Customer Story

Illuminating a green revolution

Cree Inc. uses JMP^® statistical software from SAS to identify, analyze and address imperfections in its manufacturing process.

Cree, Inc., the market-leading innovator of LED lighting based in Durham, NC, has a mission: to eliminate all energy-wasting light bulbs from the face of the Earth. Poised for some time on the precipice of a seismic shift in lighting, Cree is today demonstrating that the movement is now well under way – and the revolution is a green one.

Lighting is at the heart of city infrastructure – affecting everything from quality of life and jobs to operations and commuter safety. A well-designed, high-quality LED street lighting system can improve visibility and the performance of other safety systems, while providing years of maintenance-free energy savings. With 40 percent of its street lights rendered broken, the city of Detroit selected Cree to install 32,000 LED street lights throughout the city as part of its three-year, $185-million street lighting overhaul – an effort critical to the city’s revitalization and ability to overcome a crippled economy. As of early 2017, the project has resulted in $3 million in energy savings.

The city of Los Angeles is also on board. After installing over 150,000 LED streetlight fixtures, Los Angeles observed a monumental 63 percent reduction in energy use. Between energy savings and reduced maintenance costs, the city will see an annual savings of $10 million.

Energy-efficient LEDs translate into financial and energy use savings

Suffice it to say LED lighting is here. LED lights can last for decades and use only a fraction of the energy required of incandescent or fluorescent bulbs. LED technology has been around for about 40 years, but only in the past several years have manufacturers like Cree improved the quality of the light emitted by LEDs to make them a viable option for lighting homes, stadiums, parking lots and other venues.

Cree is now leading the charge toward the obsolescence of inefficient lighting and the proliferation of this energy-efficient, environmentally friendly alternative. Cree lighting-class LEDs now surpass a record-breaking 300-lumens-per-watt threshold, and that “is huge,” says Ed Hutchins, Product Engineering Manager for Cree’s materials division. “It means that our LEDs are very bright,” Hutchins says, “and when you can use fewer LEDs in a bulb or fixture, that can drive the cost down. Certainly our customers – the ones who make the lamps and the fixtures – have realized the benefit of these LEDs.

“And when they see how good these LEDs look in their applications – that they don’t look blue or green – and they realize that for a slightly higher upfront cost they can start saving money in a year or two, they’re on board.”

Hutchins’ group is responsible for creating the silicon carbide wafer substrate that is the foundation for Cree products. His engineers have used JMP for years in a number of pursuits – data analysis, process control, designed experimentation, Six Sigma initiatives and more. “Data mining and data visualization are two of the things we use JMP for very extensively,” Hutchins says. He describes a recent use of JMP to detect and remedy a defect in the manufacturing process.

Data visualization helps engineers spot defects

In general terms, Hutchins’ group takes a silicon carbide crystal, slices it into wafers and then prepares the surface of those wafers, polishing and cleaning them so that they’re smooth and defect-free. That prepares the wafers for the next step, which is epitaxy, a process that applies the active layers that allow a device to function in the manner it’s designed.

In an LED, light is generated by the epitaxial layers, and the quality and color of the light is dictated by the quality, structure and composition of those layers. Delivery of a consistent supply of high-quality wafers by Hutchins’ team is necessary for consistent delivery of a high-quality finished product. But a downstream customer was detecting a defect that could have a major impact on yield and traced it to the tools for which Hutchins is responsible.

“The first application of JMP was in visualizing how frequently the downstream group was seeing this defect,” Hutchins says. Hutchins’ group connected to the database the downstream group uses for storing characterization data. Once retrieved, that data was manipulated in JMP to generate control charts that illustrated the day-to-day trend of the defect.

“This is a good example of the power of visualization with JMP,” Hutchins says, “in that there was a false signal and a real signal. The false signal was how quickly the failure rate escalated. It had been zero, then jumped to 20 or 30 percent in a matter of just a few days. But that was a red herring.”

In fact, it was heightened scrutiny of the defect that was causing it to show up more frequently. It was therefore determined to limit the analysis of data to a given time period with consistent conditions. Hutchins and his team then wrote a script in JMP to query for data, including the processing date and time, the batch, the tool and the process conditions. “If we were to try to collect that data set by writing an Access query on our own,” Hutchins says, “it would take hours and hours to run. Doing it in JMP allows us to do it much, much faster.”

Team members then used JMP to build a model that included only the most significant factors. “The factors in the model were the process steps,” Hutchins explains, “and the values for each factor were the sister tools at each step.”

At this point, they introduced the interactions, looked at their effects and – using some engineering judgment and the Prediction Profiler in JMP – determined precisely where in the process to focus their investigation, narrowing it to two potential steps. To their surprise, they found that although these two tools are used in steps that are separated by great distance in the process timeline, and even separated physically in the plant, there is an interaction between them.

“Seeing the interaction so clearly in this model, as visualized in JMP, was the first time we’d ever contemplated that there might be this type of interaction,” Hutchins adds. That interaction was the source of the problem. An action plan for fixing the problem was then initiated.

The real benefit of the approach chosen using JMP, Hutchins says, “is not in finding the answer to the problem, but in pointing you in the right direction so that you know where to look to find the problem. I was able to understand the differences in those tools that I may not have appreciated before. By knowing where to look and then looking there, we were able to drive the failure rate for that defect back down to zero.”

With JMP, Cree’s engineers maximize the insights they get from their data

“I prefer JMP because it’s so powerful in visualizing data,” Hutchins says. “It allows you to generate the types of control charts, graphs and displays that then allow you to understand that data. The most important thing is to get the right data; the second most important thing is to understand what that data is telling you. And to understand it, you have to look at it as a function of a number of parameters, so that you understand the space in which you’re operating. With just a table of data, you can’t do that.

“There is other software out there that will generate graphs and other displays, but it doesn’t have the statistics that JMP has. So if I’m trying to look for a statistically significant difference between these distributions, or to run a full-factorial experiment and understand the effects of each of these factors, I can’t do it in other software; but I can in JMP. JMP has a good combination of the data visualization and statistical power that makes it a good choice for us here. We use it a lot.”

“I have no doubt that our success is due to our ability to continue to innovate and improve,” Hutchins says, “and that our ability to innovate and improve is greatly enhanced with tools like JMP.”