YUJI MINEGISHI, General Manager, Gigaphoton Inc.
When it comes to research and development, the semiconductor industry has a strong history of achieving transformational innovation by means of openness and knowledge-sharing. Since no single organization can bear the burden of the enormous expense involved in semiconductor research and design, industrywide initiatives—even among traditional competitors—have paved the way for innovation that advances the entire industry.
To achieve industrywide advancements, the semiconductor industry has formed consortia, forums, and technology roadmaps and has even shared resources to build industry infrastructure. [1] In fact, the semiconductor industry has such a long and strong history of open collaboration that it has been called “a trailblazer in pre-competitive collaboration” and is seen as a model for other industries to emulate. [2]
However, outside of research and design, the semiconductor industry looks like many others. Semiconductor manufacturing is characterized by trade secrets, siloed and proprietary intelligence, and fierce competition. Given these constraints, is anybody actually winning?
More and more, chip manufacturers are focused on gaining process efficiencies through advanced analytics. While the need for greater efficiency isn’t new, its impact was recently highlighted during the pandemic, which brought with it an unprecedented global semiconductor shortage. Yet, even as the world cautiously emerges from the pandemic and related shortages, it’s apparent that chip manufacturers will continue to find themselves behind the curve as technology advances and demand continues to increase. Existing proprietary solutions for yield optimization will not generate the efficiency required to keep pace with ever-increasing worldwide demand for output and speed.
Advanced analytics for yield improvement: Opportunities and challenges
With every new generation of chips comes a new generation of challenges, as well as new opportunities for innovation. As semiconductor manufacturing processes become more complex, there are thousands of steps, thousands of parameters, and endless combinations of these variables. This level of complexity results in enormous yield losses that are beyond the capabilities of traditional quantitative analysis to address. [3]
The answer, according to an analysis by McKinsey & Company, lies in advanced analytics that are capable of detecting defects early, identifying root causes of issues, and identifying improvement opportunities that are too complex for quantitative analysis to uncover. [4] However, only a small number of chip manufacturers have software capable of performing the most advanced predictive (14%) and prescriptive (12%) analytics. [5] Moreover, the effort it takes for engineers to pull data, clean it, and analyze it is incredibly costly and time consuming—and leads to less uptime. FIGURE 1 shows the results of a survey conducted by Gigaphoton Inc. into these issues.
Yield optimization is a competitive advantage. So, it follows that the data and analytics tools that chip manufacturers use to monitor and improve their processes are proprietary. There are so many monitoring and analytics solutions—each offering different functionality or focus and using proprietary code bases—that no single solution meets the needs of the entire industry, or even of a single fab. And since there is no integration among them, most chip manufacturers have to piece together multiple analytics solutions as well as develop their own in-house tools to gather the intelligence they need. [6] This lack of integration contributes to intelligence silos and massive inefficiencies for chip manufacturers across the globe.
No standardized analytics platform exists for the semiconductor industry. Seventy-three percent of chip manufacturers piece together multiple disconnected equipment analytics solutions to get the information and intelligence they need for improving output and yield. [7] Major equipment manufacturers have developed suites of software to make their own equipment run more efficiently and predictably, but each vendor’s solution applies only to its own tools. This means that when there’s a bad yield, chip manufacturers aren’t empowered to find their own solutions. Instead, they move from vendor to vendor in search of the source of their problem. And when chip manufacturers need custom solutions developed for their particular manufacturing operations, it often takes 3-6 months to get the software features they need—if they ever get developed at all. [8]
When asked about the time it takes to get the solutions he needs, William Haskell, site technology manager at Honeywell Sensing and IoT, expressed frustration with the time and lack of agency involved in dealing with all manufacturing software solutions: “I want it to be relatively easy to configure and model and change myself, without having to go back to the vendor all the time for every modification,” he said.
There are solutions that aim to bring together in one place all the data coming from the many tools and sensors in the production line. Manufacturing execution systems (MES) claim to offer a complete view of manufacturing operations. Yet, simply collecting the data doesn’t enable the complex analysis that is needed to optimize complex processes. Haskell noted that many factories have an enormous amount of data but “don’t know how to use it properly, or don’t know what to do with it, or don’t have a tool that brings it all together, even if they’re gathering it all automatically.” Moreover, while gathering the data may be an automatic process at most fabs, analyzing it is still very manual.
Surya Iyer, vice president of operations at Polar Semiconductor, explains that for every machine, “there are tens of sensors … that is tens of charts someone has to set up, watch, or react to when something goes wrong. And in a fab with hundreds of machines, times tens of charts … that is a crazy amount of work … all of this is very inefficient.”
What the industry needs is a way to quickly, easily, and accurately transform enormous amounts of data into actionable insights and predictions. In fact, some third-party software solutions propose to do just that by integrating data across the entire production line and applying predictive analytics. Some equipment manufacturers have entered this software space as well. However, these solutions work with proprietary code bases that don’t provide the flexibility necessary to meet the unique analytics needs of every chip manufacturer. Moreover, they are only pulling and stitching together data from existing data lakes that may not be optimized for analysis (FIGURE 2). Every piece of equipment offers a different set of parameters, and there isn’t an easy way to standardize the data or facilitate accurate advanced analytics with the depth of intelligence that only the tool manufacturers themselves can provide.
As a result of these limitations, chip manufacturers have found that none of these solutions meet all of their needs. Chip manufacturers spend as much as $220,000 per year on analytics software, [9] but 75% of them have to develop their own in-house software solutions to augment their purchases. [10]
Data and analytics platforms—vendor-specific as well as third-party—are big business. Despite the industry’s history of collaboration when it comes to research and design, when it comes to manufacturing and data analytics, the entire industry is competing, but nobody is winning. The industry needs a better way.
A better way: Open access
If the semiconductor industry is going to achieve the precision, output, and speed necessary to keep up with increasing worldwide demand and progressively more complex manufacturing processes, there needs to be a change in mindset, away from proprietary solutions and toward integration and democratization of the tools required to build more intelligent fabs. To develop something that could contribute to the success of the global semiconductor industry, what’s needed is a new, agnostic analytics framework, an industrywide open platform for equipment monitoring and analytics.
Click here to read the full article in Semiconductor Digest magazine.