Automation and comprehensive data tracking are the key means to achieving and improving manufacturing process yields. Cost-of-ownership models indicate that yield and throughput improvements, not initial machine price, have the most significant influence on per-part costs.
In response to the new market dynamics prevalent in the fiberoptics industry, manufacturers are being forced to significantly reduce the cost of their components. This push to achieve lower cost targets requires manufacturers not only to accelerate the time to market for their products but ultimately the time to volume.
The move into high-volume production is generally accompanied by a shift to a more deterministic manufacturing environment—process steps have to occur in a regular and predictable manner and there is little tolerance for process interruptions. High-volume production also involves a higher degree of commitment. A significant number of parts can fill the process pipeline before any finished parts emerge at the end to be measured and tested. As a result, a moment of neglect can result in thousands of dollars of scrapped parts.
In this deterministic production environment, the plant manager is ultimately responsible for throughput and yield, and, to an increasing extent, the variation of the process. As Six Sigma methodology becomes prevalent in a high-volume production facility, the plant manager will also be held accountable for the plant's ability to hold Six Sigma quality levels on all of the key product specifications.
The ability to closely monitor and control the process is the key to meeting the requirement for high yield and throughput and low process variability. More specifically, it is important that the critical parameters of each step in the process be measured accurately and used to influence its real-time control. This is a simple concept, but one that can be very difficult to execute in the high-volume environment. Sometimes a compromise is necessary whereby a key parameter of a particular process step is measured away from the manufacturing line and a time lag exists between the processing of the part and its measurement. This step interrupts the material flow and introduces additional handling and potential yield penalties into the process.
REAL-TIME PROCESS CONTROL
Coupling a fiber to an active photonic component such as a distributed feedback (DFB) laser is known as fiber pigtailing (see Fig. 1). One example of pigtailing involves aligning a ferruled fiber to the focusing lens of the laser and attaching it in place with a clip. Laser welding is often used because of its high throughput and the stability of the weld.
A key parameter of this process step is post-weld shift (PWS). In simple terms it is the movement of the fiber from its aligned position following the imparted energy of the welding process. The problem with PWS is that the decrease in optimal alignment between the fiber and the lens of the laser results in a decrease of coupling efficiency or attenuation of the optical signal. If the attenuation is high enough, this can lead to the component being scrapped or the need for time-consuming rework.
The key to controlling PWS in this particular application is the accurate measurement of its effect. This can be done in real time by comparing the power level of the optimally aligned position to the power level after welding. The difference in power levels can be translated via a spatial profile into a position offset, which can then be used to counteract the shift for the next cycle (see Fig. 2). The more advanced laser welding systems can create a spatial profile of the fiber-to-lens coupling by quickly mapping the power vs. alignment axis movement before welding. State-of-the-art systems incorporate the offset data into a closed-loop feedback system, which measures the offset for cycle N, and then uses this data to influence the welding recipe for cycle N+1. This is a typical Six Sigma style approach for real-time control of the variation for a particular high-volume process step parameter, in this case PWS.
Another fundamental practice in manufacturing process control is accurate process data traceability from chip to ship. Ideally, all critical data can be taken in line or in situ during each test, inspection, or assembly process. However, there are often interruptions or time delays between the processing and measurement. An example of this is post-assembly temperature cycling. After fiber pigtailing, the DFB laser package will typically undergo thermal cycling to relieve the stress in its assembly and test its susceptibility to environmental effects.
At the end of the thermal cycling process the output power of the device is measured and is used in part to determine whether the laser package passes or fails. If a package fails this test there can be a multitude of possible causes, so it is imperative that management be able to track the history of the failed assembly.
There are two primary strategies for data tracking. The first is to have the process data move with the device package. In this case a bar-code reader or RFID read/write system would be employed by each machine to read and/or write the following types of process data to the device carrier—date/time stamp, equipment ID number, cycle count, and device number.
For the process step of fiber pigtailing, the following data could also be tracked: fiber/ferrule lot number, weld clip lot number, welding recipe number, peak alignment power and position, coupling efficiency, L/I characteristics, spatial power distribution, PWS, and cycle time.
A secondary data traceability strategy is to have the process information transferred to the local area network (LAN) by each machine in the line. This requires each piece of equipment to be integrated into the LAN via an Ethernet card or equivalent. It is also not uncommon to see both strategies employed in a redundant data traceablity scheme, where the process data moves with the device and is also updated on the LAN.
LEVERAGING DATA FOR BETTER YIELD
Armed with data-tracking systems such as these, the DFB-laser manufacturer can make a number of tactical moves based on the feedback obtained from each process. By looking at the data associated with the failed device, the manufacturing or process engineer could put a hold on all devices being processed by a flagged machine until the appropriate adjustments are made. The engineer could also identify a bad batch of assembly components, or isolate an area in the manufacturing facility that has fallen out of the environmental specifications such as particle count or temperature variation. This capability can prevent the entire production line from being shut down, saving on delays, inefficiencies, and scrap.
Other requirements of the high-volume photonic assembly process include the specifications for the facility's manufacturing environment (such as particle count) and the reliability, availability, and maintainability (RAM) of the capital equipment. These specifications can have a significant influence on the design of the capital equipment. For example, low particle-count specifications require specific designs and materials to inhibit dust and particle accumulation. High-volume RAM requirements necessitate high mean time between failure and low mean-time-repair performance. More important, these requirements have a direct impact on the potential equipment utilization or uptime percentage, which drives the overall cost of ownership (COO) of the machine.
COST OF OWNERSHIP MODEL
Cost of ownership is the full cost of embedding, operating, and sustaining a processing system in a factory environment.1 The formula set forth by Semiconductor Equipment and Materials International (SEMI) in their standard E35 is:
(COO) = (F$+R$+Y$)/(L*T*Y*U)
Where F$ is the annualized fixed costs in dollars. This includes acquisition costs such as the purchase price, installation, qualification, and training, and facilities costs such as floor space usage, infrastructure needs, and decommissioning. These costs are annualized by dividing the lump sum amount by the useful life (L) of the equipment (in years). R$ is the annualized recurring costs in dollars. This includes costs associated with the factory interface, equipment management, maintenance, consumables and operation labor. Y$ is the annualized yield costs in dollars. This includes costs associated with scrap including breakage and equipment yield. T is the tool throughput in parts per year. Y is the composite yield, which includes the cumulative effects of yield loss. U is the equipment utilization, which includes the impact of unscheduled maintenance, mean time to repair, scheduled maintenance, assist time, production qualification time, and process engineering time.
When applied to a typical photonic component-assembly machine such as a laser welder, this model yields several important findings (see Fig. 3) For example, the purchase cost of the equipment has an almost insignificant impact on the overall cost of ownership of the machine on a dollar-per-part basis. In general, the throughput, labor to operate, and yield-loss costs have the biggest influence on the cost of ownership and return on investment. Because automation generally has the greatest influence on throughput and yield improvement, it is important to consider the savings associated with these factors along with the purchase price when analyzing any capital equipment purchase.
Frank Vodhanel is a product manager of laser welding systems at Newport Fiberoptics and Photonics Division, 1821 E. Dyer Road, Santa Ana, CA 92705. He can be contacted at firstname.lastname@example.org.
- E35: Cost of Ownership for Semiconductor Manufacturing, www.semi.org, SEMI (1999) San Jose, CA.