Emerging optical networks must carry burgeoning Internet traffic and will require many more optical components (and different types of components) than the initial point-to-point systems. The complexity of networks with 80 to 160 channels means that hundreds of discrete optical devices are required. One approach to reducing the cost per function of optical components is to integrate multiple functions in a single package. Fairly complex optical integrated circuits already have been developed in the form of arrayed waveguide gratings (AWGs) for dense wavelength-division multiplexing (DWDM). The AWGs are optical circuits fabricated from waveguides arranged on a silicon wafer and are made using well-established tools and techniques from the semiconductor industry.
To be successful, integrated devices, such as AWGs, must provide greatly reduced cost per function without sacrificing the performance necessary for demanding telecom applications. Key performance parameters are insertion loss, crosstalk, uniformity, and polarization-dependent loss (PDL). Equally important for overall system performance are polarization-mode dispersion (PMD) and group-delay dispersion. Although sometimes not mentioned, these parameters are critical for networks operating at OC-192 and above.
Arrayed waveguide gratings
Since AWGs were first proposed by Smit in 1988, extensive work has gone into applying them to DWDM.1,2,3 They represent, in and of themselves, a fairly high degree of integration. Commercial AWGs are available with 40 channels, each spaced at 50 or 100 GHz. The same function performed with thin-film filters would require 39 or more different filters, each with its associated beam expansion and re-focusing optics, or more than 120-piece parts. Laboratory demonstrations of AWGs with 128 and 256 channels spaced at 25 GHz have shown that AWGs can scale to even higher channel counts.4,5 Furthermore, the insertion loss of an AWG does not increase linearly with channel count as it does for thin-film filters and for fiber Bragg gratings (FBGs) (also see page 41). The AWGs have the best cost/performance ratio at high channel counts. When viewed from the perspective of what is placed on the chip, the level of integration is also impressive (see Fig. 1).
For the AWG to operate properly, each of the more than 100 optical elements on the chip must perform precisely. Such is the power of semiconductor manufacturing techniques that, once the design is correct and the manufacturing process stable, these integrated optical circuits can be manufactured reliably in high volume.
The shape of the AWG filter passband versus wavelength can be altered by changing details of the lens region and the transition to the waveguide array (see Fig. 2). The "regular" or "Gaussian" passband exhibits the lowest loss at the peak but is very "pointy," requiring greater stabilization of the laser wavelength. Furthermore, for applications in which the light passes through several AWGs (such as in optical add/drop multiplexing) the multiplicative effect of the filtering function reduces the passband to almost nothing. An alternative to the Gaussian passband shape is the "flat top" or "wide band" shape, which has a uniform insertion loss across it and is therefore not as sensitive to laser drift or cascaded filters as is the "Gaussian" one. However, the loss in a wide-band device is, of necessity, somewhat higher (about 2 to 3 dB) than that in a Gaussian one.
Crosstalk and insertion loss
As with any filter, AWGs do not perform their wavelength selection function perfectly. The data show that the bandpass filter functions do not become infinitely small (see Fig. 2). Instead, the curves level out and form what is called the noise floor. This causes some of the light from one channel to leak into other channels, where it causes noise and crosstalk.
Isolation and crosstalk are sometimes used interchangeably but, in general, isolation describes the difference between the signal power and the unwanted noise in the passband, while crosstalk is the total noise power in the signal passband, referenced to the input power into the device (see Fig. 3). In other words, crosstalk includes the device loss, while isolation does not.
Isolation is sometimes quoted as adjacent channel isolation and non-adjacent channel isolation, depending on in which channel the unwanted signal originates. This is sometimes convenient for descriptive purposes, since adjacent-channel crosstalk often is dominated by the shape of the peak, and nonadjacent-channel crosstalk is dominated by the noise floor. These two types of crosstalk have somewhat different origins. The important parameter from a systems design standpoint, however, is the total cumulative isolation or crosstalk from all other channels. This gives the total noise that the system designer must allow for. When specifying a device, worst-case values must be used. The total cumulative isolation is the sum of the worst-case isolation for all other channels at the worst wavelength in the passband and at the worst polarization. This calculation then gives the worst value of noise that the device might exhibit. This worst-case alignment of all parameters might be statistically unlikely, but will happen in some system somewhere and must be guarded against. For example, the worst-case total cumulative isolation for the device shown in Fig. 2(b) was 23 dB.
If isolation is to be improved, the origins of the different contributing mechanisms must be understood. Much of the groundbreaking work in this understanding has been carried out at Nippon Telegraph and Telephone.6 Noise and crosstalk come from scattering of light due to imperfections in the waveguides in the grating. These imperfections cause phase errors in the light emerging from the waveguide grating, which in turn cause the light not to be perfectly focused onto the proper output waveguide (see Fig. 4). These phase errors can be caused by effective index variations due to fluctuations in any of the indexes of refraction or dimensions of the grating waveguides.
Phase errors are caused by any fluctuation in the effective index b of the waveguides, which, in turn, may be due to errors in any of the dimensions or indexes of the device. The effective index fluctuation is given by:
sb2 = (c1sN core)2 + (c2sN bot)2 + (c3sN top)2 + (c4sW)2 + (c1sD)2
Where W and D are the waveguide width and height and the N s are the indexes. The potential noise floor then can be calculated by estimating the various parameters. The noise floor is a sensitive function of phase error, and all fluctuations in the effective indexes must be tightly controlled to achieve optimal performance. Obviously, this is as much a function of device fabrication processes as it is of design. All manufacturing processes must be highly uniform and tightly controlled to achieve high isolation. Process control is the key to high performance.
Dispersion and polarization-mode dispersion
Dispersion is a critical parameter for high-data-rate systems. Arrayed waveguide gratings do not utilize multiple reflections so all wavelengths travel close to the same optical path length. For this reason, AWGs theoretically have very low dispersion. As with the case of crosstalk, however, imperfections and nonuniformities can lead to dispersion (see Fig. 5). Dispersion between the two different polarizations-PMD-also is important and is typically less than 0.5 ps in an AWG.
Arrayed waveguide gratings can perform complex filtering functions in a highly integrated and efficient manner. They can replace hundreds of conventional discrete parts. Achieving high performance, however, requires advanced designs and precise manufacturing process control. It is possible to reliably manufacture these complex optical integrated circuits (OIC) in high volumes by capitalizing on the extensive infrastructure that has been created by the silicon integrated circuits industry. This infrastructure includes the equipment, processes, and methodologies that have been built up over 40 years with the investment of billions of dollars.
As important as they are, AWGs represent just one building-block element that can be integrated with other building blocks to perform subsystem and module-level functions in a single package. Other building-block functions include switches, couplers, taps, variable optical attenuators, and splitters. Furthermore, active components such as lasers and detectors also can be integrated in a hybrid fashion. Groups around the world are working to utilize the many possible OIC functions to make integrated modules.
Recent demonstrations include an integrated dynamic gain equalization chip, an integrated multichannel waveguide laser, and a frequency spectrum synthesizer.7,8,9 Such second-generation products will be released by several vendors in late 2000 or 2001, and will replace rack-mounted boxes assembled out of many traditional optical components. Future products will integrate multiple optical functions, including lasers, receivers and electronics, inside a single package.
1. M.K. Smit, Electron. Lett., 24, 385 (1988).
2. H. Takahashi, Electron. Lett., 26, 87 (1990).
3. C. Dragonne, IEEE Photon.Lett., 3, 896 (1991).
4. K. Okamoto et al., Electron. Lett., 32, 1474 (1996).
5. Y. Hida et al., Electron. Lett., 36(9), 820 (2000).
6. T. Goh et al., J. Lightwave Tech., 15, 2107 (1997).
7. B.J. Offrein et al., ECOC`99 Post Deadline Papers, 6.
8. M.W. Sckerl et al., ECOC`99 Post Deadline Papers, 16.
9. K. Okamoto et al., Electron. Lett. 35, 733 (1999).
JANE LAM is director of design engineering and LIANG ZHAO is optical design engineer at Lightwave Microsystems, 2911 San Jose, CA 95134; e-mail: firstname.lastname@example.org and email@example.com.
FIGURE 1. Basic elements and operation of an arrayed waveguide grating shows how the input waveguide (1) enters a lens region (2) that divides the power among more than 100 different waveguides in the grating array (3). Each grating waveguide has a precise length difference relative to its neighbors (DL) so that the light in each waveguide emerges with a different phase delay at the output of the waveguide array (4). The delay is given by: DF= 2p *b * DL/ l. The second lens region (5) refocuses the light from all the array waveguides onto the output waveguide array (6). Due to the precise differential phase tilt for different wavelengths, each wavelength is focused into a different output waveguide in the output array.
FIGURE 2. The shape of arrayed waveguide grating passbands can be adjusted-the two most popular passband shapes are Gaussian and wide band. Figure 2a shows data from a Gaussian AWG and Figure 2b from a wide-band AWG. The 40-channel devices had a channel spacing of 100 GHz.
FIGURE 3. Isolation in an AWG is referenced to the top of the in-band transmission function.
FIGURE 4. Calculation of the filter function of an AWG channel in the presence of different sorts of phase errors shows how, with a perfect index distribution, the transmission of the channel falls off rapidly without stopping, reaching a value of greater than 60 dB within a few channel widths (upper left). With random phase errors imposed on the waveguides the transmission stops falling and forms a "floor" at about 30 dB (upper right). This type of phase error is dominant for non-adjacent channel crosstalk. In the bottom two charts a periodic phase error is introduced so the transmission function continues to fall with no noise floor, but the width of the transmission peak is broadened. This type of phase error can be the major cause of adjacent channel crosstalk.
FIGURE 5: Dispersion over the nominal channel passband of 0.2 nm in an AWG channel with a 35 dB noise floor is less than 1 ps.