Download this article in .PDF format
This file type includes high resolution graphics and schematics when applicable.
Semiconductor manufacturing generates lots of data, making it ideal for big-data analytics. Focusing on improving product-related issues can have a positive impact on yield, quality, and productivity. One company working in this space is Optimal+. I spoke with Dan Glotter, CEO and Founder, to find out more about the company’s software and the trends in big-data analysis for semiconductor production.
Wong: There is so much manufacturing data already being produced, why big data now?
Glotter: The question isn’t “why big data now,” the question is “big data for what purpose?” From the beginning, Optimal+ decided not to concentrate on issues within the fab (process-related issues). Instead, it focused on product-related issues where the benefit of using big-data analytics was obvious to us. Some of our customers have 15,000+ products and all of them need attention in manufacturing from product, quality, and yield engineers.
In addition, annual semiconductor production is growing rapidly. The big data being generated from those thousands of devices, at the volumes that are being produced, simply cannot be analyzed quickly enough with traditional analytic tools. Optimal+ enables our customers to analyze their manufacturing data within minutes to quickly find problems impacting product yield and quality, while at the same time helping manufacturing operations improve productivity. The data challenges for semiconductor companies will only continue to grow, and Optimal+ can help these companies become proactive with the manufacturing data instead of being reactive, regardless of their product mix or volume.
Wong: Is the explosion of chip/device demand and quality a factor in using big-data analytics?
Glotter: An explosion in the use of semiconductors and sensors is being realized in new types of electronic systems, such as autonomous automobiles. These systems are moving from being “mission-important” to “mission-critical” and require semiconductor companies to take an entirely different approach to quality and yield that requires next-generation analytic solutions.
Wong: Why are semiconductor companies now concerned about real-time big-data analytics in the supply chain?
Glotter: To provide a high level of quality and reliability, it is not enough to simply analyze data. Companies need to use analytics to create knowledge that can impact manufacturing. If you look at a multichip package, it is one thing to know about the components used to create a package (the bill of materials, or BOM), but it is an entirely different challenge to have the ability, in real time, to direct the quality or performance of individual components within a package to get the best possible result (smart-pairing of devices).
Wong: What types of supply-chain decisions and other organizational decisions can be made with the help of big-data solutions?
Glotter: Many different types of decisions can be made across a company through the use of manufacturing test data. For example, we have customers that use their manufacturing data to drive product planning, financial reconciliation, and purchasing decisions, in addition to improving aspects of manufacturing operations (yield, quality and productivity). This is made possible because all of the business units of a company are using, for the first time, a common, consistent set of data for all of these decisions. In the past, every business unit collected and used its own data.
Wong: What does Optimal+ solve and for whom?
Glotter: Optimal+ solves several problems for semiconductor companies (IDM or fabless) by providing visibility into their global supply chain. We collect real-time data from across the entire manufacturing supply chain and deliver actionable information within minutes that can be leveraged by all of the different functions within a company (finance, product planning, manufacturing operations).
Prior to using Optimal+, every business unit had to collect and maintain its own data, which resulted in very little effective sharing. Using big-data analytics, these semiconductor companies can now solve issues that could not be addressed in the past because their data was so fragmented and incomplete. Our customers can now use a single, comprehensive database to identify and solve manufacturing problems, such as yield, quality, and reliability across their entire company.
Wong: Does this solution focus on a particular part of the production cycle, either at the beginning (New Product Introduction-NPI) or post-design (High-Volume Manufacturing-HVM)?
Glotter: Today, we are focusing on Characterization/NPI and post-production manufacturing. We don’t deal with fab data because we feel that process data is not necessary for the mission we are addressing for our customers. The beginning of the production lifecycle (NPI) is very important product-wise, and then from the moment the wafer is completed until final test, our solutions can be used to manage every step of the process.
Wong: How will semiconductor companies support the collection, storage, and management of the new data?
Glotter: Today, many customers use traditional storage methods for their data needs. However, with the big-data trends we are seeing, we are working with our customers to develop BKMs, or Best Known Methods, for the storage solutions that will be necessary to address significantly larger data sets. This is also why we are moving to HP Vertica to power our analytic solutions. Several years down the road, as customers want to move their older data to archive formats, we will work with customers to set up inexpensive file-based archive solutions that still meet their analytic needs (e.g., RMA analysis). Finally, we are seeing many customers moving to a cloud-based deployment, which gives them many more options for big-data storage (e.g. Amazon Cloud).
Wong: Is the Optimal+ solution an executive-suite reporting tool or an operations management tool?
Glotter: The answer is yes to both. This is the nice part of the Optimal+ solution, because it appeals to a very broad range of users. Due to the vast amount of data that is collected by our solutions, our analytics can be used by engineers and engineering management, people in operations, product planning and purchasing, and finally C-level executives. Our solution is truly an end-to-end solution that can be used by the entire enterprise.