high-index waveguide core.

This material is consolidated at ~1300\(^\circ\)C for several hours down to roughly half its original thickness, and then the waveguide core layer is patterned using reactive ion etching to form square cross section waveguide cores. Then FHD is again used, followed by more consolidation, to form the upper cladding layers. Typical index differences for the core material are in the range of 0.25 percent to 0.75 percent, with core dimensions of 6–8 μm\(^2\).

A measure of the material quality obtained using this approach is given by some of the extraordinary devices results obtained. IO power splitters have been fabricated to sizes of 1 x 128 using seven stages and a total of 127 Y-branch 1 x 2 splitters and a total device length of 5 cm. Total fiber-to- fiber excess loss for this device was 3.2 dB with a standard deviation of 0.63 dB. A large variety of devices has been made using this technique.

Another technique for Si/SiO₂ fabrication employs film deposition technology borrowed from silicon electronics processing. First a base SiO₂ layer is deposited using high-pressure steam to a thickness of ~ 15 μm to prevent leakage to the high-index Si substrate.

The waveguide and cladding layers are deposited using low-pressure chemical vapor deposition, either from silane and oxygen, or from tetraethylorthosilane and ammonia. Phosphine is added to increase the index, with guide cores typically containing 6.5–8 percent P.

The wafer is usually annealed at 1000\(^\circ\)C to relieve strain and to densify the films. Waveguide losses below 0.05 dB/cm have been reported using this technique, and a large variety of devices have been demonstrated using this approach, including splitters, couplers, and WDM devices.

One of the interesting features of this approach to fabrication is that it readily lends itself to the inclusion of other thin films common in Si processing. One such film is Si₃N₄ and this has been used as a high-index core for waveguides with much larger core-cladding index step.

Such waveguides can generate tightly confined modes that are a much closer match to the modes commonly found in active semiconductor components such as lasers. This feature has been used in a novel mode converter device that adiabatically transforms from the smaller mode into the larger, fiber-matched mode commonly employed in Si/SiO₂ IOCs.

In some instances, slow-response active devices have been fabricated in Si/SiO₂ technology using thermal effects to achieve local index changes in one arm of a Mach-Zehnder interferometer. This can either be used as a thermo-optic switch or as a tuning element in WDM components. The heating element in these devices comprises a simple metal film resistive heater deposited directly on the upper surface of the wafer.

Another characteristic feature of IOCs in Si/SiO₂ is a degree of birefringence that results from the compressive stress induced in the film by the Si substrate after cooling down from the high-temperature film deposition or consolidation. Typical amounts of birefringence are \(n_\text{TE}-n_\text{TM}=3\times10^{-4}\).

This birefringence can cause wavelength shifts with input polarization in WDM components, and techniques to counteract it include stress removal by adding strain-relief grooves in the film, or stress compensation by adding a counteracting stress-inducing film on the surface of the guide.

 

LiNbO₃ and LiTaO₃

The majority of the integrated optics R&D from 1975 to 1985 and the majority of the currently commercial integrated optics product offerings utilize LiNbO₃ as the substrate material.

LiNbO₃ is an excellent electro-optic material with high optical transmission in the visible and near infrared, a relatively large refractive index ( \(n=2.15 - 2.2\)), and a large electro-optic coefficient (\(r_{33}=30.8\times10^{-10}\text{ cm/V}\)). Probably most important, but frequently overlooked, is the widespread availability of high-quality LiNbO₃ wafers.

Hundreds of tons of LiNbO₃ are produced annually for the fabrication of surface acoustic wave (SAW) devices. This large volume has resulted in well-developed crystal growth and wafer processing techniques. In addition, LiNbO₃ wafers are at least an order of magnitude less expensive than they would be if integrated optics was the only application of this material. High-quality three- and four-inch optical-grade LiNbO₃ wafers are now available from multiple vendors.

LiNbO₃ is a uniaxial crystal which is capable of supporting an extraordinary polarization mode for light polarized along the optic axis (\(z\)-axis) and an ordinary polarization mode for light polarized in the \(x-y\) plane.

LiNbO₃ is slightly birefringent with \(n_e=2.15\) and \(n_o=2.20\). LiNbO₃ devices can be fabricated on \(x\)-, \(y\)-, and \(z\)-cut wafers. Phase modulators, fiber gyro circuits, and Mach-Zehnder interferometers are typically fabricated on \(x\)-cut, \(y\)-propagating wafers, and operate with the TE (extraordinary) mode. Push-pull devices, such as delta-beta directional coupler switches, are typically fabricated on \(z\)-cut, \(y\)-propagating wafers and operate with the TM (extraordinary) mode. Both configurations utilize the strong \(r_{33}\) electro-optic coefficient. Devices that require two phase-matched modes for operation are typically fabricated on \(x\)-cut, \(z\)-propagating wafers.

The majority of LiNbO₃ integrated optic devices demonstrated to date have been fabricated using the titanium in-diffusion process. Titanium strips of width 3 – 10 microns and thickness 500 – 1200 angstroms are diffused into the LiNbO₃ at 950 – 1050\(^\circ\)C for diffusion times of 5 – 10 h.

The titanium diffusion results in a local increase in both the ordinary and extraordinary refractive indices so that both TE and TM modes can be supported for any crystal orientation. Titanium thickness and strip width typically need to be controlled to \(\pm1\) percent and \(\pm0.1\) microns, respectively, for reproducible device performance.

Due to the high processing temperatures that approach the Curie temperature of LiNbO₃, extreme care must be taken to prevent Li₂O out-diffusion and ferroelectric domain inversion, both of which significantly degrade device performance.

Photorefractive optical damage also needs to be considered when utilizing Ti-diffused devices for optical wavelengths shorter than 1 micron. Optical damage typically prevents the use of Ti-diffused devices for optical power greater than a few hundred μW at 800-nm wavelength, although the problem can be reduced by utilizing MgO-doped LiNbO₃ wafers. Optical damage is typically not a problem at 1300 and 1550 nm for optical powers up to 100 mW.

An alternative process for fabricating high-quality waveguides in LiNbO₃ is the annealed proton exchange (APE) process. In the APE process, a masked LiNbO₃ wafer is immersed in a proton-rich source (benzoic acid is typically used) at temperatures between 150 and 245\(^\circ\)C and times ranging from 10 to 120 min.

The wafer is then annealed at temperatures between 350 and 400\(^\circ\)C for 1 to 5 h. During the initial acid immersion, lithium ions from the wafer are exchanged with hydrogen ions from the bath in the unmasked region, resulting in a stress-induced waveguide that supports only the extraordinary polarization mode.

Proton-exchanged waveguides that are not subjected to further processing are practically useless due to temporal instabilities in the modal propagation constants, high propagation loss, DC drift, and a much-reduced electro-optic coefficient. However, it has been demonstrated that proper post annealing results in extremely high-quality waveguides that are stable, low-loss, and electro-optically efficient.

The APE process has recently become the fabrication process of choice for the majority of applications currently in production. Since the APE waveguides only support the extraordinary polarization mode, they function as high-quality polarizers with polarization extinction in excess of 60 dB.

As described later, high-quality polarizers are essential for reducing the drift in fiber optic gyroscopes and minimizing nonlinear distortion products in analog links. APE waveguides exhibit low propagation losses of 0.15 dB/cm for wavelengths ranging from 800 to 1550 nm.

APE LiNbO₃ devices exhibit stable performance for optical powers of 10 mW at 800 nm and 200 mW at 1300 and 1550 nm. The APE process can also be used to fabricate devices in LiTaO₃ for applications requiring higher optical powers (up to 200 mW) at 800 nm.

In addition to offering performance advantages, the APE process also appears to be the more manufacturable process. It is relatively easy to scale the APE process so that it can handle 25-wafer lots with no degradation in device uniformity. The fiber pigtailing requirements are also substantially reduced when packaging APE devices since these devices only support a single polarization mode.

After the waveguides have been fabricated in the LiNbO₃ wafer, electrodes need to be deposited on the surface. One-micron-thick gold is typically used for lumped-electrode devices while five-micron-thick gold is typically used for traveling-wave devices to reduce RF resistive losses.

The lift-off process and electron-beam deposition is typically used for lumped-electrode devices while up-plating is typically used for realizing the thicker gold electrodes. Better than 0.5-micron layer-to-layer registration is required for optimum device performance.

As shown in Fig. 6, electrodes on \(x\)-cut LiNbO₃ are usually placed alongside the waveguide so that the horizontal component of the electric field interacts with the propagating TE mode. Electrodes on \(z\)-cut LiNbO₃ are placed on top of the waveguide so that the vertical component of the electric field interacts with the propagating TM mode.

An SiO₂ buffer layer (0.1 to 1 micron thick) is required between the optical waveguide and the electrode on all \(z\)-cut devices to reduce metal-loading loss. A thick (1 micron) SiO₂ buffer layer is also utilized on some \(x\)- and \(z\)-cut devices to reduce the velocity mismatch between the microwave and optical waves in high-speed traveling-wave modulators. A thin layer of amorphous silicon is also utilized on some \(z\)-cut devices to improve device stability over temperature.

 

III-V Materials and Fabrication Technology

In this section we will briefly review some of the epitaxial growth and fabrication techniques that are used to make PICs in III-V materials, with a primary focus on InP-based devices.

III - V Epitaxial Crystal Growth

The epitaxial growth of III-V optoelectronic materials has evolved rapidly during the last decade from nearly exclusive use of manually controlled liquid-phase epitaxial (LPE) growth to a variety of highly versatile computer-automated vapor and beam growth techniques.

These include atmospheric-pressure and low-pressure metal-organic vapor-phase epitaxy (MOVPE), hydride and chloride vapor-phase epitaxy (VPE), molecular beam epitaxy (MBE), chemical beam epitaxy (CBE), and metal-organic molecular beam epitaxy (MOMBE).

One of the critical criteria for evaluating crystal growth is the uniformity, both in thickness and in epitaxial composition. Layer thickness changes of several percent can lead to nm-scale wavelength changes in grating-based lasers and filter devices. Similarly, compositional changes leading to a 10-nm shift in the photoluminescence peak wavelength of the guide layers, which is not at all uncommon, can also result in nm-scale wavelength shifts in distributed feedback (DFB) laser emission wavelengths, in addition to potential undesirable gain-peak mismatches that may result from the \(\lambda_{PL}\) shift itself.

Proper reactor geometry, sometimes with substrate rotation, have been shown capable of percent-level uniformity both in MOVPE and in the beam techniques. One difficulty associated with the latter lies in the ballistic ‘‘line-of-sight’’ growth which prevents regrowth over re-entrant mesa geometries or overhanging mask surfaces often encountered in PIC and laser fabrication, while MOVPE and especially VPE offer outstanding coverage over a wide range of morphologies.

Other criteria to be considered are the doping capabilities. The lower growth temperatures associated with MBE, CBE, and MOMBE enable very abrupt changes in doping level, and highly concentrated doping sheets that are desirable for high-speed transistors in OEICs, for example. Both the vapor and beam techniques have successfully grown semi-insulating Fe-doped InP, a material that is playing an increasingly pivotal role in photonic devices.

The typical PIC processing involves the growth of a base structure that is followed by processing and regrowths. During both the base wafer and regrowths, selective area growth is often employed where a patterned dielectric film is used to prevent crystal growth over protected areas.

This film is typically SiO₂ or Si₃N₄ deposited by CVD or plasma-assisted CVD. This technique is readily used with MOVPE, but care must be taken to keep a substantial portion of the field open for growth to avoid the formation of polycrystalline deposits on the dielectric mask.

Caution must be exercised during regrowths over mesas or other nonplanar geometries, as well as in the vicinity of masked surfaces. Gross deviations from planarity can occur due to overshoots of crystal growth resulting from crystal-orientation-dependent growth rates on the various exposed surfaces.

 

III-V Etching Technology

A fundamenal step in III-V PIC processing is mesa etching for definition of the optical waveguides. This is usually accomplished by patterning a stripe etch mask on a base wafer that has a number of epitaxial layers already grown, and removing some of the layers in the exposed regions to leave a mesa comprised of several of the epitaxial layers. The etching process can either be a ‘‘wet’’ chemical etchant, or a ‘‘dry’’ plasma-type etch.

Wet etching refers to the use of an acid bath to attack the unprotected regions of a surface. The acids that are commonly used to etch III-V materials also tend to significantly undercut a photoresist pattern, and hence photoresist is usually used only in broad-area features or in shallow etches where undercutting is not a concern. For precise geometries such as waveguide stripes, another masking material such as SiO₂ or Si₃N₄ is first deposited and patterned with photoresist and plasma etching, or HF etching (for SiO₂).

In some instances, it is required that the etchants be nonselective, uniformly removing layers regardless of composition. This is usually the case when etching a mesa through a multilayer active region to form a buried heterostructure laser.

Br-based etchants such as bromine in percent-level concentration in methanol, tend to be very good in this regard. This etchant, along with many of the nonselective etchants, will form a re-entrant 54.7\(^\circ\) (111A) face mesa for stripes along the (011) direction (with a nonundercutting mask) and will form an outward-sloping 54.7\(^\circ\) walled mesa for stripes along the (\(01\bar{1}\)) direction. Other etchants, with varying degrees of nonselectivity and crystallographic behavior, include mixtures of HBr , CH₃COOH, or HCl, CH₃COOH, and H₂O₂.

In fabricating precise geometries in III-V integrated optic or PIC devices, it is often desirable to remove specific layers while leaving others, or control mesa heights to a very high degree of precision.

The combination of material-selective etchants and the inclusion of special etch-stop layers offers a convenient and precise means of achieving this. 100-Å-thick layers of InGaAsP can easily halt InP etches even after many microns of etching. Extensive compilations have been made of etches for the InP-based compounds, and the most common selective InP etches are HCl-based.

Typical mixtures are HCl and H₃PO₄ in ratios ranging from 3:1 to 1:3, with the lower HCl content leading to less undercutting and slower etch rates. The HCl-based etchants are highly crystallographic in nature, and can produce mesas with nearly vertical walls or outward-sloping walls, depending on the mesa stripe orientation.

A common selective etch for removing InGaAsP or InGaAs while only weakly attacking InP are mixtures of H₂SO₄, H₂O₂ and H₂O, in a ratio of X : 1 : 1 with X typically ranging from 3 to 30. Selectivities in the range of 10 : 1 and typically much higher are readily achieved. Other selective etchants for InGaAsP are based on HNO₃ or mixtures of KOH, K₃Fe(CN)₆, and H₂O.

Dry etching techniques, such as reactive ion etching (RIE), or other variants, such as chemically assisted reactive ion beam etching (CAIBE), also play a key role in III-V PIC processing.

These have often been carried out using Cl₂-based mixtures with O₂ and Ar, while in other cases the reactive chlorine is derived from compounds such as CCl₂F₂. Recent work has demonstrated excellent results with methane/hydrogen mixtures or ethane/hydrogen. In these latter cases, Ar is also often used as a sputtering gas to remove interfering redeposited compounds. Reactive ion etching has been used both to form mesa and facet structures as well as in transferring grating patterns into semiconductors through an etch mask.

The appeal of reactive ion etching is the lack of mask undercutting that can usually be achieved, allowing very high lateral precision with the promise of reproducible submicron mesa features.

In addition, the ability to create vertical-wall etched facets through a variety of different composition epitaxial layers suggests the possibility of integrated resonator or reflecting and coupling structures without the use of gratings. This approach has been used to form corner reflectors, square-geometry ring-resonators, and a variety of complex waveguide patterns using beam splitters.

Another recent application has been the use of etched-facet technology to create gratings, not as an interfacial corrugation along the waveguide, but as a grating in the other dimension at the end surface of a waveguide for two-dimensional ‘‘free-space’’ grating spectrometers.

 

Grating Fabrication

Many of the PICs employ corrugated-waveguide grating-based resonators or filters, and the most common technique for fabricating these gratings involves a ‘‘holographic’’ or interferometric exposure using a short-wavelength laser source.

Here a thin (typically 500–1000-Å-thick) layer of photoresist is spun on a wafer surface and exposed with two collimated, expanded beams from a blue or UV laser at an appropriate angle to form high contrast fringes at the desired pitch.

Since the illuminating wavelength is precisely known, and angles are easily measured in the mrad range, the typical corrugation in the 2000-Å-period range can be fabricated to Å-level precision in period.

The resist is developed and then functions as an etch mask for the underlying layers. This etching can be either a wet etch (commonly using HBr-based etchants), or a dry reactive ion etch. Commonly used lasers are HeCd at 325 nm or one of the UV lines of an argon ion laser at 364 nm. Electron-beam lithography has also been successfully applied to the generation of gratings for III-V integrated optic devices.

 

Active-Passive Transitions

Compound semiconductors are appealing for PICs in large part due to their ability to emit, amplify, and detect light. However, waveguide elements that perform these functions are not low-loss without excitation, and are generally not suitable for providing passive interconnections between circuit elements. One of the most fundamental problems to overcome is the proper engineering and fabrication of the coupling between active waveguides, containing lower bandgap material, and passive waveguides composed of higher bandgap material.

Most PICs demonstrated to date have employed some form of butt-coupling, where an active waveguide of one vertical and/or lateral structure mates end-on with a passive waveguide of a different vertical and/or lateral structure.

Butt-coupling offers design simplicity, flexibility, and favorable fabrication tolerances. The most straightforward approach for butt-coupling involves the selective removal of the entire active waveguide core stack using selective wet chemical etching, followed by a regrowth of a mated, aligned passive waveguide structure. The principle advantage of such an approach is the independent selection of compositional and dimensional design parameters for the two guides.

Another approach to butt-coupling employs a largely continuous passive waveguide structure with a thin active layer residing on top, which is selectively removed on the portions of the structure which are to be passive. Using material-selective wet chemical etches, the thin active layer (often a thin MQW stack) can be removed with very high reproducibility and precision, and the dimensional control is thus placed in the original computer-automated MOVPE growth of the base wafer. The removal of the thin active layer constitutes only a small perturbation of the continuous guide core constituted by the lower, thicker layer, and efficient coupling can be achieved.

Yet another approach to coupling between two different waveguides employs directional coupling in the vertical plane between epitaxial layers serving as the cores of the two distinct waveguides. This type of vertical coupling can either be accomplished using the principle of intersecting dispersion curves, or through the use of a corrugated-waveguide grating to achieve phase matching. Vertical coupler structures may be useful for wide-tuning applications, since a small change of effective index for one mode can lead to a large change in coupling wavelength.

 

Organic Polymers 

Polymer films are a relatively new class of materials for integrated optics. Polymers offer much versatility, in that molecular engineering permits many different materials to be fabricated; they can be applied by coating techniques to many types of substrates, and their optical and electro-optical properties can be modified in a variety of ways. Applications range from optical interconnects, in which passive guides are used in an optical PC board arrangement , to equivalents of IOCs and OEICs. Polymer devices are also being explored for third-order nonlinear applications.

While numerous methods for fabricating polymer waveguide electro-optic devices have been reported, the most attractive technique consists of spin-coating a three-layer polymer sandwich over a metal film, often on a semiconductor (Si) substrate. The three polymer layers form a symmetric planar waveguide; the middle layer is electro-optic, due to the presence of a guest molecule that imparts the electro-optic property, or the use of a side-chain polymer.

The sample is overcoated with metal and the entire structure is heated near the glass transition temperature and poled at an electric field of typically 150 V/mm. The poling aligns the nonlinear molecules in the middle polymer layer, thereby inducing the Pockels effect and a birefringence. Typical values of index and birefringence are 1.6 and 0.05 respectively. Electro-optic coefficients are in the 16–38-pm/V range. Channel waveguides are subsequently defined by a variety of methods. An attractive technique is photobleaching, in which the waveguide region is masked with a metal film and the surrounding area exposed to UV light. This exposure alters the molecules/linking in the active layer, thereby reducing the refractive index and providing lateral confinement . Losses in such guides are typically in the 1-dB/cm range.

The basic IO modulators have been demonstrated in a variety of polymers. Of particular note is a traveling wave modulator with a 3-dB bandwidth of 40 GHz and a low-frequency V pi of 6 V. Relative to LiNbO₃, polymer modulators can have higher overlap factors because the lower metal layer provides vertical, well-confined signal fields.

However, the relatively low index of polymers and their comparable electro-optic coefficient to LiNbO₃ implies a lower electro-optic efficiency. Polymers do provide a better velocity match of optical and electromagnetic velocities, which can result in very high frequency performance as described above.

For polymers to fulfill their potential, a number of material and packaging issues must be addressed. First, it is highly desirable to develop polymers that overcome the long-term relaxation of the electro-optic effect typical of many of the materials reported to date. Development of polymers with transition temperatures in the 300\(^\circ\)C range (so they can withstand the temperatures typical of device processing and packaging) is also highly desirable. Work on polyimide is particularly promising in this area. Finally, techniques to polish device end faces, pigtail, and package these devices need to be developed.

 

4. Circuit Elements

Passive Devices

Passive guided wave devices are the fundamental building blocks and interconnection structures of IOCs and OEICs. Passive devices are here defined as those dielectric waveguide structures which involve neither application of electrical signals nor nonlinear optical effects. This section will focus on the most important structures: waveguide bends, polarizers, and power splitters, and on the closely related issue of fiber-to-chip coupling.

Waveguide bends, such as those illustrated in Figs. 10, 14, and 18, are needed to laterally offset modulators and increase device-packing density. The most widely used bend is based on an S-bend geometry described by a raised cosine function. This structure minimizes the tendency of light in a dielectric waveguide to ‘‘leak’’ as the guide’s direction is altered by starting with a small bend (large effective bend radius) and then increasing the bend rate until the midpoint of the offset, then following the pattern in reverse through the bend completion.

Since the index difference between the guide and surrounding dielectric material is usually small (\(10^{-3}\) to \(10^{-4}\)), bends must be gradual (effectively a few degrees) to keep losses acceptably ( < 0.5 dB) small. In LiNbO₃, offsets of 100 microns require linear distances of typically 3 mm. In semiconductor research device work, designs with high index steps are sometimes used to form small-radius bends, and selective etching has been utilized to form reflective micro mirrors at 45\(^\circ\) to the guide to create a right-angle bend. To date this latter approach is relatively lossy.

Polarizers are necessary for polarization-sensitive devices, such as many electro-optic modulators, and in polarization-sensitive applications such as fiber gyroscopes. Polarizers can be formed on dielectric waveguides that support both TE and TM propagation by forming overlays that selectively couple one polarization out of the guide. For example, a plasmon polarizer formed on LiNbO₃ by overcoating the guide with a Si₃N₄/Au/Ag thin-film sandwich selectively attenuates the TM mode.

In some materials it is possible to form waveguides that only support one polarization (the other polarization is not guided and any light so polarized radiates into the substrate). By inserting short (mm) lengths of such guides in circuits or alternatively forming entire circuits from these polarizing guides, high extinction can be obtained. For example, annealed proton exchange waveguides (APE) in LiNbO₃ exhibit polarization extinction ratios of at least 60 dB.

Guided wave devices for splitting light beams are essential for most IOCs. Figure 7 illustrates the two common splitters: a directional coupler and a Y junction. The figure illustrates 3-dB coupling (1X2); by cascading such devices and using variations on the basic designs it is possible to fabricate N x N structures. IO splitters of complexity 8 x 8 are commercially available in glass.

 

 

The operation of the directional coupler is analogous to the microwave coupler and is described by the coupled mode equations. The coupling strength is exponentially dependent of the ratio of the guide spacing and the effective tail length of the guided mode. Thus, when guides are far apart (typically greater than 10 microns), as in the left-most portion of the structure in Fig. 7, there is negligible coupling.

When the guides are close together (typically a few microns), power will couple to the adjacent guide. The fraction of power coupled is sinusoidally dependent on the ratio of the interaction length to the coupling length \(L_c\). \(L_c\) is typically 0.5–10 mm and is defined as that length for full power transfer from an incident guide to a coupled guide. The 3-dB coupling illustrated requires an interaction length of half \(L_c\). Operation of this device is symmetric; light incident in any one of the four inputs will result in 3-dB splitting of the output light. However, if coherent light is incident on both input guides simultaneously, the relative power out of the two output guides will depend on the phase and power relationship of the incident signals.

The Y splitter illustrated in Fig. 7 operates on a modal evolution principle. Light incident on the junction from the left will divide symmetrically so that the fundamental mode of each output branch is excited. Branching circuit design follows closely from the design of waveguide bends. The Y-junction angle is effectively a few degrees and the interaction length is a few mm. Operation of this device is not symmetric with respect to loss. If coherent light is incident on both guides from the right, the amount of light exiting the single guide will depend on the power and phase relationship of the optical signals as they enter the junction area. If coherent light is only incident in one arm of the junction from the right, it will experience a fundamental 3-dB loss in propagation to the left to the single guide. This is due to the asymmetric modal excitation of the junction.

An extremely important issue in integrated optics is the matching of the waveguide mode to the mode of the fiber coupled to the guide. Significant mode mismatch causes high insertion loss, whereas a properly designed waveguide mode can have coupling loss well under 1 dB.

To design the guide, one must estimate the overlap integral of the optical fields in the two media. It is reasonable to assume that some sort of index matching between the two materials is also employed.

Figure 8 illustrates the issue with mode profiles in the two transverse directions for a Ti indiffused guide. In general the IO mode is elliptical and often asymmetrical relative to the fiber mode. It should be noted that the loss obtained on a pigtailed fiber-chip joint is also highly determined by the mechanical method of attaching the fiber to the chip. Most techniques use some sort of carrier block for the fiber (e . g ., a Si V-groove) and attach the block to the IO chip. Performance on commercially available devices is typically, 1 dB coupling loss per interface with robust performance over at least the \(-30\) to \(60^\circ\)C range.

 

 

 

Active Devices

Active IO devices are those capable of having their state altered by an external applied voltage, current, or other stimulus. This may include electro-optic devices, or devices that generate, amplify, or detect light. Active IO devices in non-semiconducting dielectrics generally depend on the linear electro-optic effect, or Pockels effect. The electro-optic effect results in a change of the index of refraction of a material upon the application of an electric field. Typical values for a variety of materials is about \(10^{-4}\) for a field of \(10^4\) V/cm. This results in a phase change for light propagating in the field region and is the basis for a large family of modulators.

The fundamental guided wave modulator is a phase modulator, as illustrated in Fig. 6. In this device, electrodes are placed alongside the waveguide, and the lateral electric field determines the modulation. In other modulator designs, the vertical field component is used.

For the geometry shown, the phase shift is \(KLV\), where \(K\) is a constant, \(L\) is the electrode length, and \(V\) is the applied voltage. For LiNbO₃, \(K=\pi{n}^3r_{63}\Gamma/g\lambda\) for the preferred orientation of field along the \(z\) (optic) axis and propagation along the \(y\)-axis. Here, \(n\) is the index, \(r_{63}\) is the electro-optic coefficient, \(\lambda\) is the wavelength, \(g\) is the electrode gap, and \(\Gamma\) is the overlap of the electrical and optical fields.

In general, the value of \(K\) is anisotropic and is determined by the electro-optic tensor. It should be noted that modulators are often characterized by their \(V_\pi\) value. This is the voltage required for a pi-radian phase shift; in this nomenclature, phase shift is written as \(\Phi=\pi\cdot{V}/V_\pi\) where \(V_\pi\equiv{\pi}/KL\). Due to the requirement that the optical field be aligned with a particular crystal axis (e. g., in LiNbO₃ and III-V semiconductors), the input fiber on modulators is generally polarization maintaining.

Modulators in LiNbO₃ typically have efficiencies at 1.3 microns of 50\(^\circ\)/volt-cm, a \(V_\pi\) of 5 V for a 1-GHz 3-dB bandwidth, and a fiber-to-fiber insertion loss of 2-3 dB. In semiconductors, modulation efficiencies can be significantly higher if one designs a tightly guided mode (i. e., one well-suited for on-chip laser coupling, but having a relatively high fiber-to-chip mismatch coupling loss).

The modulation bandwidth of phase and intensity modulators is determined by the dielectric properties of the electro-optic material and the electrode geometry. For structures in which the electrode length is comparable to or shorter than a quarter RF wavelength, it is reasonable to consider the electrodes as lumped and to model the modulator as a capacitor with a parasitic resistance and inductance.

In this case, the bandwidth is proportional to \(1/L\). For most IO materials, lumped-element modulators have bandwidths less than 5 GHz to maintain reasonable drive voltages. For larger bandwidths, the electrodes are designed as transmission lines and the RF signal copropagates with the optical wave. This is referred to as a traveling wave modulator.

The microwave performance of this type of structure is determined by the degree of velocity match of the optical and RF waves, the electrode microwave loss, the characteristic impedance of the electrodes, and a variety of microwave packaging considerations.

In general, semiconductor and polymer modulators fundamentally have better velocity match than LiNbO₃ and thus are attractive for highest frequency operation; however, in recent years, techniques and structures have been developed to substantially improve the velocity match in LiNbO₃ and intensity modulators with 50 GHz bandwidth have been reported.

To achieve intensity modulation, it is generally necessary to incorporate a phase modulator into a somewhat more complex guided wave structure. The two most common devices are the Mach-Zehnder (MZ) and the directional coupler modulator. Figure 9 illustrates the MZ modulator.

 

 

This device is the guided wave analog of the classical MZ interferometer. The input and output Y-junctions serve as 3-dB splitters, and modulation is achieved in a push-pull manner by phase-modulating both arms of the interferometer. The interferometer arms are spaced sufficiently that there is no coupling between them. When the applied voltage results in a pi-radian phase shift in light propagating in the two arms when they recombine at the output junction, the resultant second-order mode cannot be guided and light radiates into the substrate.

The output intensity \(I\) of this device is given by \(I=I_0/2[1+\cos(KLV)]\). The sinusoidal transfer characteristic is unique in IO modulators and provides the unique capability to ‘‘count fringes’’ by applying drive signals that are multiples of \(V_\pi\). This feature has been exploited in a novel analog-to-digital converter. The device can be operated about its linear bias point \(\pi/2\) for analog applications and can also be used as a digital switch.

A variation on this device is a balanced bridge modulator. In this structure the two Y-junctions are replaced by 3-dB directional couplers. This structure retains a sinusoidal transfer characteristic, but can function as a 2 x 2 switch.

 

 

A directional coupler switch is shown in Fig. 10. In the embodiment illustrated, a set of electrodes is positioned over the entire coupler region. The coupler is chosen to be \(L_c\), a coupling length long, so that in the absence of an applied voltage, all light incident in one guide will cross over and exit the coupled guide. The performance of the directional coupler switch can be modeled by coupled mode theory. The application of an electric field spoils the synchronism of the guides, resulting in reduced coupling, and a shorter effective coupling length. For application of a voltage such that \(KLV=\pi\sqrt{3}\), all light will exit the input guide.

In general, the transfer characteristic is given by

\[\tag{14}I=\frac{I_0}{2(1+(KLV/\pi)^2)}\cdot(1-\cos(\pi\sqrt{1+(KLV/\pi)^2}))\]

Directional coupler switches can also be used for analog or digital modulation. They have also been fabricated in matrix arrays for applications in N x N switch arrays. To increase the fabrication tolerance of directional coupler switches, designs based on reversing the sign of index change (delta beta) periodically along the coupler have been developed. The most common device consists of a device 1 – 2 coupling lengths long and a single reversal of the voltage formed by a two-section electrode.

Both Mach-Zehnder and directional coupler devices have been developed in semiconductors, LiNbO₃, and polymers. Devices are commercially available in LiNbO₃. Drive voltages and bandwidths achieved are similar to the values quoted above for phase modulators.

Additional effort has been focused in LiNbO₃ to make devices that are polarization insensitive so that they are compatible with conventional single-mode fiber. Mach-Zehnder modulators have also been formed in glass waveguides. Here a resistive pad is heated to vary the index of the waveguide via the thermo-optic effect.

Another important IO component is the TE-to-TM mode converter. This device, illustrated in Fig. 11, depends on an off-diagonal component \(r_{51}\) of the electro-optic tensor in LiNbO₃ to convert incident TE (TM) light to TM (TE) polarization.

In the converter, a periodic electrode structure is used to create a periodic index change along the waveguide to provide phase matching, and thus coupling, between the TE and TM wave. The period \(\Lambda\) of this index change is given by \(\Lambda=\lambda/(n_\text{TE}-n_\text{TM})\). The coupling efficiency at the phase-matched wavelength is given by \(\sin^2(\kappa{L})\) where \(\kappa=\pi{n^3}r_{51}E/\lambda\) and \(E\) is the applied field.

This type of device can be used for polarization control. In a polarization controller, a phase modulator is placed before and after the converter so that signals of arbitrary input polarization can be converted into any desired output polarization. The converter also serves as the basis for a tunable wavelength filter.

 

 

There are numerous other types of IO intensity modulators that have been reported. These include a crossing channel switch, a digital optical switch, an acousto-optic tunable wavelength switch, and a cut-off modulator.

The first two devices depend on modal interference and evolution effects. The acousto-optic switch utilizes a combination of acoustically induced TE-to-TM mode conversion and TE-TM splitting couplers to switch narrow-optical-band signals. The cut-off modulator is simply a phase modulator designed near the cutoff of the fundamental mode such that an applied field effectively eliminates the guiding index change between the guide and the substrate. This results in light radiating into the substrate.

In addition to the electro-optic devices described above, another common modulation technique employed in III-V materials employs the electroabsorption of electrorefraction effects discussed previously.

Here the bandgap energy of a bulk medium or an appropriately engineered quantum-well medium is chosen to be somewhat higher than the energy of the propagating photons. An applied field directly induces absorption, or a large index shift associated with the change in absorption at higher energy. The latter effect is used interferometrically in directional couplers, Mach-Zehnder modulators, or other designs as an enhanced substitute for the conventional electro-optic effect. The former is used as a single-pass absorptive waveguide modulator.

To achieve low operating voltages, such modulators are usually designed with small waveguide dimensions for tight confinement of the optical mode. This usually leads to a significant insertion loss of ~ 2 – 3 dB/cm when coupling to optical fibers. However, the tight waveguide mode is very similar to the waveguides employed in semiconductor lasers, and hence the primary appeal of the waveguide electroabsorption modulators lies in their potential for integration with semiconductor lasers on a single PIC chip.

 

 

A particular implementation used by Soda et al. is shown schematically in Fig. 12. A 1.55-μm DFB laser structure is mated to an electroabsorption modulator with an InGaAsP core layer having a photoluminescence wavelength of \(\lambda_\text{PL}\) ~ 1.40 μm. The entire structure uses a buried heterostructure waveguide with semi-insulating InP lateral cladding to provide good current blocking with low capacitance for high modulator bandwidth.

Optimization of the modulator core \(\lambda_\text{PL}\) is very important in this device. With proper design, devices have yielded a good compromise between high output power and high modulator extinction ratios with low voltage drive. Typical device designs exhibit mW-level fiber-coupled output power with a \(-10\)-dB extinction ratio at drive levels of 2 – 4 V.

The integrated DFB/electroabsorption modulator mentioned previously provided one illustration of laser integration. Common to most laser/detector/waveguide PICs is the inclusion of a guide containing an amplifying or gain medium, or an absorptive medium for detection, and the requirements of current drive or extraction. The design and processing associated with these guided-wave components relies heavily on a relatively mature technology associated with semiconductor lasers.

The gain section of a semiconductor laser is usually fabricated in a buried heterostructure guide as shown in Fig. 2a, and is driven through a forward biased p-n junction where the layers are usually doped during the crystal growth. With zero or reverse bias, this same structure can function as a waveguide photodetector.

In a DFB laser or a distributed Bragg reflector (DBR) laser, this feature can be used to provide an integrated detector, on the back end of the device external to the cavity, for monitoring laser power.

Alternatively, a separate gain medium external to the laser cavity can be located external to the cavity on the output side to function as an integrated power amplifier for the laser.

Such a structure is shown in Fig. 13, where a DBR laser is followed by a fan-shaped amplifier to keep the amplifier medium at a relatively constant state of saturation. These PICs are termed master-oscillator/power-amplifiers (MOPAs), and can provide Watt-level single-frequency, diffraction-limited output beams from a single chip.

 

 

The challenge of laser integration is to fabricate other guided-wave components without compromising the intricate processing sequence required to make high-performance lasers. Figure 14 shows a balanced heterodyne receiver PIC suitable for coherent optical communications links.

 

 

Here a tunable local oscillator is tuned to an optical frequency offset by a predetermined amount from one of potentially many incoming signals. The beat signals are generated in the integrated photodetectors, whose signals can be subtracted for noise reduction, and then electrically amplified, filtered, and fed to a decision circuit.

This PIC combined five different types, and a total of seven guided-wave optical devices: two tunable Bragg reflection filters, an MQW optical gain section, an electrically adjustable phase shifter, a zero-gap directional coupler switch, and two MQW waveguide photodetectors.

It also demonstrates self-aligned connections between the buried heterostructure guides, which offer current access and semi-insulating InP lateral current blocking, and the low-loss semi-insulating InP-clad rib guides used in the S-bends and input port.

The processing sequence for PICs of this complexity has been described in some detail in the literature , and can be carried out following most of the same steps used in commercial semiconductor laser fabrication.

Tuning of the DBR lasers, as used in the PIC above, is accomplished by injecting current into the (transparent) Bragg reflectors, shifting their index via the plasma and anomalous dispersion effects discussed under Carrier Effects in this tutorial.

This shifts the wavelength of peak Bragg reflectivity, thereby selecting different longitudinal modes for the laser. The laser can also be continuously tuned by shifting the frequency of any particular longitudinal mode by injecting current to provide an index shift in the (transparent) phase section of the laser. Detectors in PICs of this type often employ for absorption the same layers used for gain in the laser, and typically have capacitance of several pF dominated by the contact pads rather than the depletion capacitance of the p-n junction.

Early experimental prototypes of PICs of this type have demonstrated total on-chip losses including propagation losses, bending losses, radiation losses at the coupler and at the active/passive detector transitions, and any departures from 100 percent quantum efficiency in the detectors, of ~ 4 dB, providing encouragement for even more complex integrations.

This PIC demonstrated error-free heterodyne reception of frequency-shift-keyed digital signals with sensitivities of \(-40\) dBm at 200 Mb/s measured in free-space outside the chip. PICs such as this may in the future offer high selectivity in multichannel optical broadcasts, much in the same way radio receivers operate today.

 

5. Applications of Integrated Optics

Digital Transmission

The performance metric in digital optical fiber transmission is the ability of a transmitter to deliver a signal to the receiver on the end of the link in a manner such that the receiver can clearly distinguish between the ‘‘0’’ and ‘‘1’’ state in each time period or bit slot. Binary amplitude-shift-keyed transmission (ASK) is by far the most common format in commercial systems, but research systems also employ frequency-shift-keyed (FSK) and phase-shift-keyed (PSK) formats as well. A decision circuit at the receiver must distinguish between ‘‘0’’ and ‘‘1 , ’’ and this circuit will be more susceptible to noise when the ‘‘0’’ and ‘‘1’’ level difference is reduced, or when the time over which this difference is maintained is reduced below the full bit period.

The performance of a transmitter is thus governed by its rise and fall times, its modulation bandwidth or flatness of response to avoid pattern-effects, and its output power. Furthermore, the spectral characteristics of its optical output can impair transmission. Examples of the latter include unnecessary optical bandwidth, as might be present in an LED or a multi-longitudinal-mode laser, that can produce pulse spreading of the digital pulses due to dispersion in the transmission fiber.

While transmission sources include current-modulated LEDs, for speeds higher than, 100 Mb/s semiconductor lasers are used, and IO technology has played a fundamental role in the evolution of semiconductor laser technology. In very high speed systems (typically > 1 Gb/s), dispersive pulse distortion can cause severe degradation with directly modulated lasers unless devices which emit only one longitudinal mode are employed. The incorporation of gratings in DFB and DBR lasers has produced sources with exceptional spectral purity and allow multi-Gb/s transmission over intermediate distances ( < 100 km) in conventional fiber.

The advent of fiber amplifiers has raised interest in longer transmission spans, and here the unavoidable frequency excursions that result from directly modulating even a single-longitudinal-mode laser again lead to pulse spreading and distortion. In these instances, a CW laser followed by an external modulator is a preferred source.

The integrated DFB/electroabsorption modulator, as discussed previously, provides such a source. These PICs have demonstrated error-free transmission in excess of 500 km in dispersive fiber at 2.5 Gb/s. However, even these devices impose a small residual dynamic phase shift on the signal due to electrorefractive effects accompanying the induced absorption in the modulator. This can be especially problematic with typical percent-level antireflection coatings on the output facet, since this will provide time-varying optical feedback into the laser and further disrupt its frequency stability.

The highest performance digital transmission has been achieved using external LiNbO₃ Mach-Zehnder modulators to encode a CW semiconductor laser. Modulators similar to that in Fig. 9 have been constructed to allow separate traveling-wave voltages to be applied to each arm of the modulator in a push-pull configuration. This device design can produce a digitally encoded signal with zero residual phase shift or chirp.

Such a source has only its information-limited bandwidth and generally provides nearly optimized performance in a dispersive environment. The Mach-Zehnder can also be driven to intentionally provide positive or negative chirping, and transmission experiments over distances up to 256 km in dispersive fiber at bit rates of 5 Gb/s have revealed a slight increase in performance when a slightly negative chirp parameter is used. Recent work has focused on the integration of semiconductor lasers with Mach-Zehnder modulators using electrorefraction from the QCSE. Such a source is expected to provide a convenient and high-performance digital transmitter.

 

Analog Transmission

A second application area that is expected to use a large number of IOCs is analog fiber optic links. Analog fiber optic links are currently being used to transmit cable television (CATV) signals at the trunk and supertrunk level. They are also being used for both commercial and military antenna remoting. Analog fiber optic links are being fielded for these applications because of their low distortion, low loss, and low life-cycle cost when compared to more conventional coaxial-cable-based transmission systems.

 

 

An analog fiber optic link using IOCs is typically configured as shown in Fig. 15. A high-power CW solid-state laser, such as a 150-mW diode-pumped YAG laser operating at 1319 nm, is typically used as the source in order to maximize dynamic range, carrier-to-noise ratio, and link gain.

An interferometric modulator, such as a Mach-Zehnder interferometer or a Y-fed balanced bridge modulator, is typically used to modulate the light with the applied RF or microwave signal via the linear electro-optic effect. Current analog links for CATV signal distribution utilize a 1-GHz Y-fed balanced bridge modulator biased at the quadrature point (linear 3-dB point).

A predistortion circuit is required to minimize third-order distortion associated with the interferometric modulator response. The CATV modulated signal can be transmitted on both output fibers of the device. Analog links for antenna remoting typically fit into one of two categories.

Certain applications require relatively narrow passbands in the UHF region while other microwave-carrier applications require very broadband (several GHz) performance. Y-fed balanced bridge modulators biased at the quadrature point are again used to perform the electrical-to-optical conversion, with the electrode structure tailored to the application. In both narrowband and broadband applications, 20 – 30 dB preamplifiers are typically utilized to minimize the noise figure and maximize the RF gain of the link.

Two important modulator parameters are the insertion loss and the half-wave drive voltage, both of which impact the link gain and dynamic range. Fully-packaged Y-fed balanced bridge modulators with 2.5 – 4.0-dB insertion loss are now readily achieved in production for both UHF and microwave bandwidths.

A trade-off typically needs to be made between half-wave voltage and bandwidth for both lumped-element and travelingwave electrode structures. Commercially available lumped-element LiNbO₃ interferometric modulators typically have half-wave voltages of ~ 5 V for 600-MHz, 1-dB bandwidths. Commercially available traveling-wave LiNbO₃ interferometric modulators typically have half-wave voltages of ~ 8 V for 12-GHz, 3-dB bandwidths. The half-wave voltages of LiNbO₃ traveling-wave modulators can be reduced by more than a factor of two using a velocity-matched electrode structure.

In order to be used in practical systems, it is critical that the integrated optical modulators have well-behaved flat frequency responses in the band of interest. Modulators for CATV signal transmission and UHF antenna remoting typically required that the amplitude response and the phase response be flat to \(\pm\)0.25 dB and \(\pm\)2 degrees, respectively.

The frequency response of an integrated optical modulator is a function of both the device design and packaging parasitics. Care must be exercised in designing modulators since LiNbO₃ is both a piezoelectric and an acousto-optic material. Early LiNbO₃ modulators typically had 1 – 3 dB of ripple in the passband due to acoustic mode excitation.

When packaging lumped-electrode devices, it is also critical to keep terminating resistors and wire bonds short to minimize stray inductance and capacitance. When packaging traveling-wave modulators, it is critical to control the impedance of the launch, the transitions, the device, and the termination. Through proper device design and packaging, it is possible to achieve well-behaved frequency responses in both lumped-electrode and traveling-wave devices as shown in Figs. 16 and 17.

 

 

 

An additional issue that impacts IOC modulator design for analog links is harmonic and intermodulation distortion. Most modulators used in analog links are interferometric in nature with a sinusoidal transfer function. By operating the device precisely at the quadrature point, all even harmonics can be suppressed. Second-harmonic distortion less than \(-75\) dBc is easily achieved using an electronic feedback loop for modulator bias.

Alternative techniques are being investigated to laser trim one leg of a Mach-Zehnder to bring the modulator to quadrature. Third-order distortion due to the sinusoidal transfer function of the interferometric modulator also poses a problem, but the transfer functions are very well-behaved and predictable, and this distortion can be suppressed to acceptable levels using electronic predistortion or feed forward techniques.

Forty- and eighty-channel CATV transmitters operating at 1300-nm wavelengths with APE LiNbO₃ modulators are currently being fielded. Compared to coaxial transmission, which requires transmitters every 500 meters, the fiber optic systems can transmit over distances up to 50 km without repeaters.

Similarly, externally modulated analog fiber optic links are currently being fielded for military and commercial applications. UHF links with 115 dB/Hz\(^{2/3}\) dynamic range, 4-dB noise figure, and unity gain have been demonstrated using commercially available hardware. These systems will maintain this quality of transmission over temperature ranges of \(-25\) to \(+50^\circ\)C. Microwave links with 2 – 18-GHz frequency response, 114-dB/Hz\(^{2/3}\) spurious-free dynamic range, and input noise figure of 22 dB can also be achieved using commercially available hardware.

 

Switching

Arrays of IO switches have been proposed and demonstrated for a variety of space switching and time-multiplexed switching (TMS) applications. In space switching, it is generally necessary to formulate the switches in a nonblocking geometry, and the reconfiguration time can be relatively slow. This requirement led to the development of crossbar switches in which an N x N switch contains \(N^2\) IO switches and \(2N-1\) stages and from 1 to \(2N-1\) cross points. Typically \(N=4\) in LiNbO₃ and in InP.

More recently, much attention in IO switch arrays has shifted to the dilated Benes architecture which is only rearrangeably nonblocking but reconfigurable in short (nsec) times suitable for TMS, and has the advantage of requiring substantially fewer switches and a constant number \(2\log_2N\) of crosspoints.

 

 

A schematic of an 8 x 8 dilated Benes switch composed of two separate IO chips is shown in Fig. 18. This device uses waveguide crossovers. The performance of switch arrays of the dilated Benes architecture are much more forgiving than crossbar switches to the degradation of individual switches. The device shown contains 48 delta beta switches driven by a single-voltage arrangement. The switching voltage at 1.3 microns was \(9.4+- 0.2\) V, the insertion loss varied from \(-8\) to \(-11\) dB (93 percent of the 256 paths through the switch were within \(\pm\)1 dB), the crosstalk levels in individual switches ranged from \(-22\) dB to \(-45\) dB, and the reconfiguration time was 2.5 nsec.

An advantage of these types of IO switch arrays is that they are data-rate transparent. That is, once the switch is reconfigured, the data stream through the device is simply the passage of light and can easily be multi-Gbit/sec. Crossbar switches were commercially available and other types of arrays were under advanced development.

 

Fiber Optic Gyroscopes

Another application that may require large quantities of integrated optical circuits is the fiber optic gyroscope (FOG). A FOG is one form of a Sagnac interferometer, in which a rotation rate results in a phase shift between clockwise- and counterclockwise-propagating optical fields. The most frequently utilized FOG configuration, which was first proposed by workers at Thompson CSF in the mid-1980s, is presented in Fig. 19.

 

 

FOG IOCs are typically fabricated in LiNbO₃ using the annealed proton exchange (APE) process, although titanium-diffused IOCs with surface plasmon polarizers have also been utilized.

The IOC performs four primary functions in the fiber gyroscope. First, the Y-junction serves as the loop coupler splitting and recombining the clockwise- and counterclockwise-propagating optical fields. Second, the IOC functions as a high-quality polarizer. Third, a ninety-degree phase dither (at the eigen frequency of the fiber coil) is typically applied to one of the integrated optical phase modulators. This approach keeps the Sagnac interferometer biased at the 3-dB point where it is linear and most sensitive to rotation. Finally, in a closed-loop FOG configuration, one of the phase modulators functions as a frequency shifter. A serrodyne signal (sawtooth wave) is applied to the phase modulator to effectively cancel the shift due to the rotation.

The output signal from a fiber gyro at rest is the sum of white receiver noise, primarily dependent on the amount of optical power arriving at the detector, and an additional long-term drift of the mean value. The long-term drift in a FOG associated with a residual lack of reciprocity typically limits the sensitivity of the FOG to measure low rotation rates. Another important characteristic of a gyro is the scale factor, which is a measure of the linearity between the actual rotation rate and the gyro response. The critical performance parameters for a FOG IOC are presented in Table 1. The performance of 800- and 1300-nm APE LiNbO₃ FOG IOCs that are currently in production is also presented in this table.

 

 

One application of the FOG is inertial guidance, requiring a FOG with a bias drift, 0.01 deg/h and a scale factor accuracy, < 5 ppm. A 1300-nm LED or an erbium-doped fiber is typically used as the light source. A large coil of polarization-maintaining fiber (typically 1 km of fiber wound in a 15 – 20 cm diameter coil) and precise source spectral stability are required to achieve the desired sensitivity. The fiber is typically wound in a quadrupole configuration to minimize sensitivity to temperature gradients. With recent improvements in optical sources, integrated optics, and fiber coil winding technology, it is now possible to achieve inertial grade FOG performance over a temperature range of \(-55\) to \(+95^\circ\)C.

A second tactical-grade FOG design is more typical to aerospace applications, with bias drift and scale factor accuracy requirements ranging from 0.1 to 10 deg/h and 10 to 1000 ppm, respectively. These systems are typically designed for operation at 810 – 830 nm to make use of low-cost multimode 830-nm AlGaAs laser diodes as used in consumer electronic products. These systems typically utilize 2- to 5-cm-diameter fiber coils with 100 – 200 meters of either polarization-maintaining or single-mode fiber.

A third very low-cost, low-grade FOG design for automotive navigation is also nearing production. The required bias drift is only 1000 deg/h, and a closed-loop configuration is unnecessary since the scale factor accuracy is only 0.1 percent. Current designs to achieve low cost include low performance IOCs, laser, and fiber couplers, fully automated FOG assembly and test procedures, and only, 50 m of single-mode fiber. More advanced IOCs, including four-port designs that integrate the source/detector coupler into the IOC, are also being considered to reduce component count.

 

WDM Systems

Wavelength division multiplexing (WDM), by encoding parallel data streams at different wavelengths on the same fiber, offers a technique to increase transmission capacity, or increase networking or switching flexibility, without requiring higher speed electronics to process each channel. IO will be an enabling technology in both WDM transmitter design and in demultiplexer design.

 

 

Figure 20 shows a WDM transmission PIC, following the concept introduced by Aiki et al, where the outputs of a number of single-frequency lasers are combined into a single waveguide output port. The PIC shown here combines the outputs of four independently modulatable and independently tunable MQW-DBR lasers through passive power combining optics, and also includes an on-chip MQW optical output amplifier to partially recover the inherent losses of the power combining operation.

This WDM source PIC has successfully demonstrated a high-speed WDM transmission capability. The PIC was mounted in a fixture with SMA connectors to 50 \(\Omega\) microstrip leading to ceramic stand-offs with bond wires to the PIC contacts, and a thermoelectric cooler to provide frequency stability.

One concern in PICs employing optical amplifiers, just as in the integrated laser/modulator, is the need for good anti-reflection coatings since the source sees the facet reflection after a double-pass through the amplifier. Another area of concern is the cross-talk between channels, either through electrical leakage on chip or through amplifier saturation ef fects since all sources share the same amplifier.

Cross-talk in this PIC has been evaluated a large-signal digital transmission environment. The lasers were each simultaneously modulated with a pseudo-random 2 Gb/s signal, and the combined signal at an aggregate bit-rate of 8 Gb/s was sent over a 36 km transmission path of conventional 1.3 μm dispersion-zero fiber.

A fiber Fabry-Perot interferometer was used for channel selection. The PIC used in this experiment had two MQW-DBR lasers at each of two distinct zero-current Bragg frequencies shifted by, 50 Å, and one of each pair was then tuned by, 25 Å to yield four channels with a spacing of, 25 Å.

There was virtually no cross-talk penalty without fiber, indicating that the level of cross-talk is not significant in a direct sense, i. e., in its impact on the intensity modulation waveforms at 2 Gb/s.

When the 36 km link of dispersive fiber is inserted, a small penalty of ~ 1 dB was observed, suggesting a small dispersion penalty probably resulting from electrical cross-modulation from the other current drives on the PIC. This result indicates that cross-talk is not a severe problem, and simple design improvements could reduce it to inconsequential levels for digital applications. PICs of similar design have been extended to include 16 lasers with 8-Å spacing and an individual electroabsorption modulator for each laser.

 

 

IO demultiplexers are also expected to be important for WDM. Figure 21 shows a demultiplexer design that employs star couplers and arrayed waveguides, and it has been successfully executed both in the Si/SiO₂ and InP-based technologies.

In this design each input to a primary star coupler expands in the free-space (in-the-plane) portion of the star to uniformly illuminate each output waveguide of the primary star. The path lengths of each guide in the array between the primary and secondary star are incremented in length by an integral multiple of some base wavelength.

At this wavelength, upon arrival at the input to the second star, each wave has the same phase relation to its neighbors that it had at the output to the first star coupler, and reciprocity demands that this focus the beam, as a phased array, back to the corresponding single waveguide at the secondary star output.

However, a slight change in wavelength will produce a phase tilt across this phased array, and the focus will go to a different output waveguide. In this manner, the device operates as a demultiplexer. The incremental length difference between adjacent guides in the region between the two stars functions just as a grating would in a bulk-optic equivalent of this device.

The performance of these devices has been extraordinary. Out-of-band rejection in excess of 25 dB has been obtained, with individual channel bandwidths in the 1-nm-scale and fiber-to-fiber insertion losses of only several dB. Device sizes up to 15 channels have been demonstrated, and these devices are also amenable to fiber ribbon array connections. IO devices that perform this function will be instrumental for cost-effective WDM deployment.

 

Optical Storage and Display

IO devices are of special interest for propagation and manipulation of visible light for storage and display applications. For optical data storage, there is interest in operating at wavelengths shorter than obtainable from GaAs lasers to increase storage capacity (which is proportional to diffraction limited laser spot size).

For display applications, multiple colors are needed. The method receiving most interest is to frequency double diode lasers in materials with high nonlinear optical coefficients. Waveguide doubling can be quite efficient because of the long interaction length and high optical power densities.

Both channel and slab guide doublers have been demonstrated. As discussed under Nonlinear Effects in this tutorial, periodic domain reversal has been obtained in LiNbO₃ by the application of pulsed electric fields with segmented contacts. This approach has recently yielded waveguides capable of producing 20.7 mW of blue light from 195.9 mW of fundamental at 851.7 nm. For these devices to find practical application, considerable effort in robust packaging of laser-doubling waveguide systems will be required.

 

The next tutorial gives a detailed introduction to photosensitive fibers.