Emil S. Koteles
Advances in monolithically integrated planar waveguide etched grating devices will increase manufacturing efficiency and greatly lower cost.
Long-haul wavelength-division multiplexing (WDM) systems are too expensive for metropolitan markets. However, recent advances in monolithically integrated multiplexers and demultiplexers (planar waveguide etched grating devices) and techniques for flattening dense WDM channel spectra without increasing crosstalk promise cost-effective solutions. As a result, inexpensive WDM components can be made using photolithographic-processing techniques with increased manufacturing tolerances.
Until recently, WDM signal processing (generation, detection, multiplexing, demultiplexing, and so on) has been accomplished using discrete optical components. These perform well but their fabrication is labor intensive and therefore expensive to manufacture. The next generation of optical components will be based on integration technology. The planar waveguide is the technology of choice for monolithic photonic integrated circuits (PICs) because such circuits can be easily fabricated using standard semiconductor photolithographic processing techniques
FABRICATION CHALLENGESTwo-dimensional (planar) optical waveguides can be fabricated using a variety of materials, although glass and semiconductors are the most common. By increasing the refractive index slightly (for example, through the use of ridges), light can be directed in the plane (see Fig. 1).Some two decades ago, planar waveguides were first proposed for demultiplexing and multiplexing WDM wavelengths. The first spectrometer-on-a-chip was based on echelle-grating technology, itself over 100 years old. Light from a single-mode fiber is coupled into the edge of the chip at the input waveguide. The light follows the guide to its end, where it begins to expand in the two-dimensional plane until it reaches the grating etched through the waveguide. The grating reflects and disperses the light and focuses it onto an arc located near the input waveguide. There, a series of output waveguides collects the dispersed light and transports it to the edge of the chip, where it is either detected or collected by other optical fibers (see Fig. 2).
Although an elegant solution, the etched grating demultiplexer suffered from some serious difficulties. Since the planar waveguide was inherently birefringent, the demultiplexer was sensitive to the polarization of the optical signal. Further, the grating was fabricated by etching completely through the waveguide structure. This required a deep etch that was very smooth and vertical. Unfortunately, this processing requirement was beyond the technical capabilities of the time. For these reasons, an alternative technology, the phasar or arrayed-waveguide (AWG) demultiplexer, was developed by Meint Smit of Delft University of the Netherlands and others.1 This device solved these issues but had other drawbacks (see "How echelle gratings and AWGs demultiplex," p. 96).Recently, advances in etching technologies and breakthroughs in polarization control have brought etched (echelle) grating demultiplexers back into contention.2 It is now possible to etch several tens of micrometers into semiconductors, and some other materials, with unparalleled verticality and smoothness. Even several years ago, typical losses of etched turning mirrors in Indium Phosphide(InP)-based waveguides (a good indication of the quality of the etch) ranged from 0.5 to 1 dB.
With the development of the integrated polarization compensator (effectively an additional prism-like, integrated, dispersive element placed in the area between the grating and the input/output waveguides), polarization is no longer an issue with etched grating demultiplexers.3 Advanced techniques for eliminating birefringence in planar waveguides include a new semiconductor waveguide structure that reduces birefringence to negligible levels. The innovation also decreases coupling losses significantly and increases wafer yields by reducing the complexity of the waveguide layers.
DEALING WITH CROSSTALK
Another concern with planar waveguide demultiplexers is the issue of ultimate crosstalk due to the presence of adjacent and nonadjacent wavelengths. Crosstalk, in both etched grating and AWG demultiplexers, is generated by errors in the phases of optical signals originating from the elements of the dispersive device. In the case of the etched grating, these are the reflecting facets; for AWGs they are the ends of the waveguides that act as light sources.
Phase errors in etched gratings result solely from errors in the positions of the grating facets, and ultimately depend on the accuracy with which these facets can be fabricated. The accuracy of these structures is simply a function of the quality of photolithographic masks. A grating can be designed with a very high degree of accuracy in a computer. However, transferring that design onto a photolithographic mask introduces errors due to the limitations of the e-beam or other mask-writing machine. These errors are of two fundamental types: rounding errors resulting from the finite step size of the mask machine and stitching errors, which occur when small write fields are stitched together to form a large element (say the grating in the echelle grating demultiplexer or the array of waveguides in the AWG).
Rounding errors produce a more or less constant background (nonadjacent crosstalk) that decreases as the step size of the mask machine is reduced. A 100-nm step size produces a background of approximately -22 to -25 dB, while a 25-nm step size generates approximately a -35 dB background. The best e-beam machines have step sizes as low as 1 nm, suggesting that backgrounds of better than -45 dB are possible. Stitching errors produce mode structure on the sides of the main channel spectra that die away in intensity as a function of distance. These "ghosts" can be eliminated by using only one field (no stitching) or by increasing the field size. This cure causes some slight broadening of the main channel spectra, but no sidemodes. It is possible for an etched grating demultiplexer in InP to be written in a single field since the grating size is comparable to the largest field size (about 1 mm) in state-of-the-art e-beam writers.
Arrayed waveguide gratings have exactly these issues with respect to rounding and stitching errors. However, since the dispersive element in an AWG (the array of waveguides) is so much larger than a grating, it is unlikely that it can be written in a single field. Furthermore, phase errors can be generated in the waveguides themselves as the light travels from the entrances to the exits of the waveguides. Coupling between waveguides, nonuniformities of the waveguides themselves (due to material or ridge-processing nonuniformities), temperature nonuniformities, and other effects will introduce additional phase errors. Thus, in principle, etched grating demultiplexers are capable of much lower crosstalk than AWGs.
LOSS AND RELIABILITY FACTORS
Insertion loss is the total of on-chip propagation and structure loss due to turning mirrors, grating, and other components, along with coupling loss. Coupling losses can be reduced to 1 to 2 dB by shrewd waveguide design and on-chip losses of several dB are possible. Thus insertion losses of -5 to -10 dB are possible for both etched grating and AWG demultiplexers. Note that, unlike discrete component demultiplexers, insertion losses in planar waveguide demultiplexers are very uniform (less than about 2 dB across the whole wavelength range) and do not scale directly with channel count. Thus planar waveguide demultiplexers are the technology of choice for high channel count (for example, greater than 16 wavelengths) WDM systems.
Absolute channel accuracy can be controlled by adjusting the waveguide temperature (since the waveguide refractive index is temperature sensitive). Thus, it is possible to lock all of the wavelength channels of the demultiplexer to the ITU (International Telecommunications Union) WDM grid using a single feedback mechanism. This temperature sensitivity lowers manufacturing costs because slightly off-spec demultiplexers can be temperature tuned onto the ITU grid.
Reliability is a function of the stability of the materials and other components used in the manufacture of demultiplexers. Glass and semiconductor waveguides are inherently stable. Other materials have not yet demonstrated this important feature. Successful answers to the challenges of demonstrating the reliability of techniques for coupling light onto and off of planar waveguide chips and for controlling the chip environment have recently been confirmed by several manufacturers of AWGs.
BOTTOM LINE DRIVES RESULTS
Cost will be the most important issue for future WDM systems. It is closely related to the manufacturability of the components. An important aspect is fabrication yield. The preferable way of increasing yield—not only of filters but also of lasers required to meet ITU specifications—is to alter the filter shape so that wavelength tolerances are increased.
Typically, channel spectra possess Gaussian shapes with relatively sharp peaks. These shapes require that in a WDM network environment in which the optical signal passes through several filters, each filter, as well as the laser wavelength, must be closely matched to the ITU grid. If the channel shape were rectangular rather than Gaussian, then wavelength tolerances could be increased because the flatness of the peak would ensure that light would pass undiminished, even if wavelengths were slightly off the ITU wavelength. Furthermore, such filters would be more tolerant to variations in parameters such as birefringence and temperature.
A number of techniques have been proposed and demonstrated to achieve this highly desirable result. Generally, such filters were either difficult to fabricate, required mode-altering structures on the input or output waveguides, or increased adjacent channel crosstalk because the full width at half maximum (FWHM) of the wings of the channel spectra had a tendency to increase dramatically. The Telcordia FWHM at -30 dB of a "flattened" filter is almost double that of the Gaussian, suggesting that such broadening was realistic and inevitable.
A recently developed procedure involves a precise dithering of phases of the grating facets, which permits a flattening of the top of the channel spectra without increasing adjacent crosstalk. In fact, the sides of the spectra can be steepened so that the FWHM at -30 dB is identical to that of the original Gaussian. At every point, the FWHM of the flattened filter exceeds
Telcordia specifications (it is larger than Telcordia FWHMs at -1 and -3 dB and smaller at -20 and -30 dB). The importance of this result lies not only with the superior shape achieved but also in how it was accomplished. No additional processing procedures are required; both the input and output waveguides remain single mode (thus allowing bidirectional use), and the technique can be used both with etched grating and AWG multiplexers and demultiplexers in any material system (see Fig. 3).
The exact shape of the channel spectrum can be readily tailored for specific uses. In the example in Figure 3, the FWHM at -1 dB was doubled with no increase in adjacent channel crosstalk. Probably its most important use will be to increase manufacturing tolerances for WDM filters and lasers, thereby decreasing component costs for WDM optical networks.
Although etched grating demultiplexers have not reached the same level of maturity as AWGs at the moment, the issues preventing them from advancing in the past have now been resolved. They will achieve similar, or better, levels of performance compared with AWGs with higher yields and lower costs. A simple calculation suggests that for approximately $10,000, it would be possible to fabricate approximately 2000 filters (that is, about 120 x 16 channel, unmounted, InP etched grating demultiplexers) using 3-in. diameter InP wafers. With a 50% yield and typical packaging costs, selling prices as much as an order of magnitude lower than the present cost of approximately $1000 per wavelength, typical of discrete thin film filter demultiplexers, will be possible.
REFERENCES
- M. K. Smit, IEEE J. Sel.Topics in Quant. Elect., 2, 236 (1996).
- E. S. Koteles, Fiber and Int. Optics, 18, 211 (1999).
- J. J. He, E. S. Koteles, B. Lamontagne, L. E. Erickson, A. Delâge, and M. Davies, Phot. Tech. Lett., 11, 224 (1999).
Emil S. Koteles is chief technology officer at MetroPhotonics, 3701 Carling Ave., Building 14, Ottawa, Ontario, Canada K2H 8S2. He can be reached at 613-828-8717 X239 or [email protected].
_____________________
How echelle gratings and AWGs demultiplexThe echelle grating and the arrayed-waveguide demultiplexer both work on the same physical principle. The facets of the grating and the ends of the waveguides act as a series of coherent light sources. By precisely adjusting the phase shift from each source relative to all the others, an interferometric pattern is set up that results in light of different wavelengths being focused at different spatial locations on an output arc.The phase shift in the case of the echelle grating is simply determined by the relative position of each facet with respect to the others. In the case of the AWG, the phase shift is accomplished by adjusting the length of each waveguide with respect to the others.The major differences between the two devices lie in size and phase-shift accuracy. An array of waveguides is generally much larger than a grating, usually by a factor of four or five, for similar performance (see figure). This affects not only the number of chips per wafer, but also yield, as nonuniformities across the chip become more critical the larger it is. Further, as channel spacing decreases, necessitating a significant increase in the number of waveguides required to increase resolution, total component size increases dramatically. Echelle gratings require an increase in the number of facets, which is easier to accommodate without resorting to significant size increases.
Phase-error accuracy is easier to control in etched grating demultiplexers because only facet positions affect it; in AWGs, additional phase errors can be generated in the waveguide array itself.
_____________________
Battling waveguidesGlass waveguides are typically formed by growing thin layers of glass (SiO2) of various compositions on silicon substrates. This is a complex process requiring precise control over all aspects of the layers. Uniformity and reproducibility of the waveguide refractive index are important issues. Because the layers are grown at elevated temperatures and the temperature coefficients of expansion of silicon and SiO2 differ significantly, room temperature strain is an important issue. Despite the challenges, a number of companies have developed the technology for fabricating glass waveguides and AWG demultiplexers. The important advantages of glass waveguides are their stability, low propagation loss, and low coupling loss. However, since they are dielectrics, it is difficult to envision their monolithic integration with active devices such as optical amplifiers, photodetectors, or lasers.Semiconductor waveguides are grown in epitaxial growth chambers by processes such as molecular beam epitaxy, metal-organic chemical vapor deposition, or chemical beam epitaxy. These are the same techniques used to fabricate active optoelectronic devices for WDM systems, such as laser diodes, semiconductor optical amplifiers, and photodetectors. Therefore, quality semiconductor waveguides can be obtained in quantity from the same manufacturers that supply the WDM active-component industry.
Semiconductor waveguides typically have a somewhat larger propagation loss than glass waveguides, although the loss is offset by the smaller device size because the refractive index is more than double that of glass. Waveguide uniformity, reproducibility, and stability are equal to or better than glass waveguides. And they are commercially available. Furthermore, novel waveguide designs can match glass waveguide coupling losses while increasing wafer yields.
Finally, with semiconductor waveguides it is possible to monolithically integrate active elements such as digital optical switches and photodetectors with passive planar waveguide demultiplexers, thereby achieving true photonic integrated circuits (PICs). It will also be possible to integrate semiconductor optical amplifiers with passive demultiplexers, thereby producing truly lossless demultiplexers.