Packaging is key to optical-MEMS production

Acfd12

Jeff Bonja and Bob Rubino

Vendors in the MEMS-based switch market face obstacles in assembly and testing processes that result in economic and planning challenges for high-volume manufacture.

The microelectromechanical systems (MEMS) optical-switch market reintroduces packaging problems that have long plagued optical-component manufacturing. While packaging a MEMS-based switch involves particular design and manufacturing difficulties, ensuring the reliable, high-performance, single-mode optical connectivity required in three-dimensional (3-D) MEMS-based switching is especially challenging. And if this were not daunting enough, the exponential growth rate in channel count has quickly made the standard single-channel optical-device packaging, testing methods, and procedures developed for the WDM industry inadequate for MEMS devices with channel counts that will soon number in the thousands.

In addition to the relative technical complexities of the assembly and testing processes, there is an added dimension of difficulty in planning. Initial projections for the optical-MEMS market can only support numbers of units in the hundreds to thousands for the next few years, but this translates into hundreds of thousands to millions of aggregate channels. Typical economies of scale for high-volume manufacturing will be difficult to realize.

Business decisions regarding entrance to this marketplace must take into account three success factors: the technical requirements of the components, the capital amounts needed to build the technical competencies to address these issues, and the capital and infrastructure required to support long-term manufacturing of these devices. While any one vendor may have a lead position in one of these areas, only those vendors who have sufficient positions in all three areas will be able to deliver optical interconnects with the performance, volume, and cost demanded in the marketplace.

SUCCESS FACTORS
The first success factor requires sufficient technical competence to solve issues in developing high-performance optical-MEMS interconnects. In spite of growing demand for higher-channel-count connectors, there are obvious technical reasons why the multimode market is limited to 1 x 12 and 2 x 12 connectors. In a similar sense, while single-mode connectors are currently available in 1 x 12 formats, the market leans towards the lower insertion loss offered by a limited number of high-performance single-channel designs. In both of these formats, connector performance over environmental extremes is a direct function of positional accuracy and design stability. Difficulty achieving satisfactory performance with current designs explains why we do not see a proliferation of higher-density connector designs in the market today.

Similarly, the optical-MEMS switch interconnect shares some of these same issues, though magnified by channel counts numbering in the thousands. As with connectors, positional accuracy of the MEMS fan-out array is the main problem. To manufacture parts that even marginally meet the stringent requirements of this application, special attention must be paid to the stack-up of positional tolerances between the light-guiding core of the single-mode fiber and the corresponding beam-shaping optics of the 3-D MEMS switch (see Fig. 1). Although registration tolerances of the beam-shaping lens array, with respect to the MEMS switch array, are relaxed by virtue of the lens magnification, the positional tolerances between the lens array and the fiber fan-out become correspondingly more demanding.

For a typical 3-D MEMS optical fan-out, the maximum dimensional tolerances for core offset, axial displacement, and beam-pointing error are significantly more stringent than what is expected for traditional interconnect technologies. There are many technologies and methods of fiber alignment that approach, and in some cases exceed a subset of these specifications. What is of more interest (and problematic to the industry as a whole) is the technique of testing required to verify positional accuracy, not only for developmental parts but also for the thousands or hundreds of thousands of measurements required to support full-scale manufacturing.

The traditional fiber-geometry measurement techniques—fiberoptic test procedures (FOTP)—prove insufficient in measuring the dimensions and tolerances of cores of individual single-mode fibers, let alone arrays of hundreds or thousands. What is required is the investment and development of a "golden" test-bench setup that can accurately measure positional accuracies without the burden of having to complete these measurements on many manufactured parts. This setup would allow a significant amount of the design space to be bound by technical rather than economic criteria.

For this proposed testing scenario to be effective, there has to be a production-level test setup that can provide more of a go/no-go test without providing any specific quantitative analysis for diagnostic purposes. The development of the production-level testing setup will be discussed later, in terms of an all-inclusive test that will screen for multiple variables at once.

TESTING
Testing a MEMS switch under development is a straightforward but costly task. To be able to accurately identify the absolute position of every core in an array of fibers, the testing system must produce optical-core position measurements with submicron accuracy over tens of millimeters across the fan-out's endface.

Commercial equipment is available that comes close to being able to position devices under test (DUT) to this accuracy, but they are expensive and demand a significant amount of setup and maintenance to maintain accuracy. Commercially available noncontact micrometers provide a convenient starting platform for the positional measurement using broadband illumination. However, because these systems are designed to operate in the visible region of the spectrum, one must pay special attention to inaccuracies that may arise in the location of the optical core because modal noise may arise from illumination of the fiber core at wavelengths below cut-off.

Unlike standard fiber connectors, interfaces to optical-MEMS systems depend on the absolute and relative pointing accuracies of each fiber. This measurement dependency is quite new in the interconnect industry and, to date, there are no FOTPs available that provide even initial guidance for measurement. Again, as with the positional measurements, it is apparent that measurement strategies need to be segregated into a developmental "ideal" measurement system with a production-level test.

The developmental system can be based on high-quality, commercially available, and expensive beam-profiling equipment. At a minimum, this system provides the ability to accurately profile each of the output beams of the fiber cores in the array. To complete the pointing measurement, one must first normalize the position of the DUT to the beam-profiling equipment and then accurately translate the beam profiler along an axis normal to the surface of the DUT (see Fig. 2). While seeming trivial, this task must be completed with subradian angular accuracy.

Finally, a less technically challenging but absolutely required task is to automate the x-y positioning of the DUT. Each of the fiber cores must be positioned in the beam profiler's field of view without negatively affecting the previous alignment. The appropriate equipment also is necessary to illuminate each of the cores under test, while having a sufficient level of illumination at a wavelength above fiber cut-off.

In both of the prior measurements, it may not be economically viable to complete 100% testing of all optical channels because counts exceed one thousand. To address this issue, it is necessary to develop a simpler go/no-go testing setup that ensures compliance with stated specifications. It is easy to envision a system that mimics the operational environment of the optical-MEMS system being tested (appropriate lens systems, apertures, detectors, and so on) and provides usable feedback as to whether a particular DUT meets the necessary performance tests to satisfy a customer's specifications.

In addition to pointing and positional accuracies, optical-MEMS-based interconnects suffer from an additional degree of freedom in the z-axis along the axis of the optical core. Since these are either directly coupled or air-coupled to lens systems, the absolute position of the end of the optical fiber, with respect to the lens array, will have significant impact on collimation and therefore insertion loss of the system.

Commercially available white-light interferometers have the resolution and accuracy necessary to characterize both developmental and production parts. The true challenge in this arena is choosing the appropriate design to maintain this positional accuracy (or array flatness) during all manufacturing and environmental conditions.

Finally, as with many passive single-mode components, stringent insertion-loss and return-loss testing must be completed for the assembly. Suitable commercially available test sets are appropriate for both the developmental and production testing of these systems. The difficulty arises in dealing with the high overall channel count for each assembly, as well as for capturing the received optical power from the array ends of the assemblies.

While all of these testing systems must be developed in order to fully characterize an optical-MEMS interconnect, there is significant engineering required to design a device that is essentially a thousand-fiber (or more) connector. Initial impressions may be that this engineering effort is a straightforward scaling of the existing technologies, but many of the issues do not translate linearly and additional issues develop when working with devices of this magnitude.

On top of all the design concerns associated with developing a feasible technical solution, any testing solution ultimately must be manufacturable. Many promising lab experiments have created great interest but are nearly useless when they cannot be reliably manufactured in volume. Designing for manufacturability has to be included in the initial development of the components.

CAPITAL REQUIREMENTS FOR TESTING
Outside of the technical considerations mentioned above, the investment side is another factor that must be well-understood for success in the optical-MEMS interconnect marketplace. The equipment and intellectual capital required to appropriately develop and test these components is significant, and this does not include the investment required in order to scale up a manufacturing facility. Typical connector manufacturing houses can start production with minimal investments of a few handheld test setups and polishing tools. They can rely on the significant amount of design and testing completed by the connector parts manufacturer to ensure that their components meet a minimum performance requirement.

Unfortunately, the optical-MEMS interconnect market is still immature. It will be a long time before the major components in the devices become as available as an MT style or LC connector is today.

At a minimum, vendors interested in developing optical-MEMS interconnects must be willing to either invest in or have almost full-time access to precision dimensional metrology equipment. It must be paired with vision systems possessing submicron resolution and absolute accuracy; precision beam-profiling equipment tailored for the wavelengths of interest; large-area surface profiling equipment capable of resolving 10-µm spatial features and better than 100-nm height resolution; and insertion- and return-loss measurement equipment (see Fig. 3).

In addition to this specific test equipment, all of these components must qualify under the methods and procedures called out by Telcordia test standards. This implies frequent and potentially long-term (2000 hours) access to environmental test equipment such as environmental chambers and vibration and shock equipment during the development of these components, as well as during the final qualification phases. Including all of the optical sources—detectors, optical switches, and the specialized automation systems and software necessary to quantify channels—the investment in this equipment alone can exceed the amount of most startup seed money.

MANUFACTURING CAPITAL REQUIREMENTS
Moving beyond what has until now been described as a craft industry, maturation of the optical industry as a whole will require significant investment in automation (the third success factor) to greatly increase product volume at lower component cost to satisfy the market. In this regard, there is a need to develop a testing system that will allow for go/no-go production testing. This equipment, not unlike assembly automation, will require a significant investment in capital and intellectual property to develop.

There is also the capital required to ramp up manufacturing of the components themselves. Unfortunately, the economies of scale that have come with deploying WDM components do not necessarily apply to their manufacture. Automation may not be the best solution, depending on how the technologies specific to this type of assembly develop.

Automation has its own pitfalls in that many machines are limited in their scope of application, therefore reducing their returns on investment. Manual assembly or semiautomated assembly may prove to be a viable route for these components in the near-term, or at least until some standard designs take hold. Those companies with skills in the management of and with global access to this type of manufacturing environment may hold the best advantages in the optical-MEMS interconnect marketplace. While potentially increasing the requirements for inspection, this type of manufacturing system does allow for a very nimble manufacturing process with the ability to react quickly to the changing technical requirements of the end user with little reconfiguration lead-time.

Demand for optical-MEMS interconnects is very clear and there are a few vertically integrated or independent vendors operating in the market. Customers who require the services of these vendors must realize that in addition to being able to deliver a few developmental components, the vendors must also be able to address all of the technical and financial issues associated with long-term commercial development and production.

Jeff Bonja is product manager and Bob Rubino is director of product development at Schott Communication Technologies, 15 Sandersdale Road, Southbridge, MA 01550-2855; tel.: 508-765-1112. They can be reached at jeffrey.bonja@us.schott.com and bob.rubino@us.schott.com.

More in Home