Automating Photonic Production

The optical networking community underwent explosive growth from 1996 through most of 2000. However, for most of this period of rapid growth, assembly of photonic modules (couplers, transceivers, etc.) tended to be, at best, semi-automated. In some instances this occurred because the total product volumes at any given company were not large enough to justify full automation. In other cases, particularly when fiber pigtails were involved, machines capable of fully automating some functions had not yet become available. But the automation companies have been pursuing the problem energetically with improved automation equipment.

This article will look at the automation involved in producing photonic assemblies—where optics and electronics come together. Although semiconductor manufacturing, as well as optical fiber fabrication and cabling, are highly automated procedures, they represent a different "brand" of automation, performed by a different set of manufacturing companies, with a much longer pedigree. Therefore, this article will avoid these areas, as well as mass fiber polishers and mass fusion splicers, which are more related to the cabling process.

1. Block diagram of a fully automated workstation for assembly of photonic equipment. (Source: Adept Technology)

It is instructive to look at the technical problems inherent in automating production of photonic subsystems and systems by comparing those problems with similar problems encountered in automated semiconductor manufacturing on one hand and electronic manufacturing on the other. The technical requirements inherent in automating photonic production lie between the two other fields.

In the world of manufacturing integrated circuits, precision motion is the order of the day. Movements must be accurate and repeatable to a 10th of a micron or better. Historically, the range of motion has not been too large, although with 300 mm wafers the demands on range of motion of such systems have been increasing. The rate of motion need not be particularly high but, in many cases, the motion must be performed in vacuum. The combination of these requirements has driven automation in IC fabrication toward high-cost, high-precision equipment.

For general electronics manufacturing (often impolitely called "board stuffing"), the demands and constraints of automated production point in somewhat different directions. Precision is usually a good deal less important than in IC fabrication. Motion distances are relatively large, with working envelopes of perhaps half a meter by half a meter horizontally and at least a couple hundred centimeters vertically. Speed of motion is more important in board stuffing than in IC fabrication, to maintain satisfactory throughput.

	2. Melles Griot’s Model 17 AWG 001 workstation, designed specifically for fiber-to-arrayed waveguide-to-ribbon fiber applications. (Source: Melles Griot)

The "wild card" in automated photonic manufacturing is that the automation system must not make any motions that might overly bend or kink (or worst of all, actually break) fibers typically 500 cm or more in length and attached to devices. That requirement alone strongly limits what might otherwise be combinations of simple linear motions. Indeed, the most complex automated workstations offer as many as 12 degrees of freedom of motion—three linear and three rotational degrees of freedom for each of two components being joined.

If the technical aspects of the problem can be stated succinctly, the potential answers are a bit more complex because economics come into play in a big way. High-precision motion equipment, the heart of any automated production system for photonic assembly, is inherently slow, in terms of millimeters per second of motion. Therefore, it is difficult and expensive to maintain high-precision motion across long travels. Also, what might be suitable in terms of visual control for a basically manual process (a good light and a suitable microscope) must give way to some sort of machine vision system—cameras, frame grabbers and other dedicated machine vision hardware—with a full load of software controlling both the motion and machine vision systems.

Most of the dedicated photonic automation equipment available today as complete systems consist of three separate subsystems: an automated motion system, typically with four to six degrees of freedom (X, Y, Z, and one to three angles of rotation), a controller, and motion software; a machine vision system with cameras, suitable optical components, and image capture/analysis hardware and software; and some form of connection-forming hardware (typically a die bonder or a laser welder, possibly some form of epoxy-applying bonder).

The problem of achieving high-resolution motion over large areas without slowing down the process is usually addressed by starting with a long-travel, low-resolution X-Y positioning system and then mounting a high-resolution, short-travel X-Y positioning system (often a piezoelectric X-Y positioner) atop the low-resolution system. The latter then delivers the precision portion of the system to the immediate vicinity of where its high-precision motion is needed. Tying two or three sets of software together seamlessly is a key factor in making such systems work effectively.

Seven or eight years ago, companies active in building photonic assemblies typically bought motion control equipment and machine vision equipment from various suppliers and integrated the equipment themselves. The confident companies said something like, "Why don't we buy a couple of motion stages from Company A, a couple more from Company B, a motion controller from Company C, a camera-and-microscope combination from Company D, write our own proprietary software to control everything, and go from there? How difficult can it be?" The answer to the latter question often turned out to be, "Relatively simple to get the integrated system to move around, but surprisingly difficult to get all the parts to operate smoothly together." Proprietary "combination" systems often required the infusion of a lot of engineering time and talent.

At about the same time, motion control companies found themselves spending significant engineering support time answering broadly similar questions from what was then a relatively small group of companies involved in telecommunications. A new market niche was born—integrated photonic assembly workstations—initially as semi-automated workstations and more recently as fully automated workstations. The principal players here are motion control companies that have taken it upon themselves to integrate their own positioning offerings with machine vision systems and die bonders or laser welders.

At what point should manufacturers think about automation? That varies from industry to industry and from process to process, of course, but a general pattern is apparent. Manual assembly makes sense for low production volumes; fully automatic assembly becomes imperative at high volumes; and some form (or forms) of semi-automatic assembly makes sense at intermediate volumes.

In the area of photonic assembly for optical networking, the lower boundary is fairly clear; annualized run rates of less than 1000 units per year, or 20 units per week, typically do not merit anything beyond hand assembly. The upper boundary, which separates fully automatic from semi-automatic assembly, is a bit fuzzier. Some companies think in terms of annualized run rates of 25,000 units, while others espouse numbers of 50,000 units per year or higher.

Let's take a closer look at the major electronic manufacturing service (EMS) companies as a group to see why they will become more important in the future. All manufacture electronic equipment for a wide variety of applications, including telecommunications, computers, wireless equipment, medical devices, industrial electronics, and other fields. Most had large exposures to telecommunications, which has made 2001 a painful year for them financially.

Historically, companies that make equipment for the telecom market have farmed out the electronics and sheet metal packaging portion of production of their products to EMS companies. But until recently, these same companies have retained the assembly of the photonic portion of the final product to themselves—both because those companies believe they have more expertise in handling optical components and because they could handle their own relatively modest production runs with the semi-automated equipment they already had on hand.

Thus, the big EMS players have, to date, done relatively little in photonic assembly, although all the big players have some exposure to the field. Even as the EMS companies are going through a period of turmoil today, some of them are expanding their reach into photonic assembly. Sanmina, for example, has bought manufacturing plants from Nortel, Lucent and Alcatel in recent months. Similarly, Solectron bought a dense wavelength division multiplexing facility from Cisco Systems in May.

Cahners In-Stat expects that to change significantly over the next few years. Historically, in a down economy, the EMS companies have done well in capturing new clients; the EMS value proposition becomes relatively more attractive in a down economy. This factor is enhanced by the fact that an EMS company serving multiple customers—almost by definition—has an annualized run rate high enough to justify full automation of its processes rapidly. In addition, numerous recent start-up companies plan to not manufacture anything other than prototypes, expecting to use EMS companies to do all their manufacturing.

The major players in the photonic automation equipment realm come from various segments of the motion control business. Some, like Newport Corp., come from a background in R&D-based motion control. Others, like Adept Technology, come from the high-precision semiconductor industry. And still others, like ATS (Automation Tooling Systems), come from the more generalized robotics/factory automation field.

Bear in mind that numerous other general factory automation manufacturers have the ability to move into this market space without an extreme amount of effort. They already have motion expertise, often work with lasers and machine vision systems, and have marketing channels already in place. About all they need for market entry is an alliance with a company with high-precision motion experience, probably from the R&D community.