2.1

Loaded-Line MZ Modulator

The Mach-Zehnder modulator (MZM) consists of an optical splitter feeding into a pair of electro-optic waveguide phase modulators. The capacitive electrodes are set-up to operate in series push-pull², whereby a built-in back-to-back connection, via an in-grown n-doped backplane, creates a series-chain, halving the effective capacitance and splitting the RF potential, producing balanced, anti-phase phase-modulation contributions. Optical recombination produces a raised-cosine output-intensity function with no net phase modulation (i.e. zero chirp) and characterized by the half-wave (ON-OFF) voltage Vπ.

In travelling-wave versions of such a modulator, the aim is to match the velocity of a forward-propagating RF wave to the velocity of the optical modulation group which it is creating. Velocity-matching permits the modulation to accumulate monotonically along the entire length of the electrode, which needs to be long because electro-optic effects of most materials are very weak.

Our devices achieve velocity-match by means of a capacitively loaded line configuration. This slow-wave structure is required in semiconductor materials because, while the microwave and optical velocities generally have similar values, the ~50% free-space loading from the superstrate makes a pure coplanar RF wave run faster than the fully buried optical wave. The modulation electrodes are divided into a linear array of short segments isolated by passive spaces (see Figure 1 below). These sample the line-voltage via air-bridge connections to accumulate the modulation effect while adding shunt capacitance to slow the RF wave. For preference we use CPW (coplanar waveguide), which has double surface ground-planes. These are strapped together periodically by bond-wires to prevent the asymmetric loading generating higher-order RF modes. The loaded characteristic impedance and RF velocity are both subject to the same slow-wave factor (see section 5). In GaAs a good velocity-match can be designed to coincide with impedance a little under 50Ω.

Figure 1

Schematic of loaded line travelling wave modulator based on strapped CPW transmission-line, with waveguide cross-section showing typical optical mode-profiles

It will be apparent from the schematic of Figure 1, that the essential co-linearity of RF and optical paths results in a topological conflict between the RF and optical inputs. In practice, the RF is usually input from the side of the device and bent through 90° while the optical line is kept perfectly straight. For frequencies above 30GHz, this is not good enough, hence our recent emphasis on optical path-folding so that the RF input run can be kept as short and straight as possible and is in-line with the optical travelling-wave section from the outset. This requires the extensive use of optical 90° and/or 180° bends. The availability of such bends enables extension of the GaAs modulator technology to more advanced configurations, such as IQ modulators (used for QPSK and single sideband formats) as well as monolithic arrays with all RF inputs at one end of the module and optical access entirely at the other (Figure 2).

Figure 2

Folded-path dual-parallel modulator (1x2 array, SSB or IQ modulator) using 90° and 180° waveguide bends.

Arguably, monolithic arrays of more than two modulators would hardly be possible without such provision. It allows four or more channels per chip with only minimal additional complication, and in a package of similar size.

This topology is particularly suitable for phased-array applications with fiber-optic antenna-head remoting. Within the modulator package itself there are further gains in compactness due to the concentration of space-hungry fiber interfaces at one end.

Clearly such a concept is not viable if the 90°corner and 180° ‘U’ bends contribute more than a negligible loss, or cause mode-coupling. These need to be low-loss and designed carefully to maintain good optical characteristics. There is also a potential additional overhead of straight waveguide loss from the input-run which, as illustrated in Figure 2 is largely parasitic. In practice, this length of waveguide can be used to accommodate DC phase control electrodes, reducing the lengths needed in the output foldback sections.

3.1

‘Matched’ Corner Bends

The waveguides used for the bend-sections are of the deep-etched ridge type. Unlike the rib waveguides illustrated in Figure 1, these are etched fully through the GaAs core-layer and have a very high lateral index-contrast (both types of waveguide are used within the device for different purposes, with low-loss taper transitions between). Thus, bend behavior is determined by mode-coupling and mode-interference mechanisms rather than by radiative loss.

Bend-design emphases found in literature include the matched bend³ (in which the total path-length is matched to a dominant intermodal beat-length), and gradation of curvature, for example using multi-parameter geometric functions with computer-optimization routines; e.g. cubic-splines⁴, euler-spiral⁵ or other non-explicit function. Our bends use both principles, but with our own proprietary functions and methods to identify the lowest-loss and most process-tolerant designs.

Our design simulation method closely replicates the way the bends are constructed in the CAD layout, as a succession of angled straight segments. Successive pivoting of the graphics coordinate system allows the bend to be displayed semirealistically as in Figure 3, showing up a ‘whispering gallery’ type of behavior.

Figure 3

Simulation detail and measured losses of optimized bends.

While it is possible to devise one-piece U or S bends it is more usual to base the waveguide configuration on a concatenation of identical standard optimized bends. These are mainly 90° corner bends, but other types, e.g. 45° (allowing an orthogonal mutual crosspoint) also have their uses. Our optimization of a bend-profile is not based on simple best performance (e.g. lowest loss, least higher-mode residue) but on worst-case performance over a range of likely process and wavelength variation.

Different bend optimizations have been validated for several recent device designs using dedicated test waveguide structures. Each new device design carries such ancillary test-sets for the exact bend used within the design, and test bends typically comprise concatenations of many corner-bends. Losses are measured by a fabry-perot resonance technique⁶ using test-pieces without antireflection coatings. Without exception, the excess loss per bend has been measured at between 0.012dB and 0.02dB, averaging about 0.015dB (Figure 3), consistent with models, and with low sensitivity to wavelength as seen in Figure 3.

3.2

Couplers, Splitters and Recombiners

Most guided-wave optical circuits will require splitters and recombiners to perform basic interferometry. Historically, Y-branches and waveguide directional-couplers were the first port-of-call for these functions – and this is still the case for lithium niobate modulators based on diffused channel waveguides. In semiconductors, directional couplers are problematic because the weak lateral confinement required demands shallow etches with consequent high sensitivity to etch depth.

The preferred multimode interference (MMI) couplers⁷ are lengths of wide, deep-ridge waveguide with suitably placed waveguide ports. When etched fully through the GaAs core-layer the properties are stable against any variability in the etch-depth, though a tight tolerance on the width is mandatory. The MMI concept makes use of the tidy geometrical relationship between the propagation velocities of waveguide modes when the walls are of high reflectivity (so that all modes are subject to the same effective width) – a condition well satisfied by the deep-ridge GaAs waveguides used.

Any launch condition can be decomposed as a superposition of the MMI modes and radiation. With the launch phaserelationship restored cyclically at multiples of a fundamental beat-length along the MMI waveguide, lens-like re-imaging of the launch profile results. Most useful is the fact that a double image appears at half this re-imaging length. This provides 2x2 split and recombination functions in a process-tolerant structure.

Figure 4 shows propagation modelling on a 2x2 MMI recombiner which has been used in our GaAs modulator designs; these can also be used for the optical split function. 2x2 MMIs of this type provide a 1:1 split/recombination ratio which is closely maintained even when the geometry (or wavelength) is far from the design-center; the excess loss can be significant before the split-ratio is degraded. This means that (unlike the directional-coupler) the modulator extinction-ratio is maintained out of band, even if loss increases.

Figure 4

Simulation and loss measurements on MMI optical recombiners

Much of the subtlety in guided wave design using components such as these lies in choice of geometry to maximize tolerance to process and wavelength variation. As with bend designs, MMIs are dimensioned to this end.

The chart in Figure 4 compares the measured excess loss with simulation for one split or recombine unit (without I/O bends). The test-set for this used a concatenation of 10 full-cross MMI units (the simulation also calculates the full concatenation). Excellent agreement with measurement is obtained; moreover, the excess loss of an optimized MMI is clearly very low.

Despite the virtues of MMI couplers, there remains a sensitivity to wavelength in terms of excess loss (though not splitratio) which limits the modulator design to a specific wavelength band. Broader waveband operation, for example encompassing 1300nm and 1550nm bands, would require the exclusive use of a non-interferometric split/recombine structure such as the Y-branch. For this reason, we continue to improve our Y-branch designs (Figure 5). These are of low loss but do not offer the excellent process tolerance and reliably high modulator extinction-ratio of MMIs.

Figure 5

Y-branch simulations contrasting sub-optimal and optimized tapering of the divergent waveguides.

5.1

Velocity-match at short wavelength

Optical dispersion must also be considered when re-designing for a new wavelength or for a broad wavelength band, as this determines the waveguide group-velocity. The RF power advantage of short wavelength operation is very significant, as has been shown. Near the semiconductor band-edge, dispersion is high, resulting in significantly enhanced group-index values (see Figure 6); this strongly influences the design approach for a short-wavelength V-band modulator and requires a different velocity-matching strategy compared to 1550nm. A higher slow-wave factor is required, and this can be readily achieved with the flexibility of the GaAs/AlGaAs semiconductor system by using a coplanar base-structure of increased impedance, by adjusting the waveguide geometry and by specifying epitaxial layers with thinner depletion and increased aluminium.