Almost from the beginning, stabilization of laser wavelength and narrowing of the spectral bandwidth was an important technology needed for some laser applications. At the time—three decades ago—atomic transitions were used as the reference. The primary motivation for laser stabilization was to perform high-resolution spectroscopy on small molecules. In the past decade several important applications have evolved to a point where laser stabilization and narrowing of the bandwidth are necessary conditions for the success of these applications.
The first application is laser lithography, in which the continually narrowing linewidths on the semiconductor require ever more precise lithography. The consequence of Moore's law has been an ongoing shift to shorter wavelengths, increased stability of the laser wavelength, and a narrowing of the spectral bandwidth. For lithography this control is required because the focal length of the refractive lenses that are used are dependent on the wavelength. Changes in wavelength will shift the focal plane of the projected image of the mask on the wafer, increasing the width of the lines. Stability of at least 1 part per million is required. Laser isotope separation, the second application, also requires maintaining a narrow fixed wavelength to selectively excite molecules containing the required isotope.
A third application requiring wavelength stability is optical communications. As the number of channels increases, the spacing between channels has dropped from 200 to 100 GHz, and soon will be only 50 GHz. The increase in channel density requires the laser to be narrow and stable. At 1.553 µm the frequency is approximately 193 THz. For a 100-GHz separation a ±2.5-GHz stability is required. This means a stability of ±13 parts per million is needed. Not quite as severe as the lithography laser, but the stability must last for 20 years.
The structure of the laser diode ensures that the output will be narrow. However, it does not ensure that it will be stable over days, let alone the decades it must operate in the field. Careful thermal design and control of the temperature make it possible for one manufacturer to specify the lifetime drift of its digital laser products to only 12.5 GHz. Such a product could not be used in a system with 100-GHz channel spacing.
In addition, as the number of channels increases, the cost of maintaining inventory by the manufacturers of the lasers and line cards increases. This is also a problem for the service facility that must stock spares for maintenance. Therefore, a number of companies have developed tunable laser diodes. Depending on the mechanism for tuning, this can reduce the required number of parts on average by a factor of ten.
The tunable laser creates a new set of problems. First, a fixed wavelength laser can have a wavelength locker manufactured to stabilize its particular wavelength. For tunable lasers, the locker must also be tunable or the problem of managing 100 different parts is not resolved. The increase in wavelength density as well as the pressures to reduce part count requires that the wavelength locker no longer be an external module but be incorporated into the laser package.
The best stability could be obtained by locking the laser to an atomic transition because these would be pretty insensitive to most, if not all of the environmental conditions that the stabilizer would encounter. This approach, of course, is completely impractical. However, locking to molecular transitions such as carbon monoxide, methane, acetylene, or hydrogen cyanide spectra can be used for high accuracy wavelength references.
Several approaches to diode-laser stabilization can be found in the patent literature. All of them rely on passing the laser light through a wavelength-selective element. One example is a thin-film dielectric filter, which requires many deposition layers to get the bandwidth narrow enough. The main problem with this approach is that it is adjustable over 20 nm by angle tuning and so has the same problem with needing to stock a filter for each channel. Optical fiber Bragg gratings are easy to manufacture and are used extensively as multiplexers and demultiplexers but have not been used as lockers for telecom wavelengths. They have the same problem as the filters in that a Bragg grating would need to be available for each channel.
The component that is best suited for wavelength stabilization is the Fabry Perot etalon. It was invented in 1899 and has been used extensively. The etalon consists of two parallel reflectors with reflectivity R separated at a distance, l, by a spacer. The spacer can be a ring so that air with index n = 1 is in the space between the reflectors or it can be a solid material with a higher index of refraction such as quartz (n = 1.44) or silicon (n = 3.48).
The peak in the transmission occurs when 2n/cos(θ) = mλ where θ is the angle of the refracted light inside the etalon. Since m is an integer, many different wavelengths can be transmitted by the etalon. As the reflectivity of the mirrors is increased, the etalon becomes more selective. The frequency width corresponding to the full width at half height of the transmission through the etalon is given byWhere F is the finesse of the etalon and is given by πR1/2/(1 - R) and c is the speed of light in vacuum.
This leaves the optical designer a number of options. The finesse needs to be fixed during the manufacturing process. The index of refraction of the material between the mirrors has two effects of the design. The higher the index, the more compact the etalon can be, which is important in trying to make the etalon as small as possible so that it can be easily packaged inside the laser.
The temperature of the etalon provides one method for changing the effective separation between the mirrors and, hence, the transmitted frequencies. For example, etalons made from silicon are more than three times as thin as those filled with air. However, silicon has two properties that work against it. The electro-optic properties of silicon make it difficult to maintain polarization and the temperature dependence of the index of refraction is too high.
In a practical etalon this condition would respond to a shift in transmitted frequency by about 10 GHz/°C. This puts too large a demand on the thermal stability of the package that houses the locker. Sapphire is an excellent material except that its price is too high. I believe that most people have chosen fused silica. It does not change the polarization and has an effective temperature dependence of only approximately 1.5 GHz/°C.
The etalon lends itself naturally to supporting any of the ITU spacings. It just needs to be manufactured so that the ITU wavelengths divided by an integer equal nl. Small errors in the manufacturing process can be corrected by simply changing the temperature of the etalon. Alternatively, the etalon can be tilted so that the internal angle θ is somewhat greater than zero. In fact it is important to make θ larger than zero to prevent back reflections into the cavity.Another approach is to use a Fabry Perot structure at only a slight angle with respect to the beam. A photodiode monitors a reflection from the front surface of the first reflector to compensate for any changes in laser output power. A second photodiode monitors the light passing through the etalon. The temperature of the etalon is adjusted so that the transmitted wavelengths correspond to the ITU grid.
Any decrease in the ratio of transmitted-to-reflected light means that the laser wavelength has shifted. However, since the etalon is centered on the grid, it is impossible to determine in which direction the source has drifted without using a dithering of the laser. Alternatively, the etalon is adjusted to transmit the ITU grid at about the 50% point. In this way it is possible to generate an error signal that is useful (see Fig. 3).
Finally, to use an etalon with any of these methods with a tunable laser it is important to be able to tune the laser close to the wavelength desired, and then allow the locker to center the output. The etalon is excellent at detecting relative deviations but cannot measure absolute wavelength without some other measurement. Thus, a tunable laser is characterized during manufacturing to determine the correspondence between operating parameters and the grid location. Of course this characterization must be stable. It is interesting that while DWDM is a cutting-edge technology, it requires a key component that is more than a century old.
Ken Kaufmann is with Hamamatsu, 360 Foothill Rd., Bridgewater, NJ 08807. He can be reached at [email protected].