+44 (0)24 7671 8970
More publications     •     Advertise with us     •     Contact us
 
Technical Insight

Early implementation of SPC at Lasertel (Cover Story - FAB Operation and Control)

Statistical process control (SPC) is one of the keys to creating an efficient manufacturing process for the production of semiconductor lasers, writes Stewart Wilson of Lasertel.
Traditionally, high-power semiconductor laser companies have created and assembled products by hand, using experienced engineers in a research and development setting. This is typically a slow process, further handicapped by a high scrap rate. Today, the demand for high-volume manufacturing is making the creation of handcrafted diode lasers a thing of the past. To succeed in the current marketplace, companies must take the alchemy of semiconductor laser production and transform it into an efficient manufacturing process. This has been at the core of the startup success at Lasertel, a high-volume manufacturing facility in Tucson, Arizona that makes high-power semiconductor lasers (see Compound Semiconductor May 2001, p63). Along with lean manufacturing and demand-flow tools, Lasertel has used statistical process control (SPC) to streamline production, improve repeatability and yield, and reduce the overall costs of its single- and multi-emitter diode lasers. SPC theory An ongoing and continuously evolving process rather than a quick-fix quality management system, SPC uses statistics to identify variation in a process. Steps are then designed and implemented to correct these variations, bringing processes under ever-tighter control. Traditionally, manufacturers controlled quality by inspecting products coming off the line, auditing factory processes, and taking corrective action when yields started to slip. With SPC, companies monitor their processes throughout production, analyzing data and taking corrective action in real-time. Such real-time data collection, analysis and process correction requires significant investment by a company, and certainly demands top-level management support. To objectively pinpoint variations, SPC analysis needs a foundation of statistical information; obviously it works better as its database of information grows. Indeed, some startups have often viewed SPC as not that valuable, feeling that data sets are too small to be statistically valid, and that too many variables affect brand-new processes. However, the experience at Lasertel has been quite the opposite SPC has offered significant start-up advantages. Implementing SPC at Lasertel A key to Lasertel s SPC implementation was its lean manufacturing methodology, incorporated into the design, start-up and operation of production early on. This methodology which maintains a small WIP (work in process) inventory, helps improve yield and reduces scrap has paved the way for the successful implementation of SPC. Integrating lean manufacturing principles, such as a flow-enhanced floor plan, a visual factory, paperless travel, cross-trained employees and one-piece flow, has led to Lasertel s efficient and easily monitored operation. In addition to lean manufacturing, Lasertel organized its flow of production to incorporate inspection, verification and data collection at all critical steps. Examples include the following:
  • Every piece of equipment has its own SPC system. Data is extracted from microscopes set to analyze visual defects, or spectrometers that measure reflectivity, and control charts analyze raw power, temperature, water settings and other parameters.
  • All stations, from bonding to lensing to fiber coupling, have customized tracking databases. Data is either collected automatically or entered by the operator.
  • Tests are designed to measure and report on performance, environmental factors, processes and material quality. In measuring operator performance, for example, the inspection yield per operator has varied from 36% to 90%, achieving 100% at some stations. While a low yield may indicate a need for more training, it can just as easily point to a problem with equipment or with raw materials.
  • Programmers work in close partnership with line managers and suppliers to automatically extract data from all areas, pulling it together with the help of customized SPC software to produce real-time reports.
  • Data is entered into touch-screen PCs at each station so that the real-time advantage of SPC is maintained, rather than being slowed by report filing and paper administration. With this initial commitment to an efficient process, Lasertel was able to use SPC tools almost immediately, enabling more objective decisions based on real-time information and allowing problems to be addressed proactively. For example, during chip-on-submount bonding which forms a building block for a laser package, the quality of the bond is monitored to determine whether there are voids or other defects that may impact performance, and whether performance falls within specification because of that bond. Rather than have a quality department that actively audits the final products, SPC integrates inspection within the line. Operators at each station test or inspect the product, enter relevant information, perform their task, and send it to the next step. Defects are caught early, and this points directly to a possible process problem and saves the time it would later take to identify where the problem originated. It also saves money by minimizing the loss of value-added product that would occur if the defect was not caught prior to submount attachment. SPC analysis A myriad of statistical analysis tools can be tailored to each company s unique processes and perfected through trial and error. Some of the basic SPC analysis tools used at Lasertel include:
  • Flowcharts that graphically represent process flow, showing process inputs, activities and outputs in the order in which they occur.
  • Check sheets that list items inspected in a standardized format so that required information is gathered in the same way each time.
  • Attributes charts that qualitatively plot defect information with constant sample sizes (C-charts), varying sample sizes (U-charts), actual number of defects (NP-charts), or percentage of defects (P-charts).
  • Pareto charts that identify the problems of greatest concern and provide a prioritized order in which to address these problems.
  • Cause-and-effect diagrams that show the interrelationships between groups of causes and effects, usually in the form of a tree or fishbone diagram.
  • Control charts that shows statistics on important characteristics of a product to provide information for correction and improvement of processes.
  • To date, Lasertel has made a great deal of use of the X-bar and R-bar control charts. These charts plot data as a histogram to show average (X) values of samples and variations in the range (R) of data subsets. A basic X-bar chart has a center line and upper and lower control limits, most often set at 3 standard deviations from the mean. A certain number of points in a row outside the limits indicates that the process is out of control. As the process gets tighter through SPC, the allowable number of points can be adjusted, further fine-tuning production. Once SPC has indicated a variation or potential problem in a process, a Lasertel team of engineers, technicians and operators immediately addresses it. If the cause is not apparent, a design of experiments may be used to run tests that vary critical characteristics to see what occurs. When certain cause-and-effect relationships are known through prior experience, fault tree analysis is used to track the origin of a problem. Gauge R&R (repeatability and reproducibility) studies are crucial for making sure measurements are valid. They test whether the same result will occur if the same part is measured with the same machine over and over again or if it is measured on several different machines. SPC at fiber coupling SPC has already contributed to significant improvements in Lasertel s semiconductor laser production. One of the best examples to date is a turnaround in fiber coupling. Through tight SPC tracking, it recently proved possible to resolve low power readings at the fiber coupling stage. The flow chart in shows a simplified version of process flow in the fiber coupling of Lasertel s Quad Blaster, a four-channel multi-emitter package, along with some of the SPC analysis controls we used to monitor the process. First, the Quad Blaster is assembled and an attributes P-chart (percentage of defects) is used to qualitatively measure defects. Second, the package is tested and an X-bar/R-bar chart (X-range chart) is used to monitor power. Lastly, a fiber jumper is coupled to the Quad Blaster, and another X-range chart is used to monitor the power after coupling. Initial monitoring of the P-chart revealed a greater percentage defect than expected, suggesting that it should be possible to improve the process and have an impact on the final yield. Several modifications were introduced, which were verified by a reduced mean percentage of defects on the P-chart (). However, after process optimization of the Quad Blaster assembly and revised test criteria we continued to see low yield and poor process control at fiber coupling. Since our charting of the fiber jumper line () indicated that we were manufacturing good jumpers, we performed a gauge R&R test on the jumper test measurement. We found that our current measurements were not adequately quantifying the jumpers and, hence, we were not properly filtering all the low-performing ones. We developed and implemented a new technique for quantifying the jumper performance and immediately saw a 14% increase in mean fiber-coupled powers, from 0.92 to 1.05 W (see ). Perhaps more importantly, there was also a tighter distribution about the mean, indicating better process control. As a result of these findings and our improvements, we were able to increase the lower control limit from 0.76 to 0.90 W. In this way, SPC continues to refine the process. Conclusion Lasertel has benefited from a startup infrastructure that is based on its lean manufacturing and mathematical controls. A top-down commitment to establishing SPC throughout the semiconductor laser production line, starting with epitaxial wafer growth, continuing through photolithography, and finishing with final packaging, has helped to more than double throughput less than one year into production. As Lasertel continues to use SPC to narrow the acceptable range of variation in our processes and document normal parameters, complex processes continue to be standardized. Machines can be calibrated even more accurately, and preventative maintenance can be scheduled more often. Formulas for applying metals, and the complex ratios for layering and temperature control can be broken into a series of more simplified steps. The end result is that as processes are tracked and continuously improved, technical staff will be able to turn more of the operation over to trained operators and technicians, production will become more streamlined, volume will continue to grow, yield will improve, and prices will go down. In a market currently challenged by high scrap rates and low-volume manufacturing, SPC has helped Lasertel create a repeatable recipe for success. Further reading R Clements 1988 Statistical Process Control and Beyond. Robert Krieger Publishing Co, Malabar, Florida. L Doty and A Leonard 1996 Statistical Process Control 2nd edn. Industrial Press Inc, New York. N Gianaris and R Green 1999 Statistical methods in material manufacturing and evaluation Materials Evaluation 944. C Rauwendaal 1993 SPC--Statistical Process Control in Extrusion. Carl Hanser Verlag, New York. T Robinson et al. 2000 Statistical process control: it s a tool, not a cult Manufacturing Engineering 3 104. W Woodall and D Montgomery 1999 Research issues and ideas in statistical process control Journal of Quality Technology 31(4) 376.
    ×
    Search the news archive

    To close this popup you can press escape or click the close icon.
    ×
    Logo
    ×
    Register - Step 1

    You may choose to subscribe to the Compound Semiconductor Magazine, the Compound Semiconductor Newsletter, or both. You may also request additional information if required, before submitting your application.


    Please subscribe me to:

     

    You chose the industry type of "Other"

    Please enter the industry that you work in:
    Please enter the industry that you work in: