By Scott Jordan
Aligning optical fibers and V–groove arrays to waveguide channels is difficult and can result in low production yields and high losses. Alignment automation based on microrobots, cameras, and software can greatly increase productivity.
The application of microrobotics to optical–component production represents the convergence of semiconductors and photonics in several ways. Most obviously, the silicon arrayed waveguide (AWG) is a lithographically produced pattern imprinted en masse on a single wafer, then diced by the dozens. But there is another, less evident parallel: the advent of production–robot configurations can address unique characteristics of optical applications.
The trend is reminiscent of the introduction of the first wafer–handling robots in the semiconductor industry 30 years ago. Early handlers were stacks of linear slides, which shuttled wafers from station to station in process tools. A radical departure occurred in the 1980s, with the first multilink wafer–handler robots. These used coordinated counter–rotating axes to fold and unfold arms, giving the robots long reach with a more compact footprint than was possible with stacked linear configurations. Cleanliness and speed were significant benefits because the exposed, particle–producing bearings and substantial moving masses of the stacked approach were eliminated. Still, the new configuration was outlandish, and the necessary controls were more complex. Tool engineers took a few years to accept it, but with the industry well along in its long–term initiative to improve yield by eliminating manual handling of wafers, these robots became the norm.
A similar evolution is playing out in photonics process automation. The first production–worthy alignment subsystems—such as the earliest digital gradient–search units introduced more than a decade ago—were stacks of off–the–shelf linear stages. In 1997, however, the first hexapod–based photonics microrobot was introduced. Driven by the photonics industry's yield–improvement initiative to eliminate manual alignment processes, this configuration addressed emerging needs, including increasing device complexity, angular as well as transverse alignment, instant placement of the physical rotation center–point at specific optical points (something not possible with mechanical rotary bearings), and improving process throughput by eliminating more than 90% of the moving mass of conventional stacked approaches.
After deployment as an enabling technology in several otherwise–intractable micromechanical systems (MEMS), DWDM, and waveguide applications, the hexapod principle became mainstream. The hexapod–based photonics microrobot has been proven in this and other waveguide applications because of its high–precision, six–axis functionality, as well as its programmable rotational pivot point, integrated high–speed metrology, compact footprint, integrated high–speed automated–alignment routines, supporting software (including LabVIEW libraries), and COM–compliant 32–bit dynamic–link libraries.
WAVEGUIDE ALIGNMENT ISSUES
Waveguide–class devices are becoming more common as the metro market improves. Their complex functionality, parallel multichannel operation, and cost–effective planar fabrication offer many opportunities for systems designers and for eventual integration into hybrid optoelectronic microcircuits. Waveguide devices include semiconductor optical amplifiers (SOAs), AWGs, and planar lightwave circuits (PLCs).
Economical packaging of waveguides depends on the automated alignment of fibers and V–groove arrays to the waveguide channels. Waveguides can be paradoxically simple yet complex to align. Alignment of a fiber to a waveguide channel can closely resemble the straightforward fiber–to–fiber butt–coupling process. However, waveguide channels can be as small as 0.2 μm, necessitating very high–resolution/high–stability mechanics. Waveguide channels are generally rectangular and yield a somewhat elliptical, astigmatic wavefront; this can be problematic for some alignment mechanisms and methodologies.
Many waveguide devices act as wavelength splitters, with mixed wavelengths at the input resulting in discrete wavelength outputs arrayed by channel. Combined with fairly large insertion losses, this can result in comparatively meager light output, which poses further challenges for some alignment methodologies. Further, angled cleaves must frequently be accommodated. And achieving "first light" at the input is often a time–consuming, blind process. Bulky, awkward fixtures and process inconsistency complicate this fundamental initial task.
The attributes of a microrobot can be examined in an application we developed for aligning waveguides and fiber. Using two hexapod–based, microrobots with 12 degrees of freedom, we aligned input and output fiber arrays to a multichannel, silicon–AWG demultiplexer (see Fig. 1). This is one of the most challenging devices to align because it involves multiple input channels carrying photonic bit streams in mixed wavelengths and the output channels are ordered by wavelength (see Fig. 2).1
Two automated microrobots were mounted on an isolation table. Initially, an infrared (IR) camera imaged the output of the silicon AWG to facilitate first–light acquisition. A downward–looking camera was mounted on two stages for viewing the gaps. These stages were controlled by one of the microrobot controllers. The output fibers from the 0th and Nth channel were connected to a power–meter card. The output of the photoamplifier was connected in parallel to both of the analog–to–digital inputs.
A LabVIEW program was constructed to sequence the two microrobots for six–degree–of–freedom alignment of both sides of the AWG. A serpentine seek routine for the two microrobots was programmed to automatically capture first light at the input, eliminating the need for the back–end IR camera. Fine–alignment commands were then issued to automatically optimize the coupling of the input and output fiber arrays.
The transverse alignment of the single–mode–fiber silicon V–groove arrays to the waveguides was shown to be highly reliable. Because of the thermally induced spectral instability of the broadband source (an erbium–doped fiber laser with only the aggregate integrated power—not the spectrum—stabilized by closed loop) we statistically removed any monotonic trend observed from variation of the source. Both the raw and processed data were plotted (see Fig. 3).
FIGURE 3. Statistical analysis of ten successive dealignment/realignment operations demonstrates excellent repeatability. The terminal voltage of each run's optical coupling measured by the two F–206's power–meter cards can be presented in several ways: the raw data (red lines, top left); a bar graph showing the detrended value of each run (blue bars, top left); a histogram (top middle); mean, standard deviation, and variance (top right); and an XY plot of the terminal position (lower left).
We selected the fast–scan and align (FSA) algorithm available in the software for the demanding alignment of the quasi–Gaussian couplings to these waveguides. Typical transverse alignment time was less than 5 s from a 200–μm initial offset.
Transverse alignment repeatability was approximately 0.1 dB—virtually identical to or better than a steady state test with no alignments (see Fig. 4). Terminal positions clustered typically within ±0.1 μm. Alignment results were highly dependent on device coupling characteristics. The performance of the microrobot in the demanding transverse alignment for these devices allows the fully integrated system to achieve 0.2–dB consistency for the full 12–degree–of–freedom production process.
We chose a video approach to achieve the desired 10–μm Z separation between the fiber–array V–groove assembly and the waveguide input and output faces. Placing the pivot point on the optical axis of the 0th array channel enables the microrobot to position its optical axes anywhere in space via a single software command. The θZ alignment (that is, adjustment of the input/output roll angles of the array to bring all fiber channels into alignment with the waveguide channels) is achieved by the microrobot's fast angle–alignment command. The θY and θX angles are readily optimized via the fast angle scan to maximum (FAM) and automated gradient search (AAP) commands.
Scott Jordan is director of Nano Automation technologies at Polytec PI, 6537 Fall River Dr., San Jose, CA 95120. He can be reached at firstname.lastname@example.org.
1. E. S. Koteles, WDM Solutions, 95, (March 2001.)