Designing for automation

Photonics products require assembly tolerances and process control beyond the capabilities of manual labor. When technicians attempt to assemble these products, the difficulty of consistently getting the best alignment, maintaining peak power during epoxy cure, and delivering exact adhesive volumes results in uneven quality and low yields. Companies that make low-volume telecommunication devices risk losing this market to experienced high-volume manufacturers if they do not bring down costs and increase yields quickly.

For most photonics products today, annual scrap costs due to poor yield exceed annual labor costs by an order of magnitude. Let's examine the case of a photonics component that sells for $1000, has a cost of $500, and currently requires two hours of assembly and test labor. Typical yields for this class of product are on the order of 60% when assembled by hand. In other words, for each good product, 40% of $500--$200--worth of scrap is generated. Two hours of labor might cost $30 in North America or $4 in China, but either amount is dwarfed by scrap cost.

Figure 1. Manually assembling a 10 Gb/s telecom transceiver (left) with laser diode, detector, and packaging takes three hours and produces a $2000 device with a 60% yield, while automated assembly in Japan of a $1, 10 Mb CD transceiver takes 2.5 s and has a 99.7% yield.

Automated assembly processes, on the other hand, typically generate production yields on the order of 90% or more. For example, typical first-pass yields for a surface-mounted-device circuit board containing 300 to 500 components run 94% to 96%. A 30% savings in scrap for these assemblies amounts to $150 per unit. At a very modest annual volume of 10,000 units, scrap savings provides a one-year payback for $1.5 million of automation in both North America and China (see figure 1). As photonics manufacturers come to understand yield and cost issues more fully, they are quickly concluding that automation will be critical to their success, regardless of manufacturing location.

A factory is a collection of processes, each of which costs money to develop and maintain. The incremental cost to add a new process is higher than the cost to increase the throughput of an existing process. The first strategic question to consider in automating fabrication of a product line is how to design a device that requires a minimal number of unique processes. Once a set of manufacturing processes is implemented, new products should be designed to run on these existing processes.

It is also important to develop and troubleshoot a high-yield process before actually trying to scale up production. In many cases this means that critical processes must be refined with semiautomatic (manual-loading) bench-top tools until yields reach percentages in the high 90s. Only then should additional money be spent on scaling up throughput per station via automated part feeding and material handling. When evaluating semiautomated systems, manufacturers should look for equipment designed with full automation in mind, so that upgrading re-uses as much equipment as possible and minimizes incremental expenses per existing station.

Automation requires process engineers who understand how to operate and maintain equipment. It also requires design engineers to understand the capabilities of the equipment and to design products for these capabilities.

The perfect part for automation would be a domino. It is insensitive to scratches, shock, or electrostatic discharge. It has a very simple geometry with no delicate protruding parts. It does not roll around or get tangled, and it has nice, high-contrast fiducial marks printed on it to act as a standard reference for machine vision. Not surprisingly, it is fairly easy to design a machine to stack dominoes.

An optical part with a delicate pigtail attached to it is pretty much the opposite of a domino. Optical parts are extremely small, difficult to handle, often fragile, and generally not well-suited to automation. The typical diameter of optical fiber is 125 µm, and ferrule diameters are less than 2 mm. In addition, diode laser components have less than 1 x 1 mm surface area for automated handling and often lack registration marks to determine proper orientation.

To allow high-volume production and yields greater than 95%, optical devices must be designed for cost-effective automation by limiting part variants and assembly methods used for each component. Parts should be designed to ease feeding, locating, and gripping. Material handling is simplest when there are only a few standard shapes to handle and these shapes can be accommodated by existing equipment. Many optical packages do not fit existing electronic package form factors. There is no apparent reason for this other than lack of awareness on the part of optical designers. Fortunately, work is currently underway to develop packaging standards. Standard part shapes, assembly processes, and test strategies will allow the development of standard automation equipment and product packaging that will reduce deployment time and capital expenditures.

Figure 2. Standardized fiber-handling cassettes could facilitate the automation process by protecting fragile fiber pigtails. An effort to specify such a cassette is currently underway, sponsored by the National Electronics Manufacturing Initiative (Herndon, VA) and IPC (Northbrook, IL).

Where feasible, optical subcomponents should be designed with fiducials or asymmetric silhouettes to allow machine vision systems to determine proper angular position. It is also imperative that optical components with sensitive coatings be designed with dedicated gripping areas for part handling. Since it is difficult to eliminate fibers from fiber optic components, the industry needs a highly standardized way to handle and control fiber pigtails during the assembly and test process (see figure 2).

Some optical parts require translation and angular adjustment in order to direct a light beam. The mechanical process to perform this adjustment should be identified before the part is designed. A standard optical alignment and fastening machine that can be used for all optical devices will greatly simplify automation. Clearance for the tooling or gripper required to hold parts during assembly is an important consideration.

Cleanliness is also an important consideration for many optical components. The semiconductor industry is shifting from using large, expensive clean rooms to moving material in clean containers, known as standard-mechanical-interface or front-opening unified pods. These containers feature horizontal slots to handle either 200- or 300-mm-diameter wafers and could be adapted to hold trays of optical components. Standard automation equipment exists for interfacing to these clean containers. This approach is typically more cost effective than large clean rooms.

Three fastening processes are used in photonics assembly: epoxy bonding, eutectic soldering, and laser welding. Each has its benefits and drawbacks.

Low-viscosity epoxy wicks into fissures. Parts should be designed both to take advantage of this phenomenon and to prevent wicking where it is not desired. The best use of epoxy is in one or more wicking channels so that part surfaces are not separated by epoxy. Epoxy used as a filler material tends to change shape with time and temperature.

Epoxy bonding is cheap, and curing time varies depending on the technique used. Thermal cure cycles of 8 to 10 minutes are required for the most stable epoxy bonds. UV cures can be as short as 5 to 10 s, but offer less dimensional stability. Epoxies also outgas organic material. This has eliminated epoxy for ultra-high reliability applications, as epoxies can undergo dimensional changes and may contaminate critical optical surfaces over time.

Soldering is fast, cheap, and well supported by the equipment industry. Once solder has cooled, it tends to maintain its dimensions. Unfortunately, its dimensions typically change 4 to 10 µm during the cooling and phase-change process. As in the case of epoxy, the impact of these changes can be minimized if designers avoid using solder as filler between surfaces, but few designs take advantage of this approach.

Laser welding is used for the most demanding optical bonding applications. The machines are expensive, but after the welds are formed and stress-relieved they are stable and do not generate contaminants. Several vendors are working to improve the throughput of laser-welding machines and the process of reducing stress in the welds.

Choosing a fastening process for a particular class of optical designs is one of the most important decisions an optical designer can make. Using more than one in a single design, or even in a single factory, will add cost and complexity.

Testing of optical parts is time consuming and expensive. Current industry practice is to assemble optical components manually, which often requires using expensive test equipment to perform optical alignment. Because the bonding processes are not stable and must be stressed, parts must undergo thermal cycling and retesting. It is clear that product designs and bonding processes that are stable enough to eliminate thermal cycling will improve cycle time and increase production throughput.

Integrating testing into automatic assembly stations eliminates the need to handle components separately and frees up the floor space occupied by dedicated testing equipment. Issues such as test parameters and test points should be specified during product design to minimize costs associated with adding test features. Testing the product as part of the assembly process can increase the throughput of test machines, reduce equipment costs, and improve the economic justification for automation.

Installing automation is not a one-time event. Especially in an industry as young as photonics, both product designs and automation equipment designs will evolve quickly. Companies that install one generation of automation, learn from it, and quickly install the next generation will have the most competitive products. It is likely to take several generations of product and equipment design for this technology to mature.

Using design-for-assembly tools and strategies, engineers should limit the number of part variants and assembly methods per component (see sidebar). Devices must be structured to minimize the complexity of the assembly, test, and packaging processes. With a more consistent approach, standard, cost-effective automation equipment can be developed for the assembly of optical components, mimicking the evolution of electronics assembly equipment in the last several decades. oe

simulating success

To minimize ramp-to-volume time, it is necessary to identify and resolve problems in product or automation design before committing the design to hardware. Most product engineers, however, do not have direct manufacturing experience that would enable them to know how easy or difficult a product is to manufacture by automated means. Simulation and design-for-assembly (DFA) software tools allow teams to evaluate various assembly and product design scenarios, improving eventual yields and throughput by avoiding costly mistakes and high-risk processes from the beginning.

Pilot Yield (Adept Technology Inc.; Livermore, CA) is a simulation package that assists both product designers and manufacturing engineers in making key decisions for part-assembly process design. The software includes a DFA routine that incorporates a scoring engine able to systematically break down the assembly process into a series of tasks with attributes: part features, assembly features, and process features. The software generates a score for each task that represents the degree of assembly difficulty (time and cost) together with a composite automation grade for the total assembly process. Low automation grades don't necessarily mean failure but point to risk areas that should be addressed to maximize production yield and throughput.

The package is compatible with all major 3-D CAD files, which allows designers to use digital mockups of the assembly process to visualize step-by-step tasks and understand the issues. Process yield analysis checks both product and process tolerance stack-ups for fit and clearance problems.

--Joseph Campbell, Adept Technology Inc.