Capital equipment suppliers must provide advanced analytical systems that leverage data generated by their tools to help their fab customers address the challenges of Big Data and advanced analytics. 

BY TOM MARIANO, Foliage, Burlington, MA 

We live in a highly-connected world. Powerful intelligent devices for personal and home use are pervasive and proliferating at an accel- erated rate and will number in the tens of billions in the years to come. These devices are connected to powerful back-end software creating intelligent systems. The semiconductor industry is a major enabler of these intelligent systems. The industry’s drive to adhere to Moore’s Law has resulted in extremely low-cost memory, tremendous computing power and high-speed connectivity, in packages that are low cost and have low power consumption.

These device-level advances when combined with innovations in information technology such as Cloud computing, Big Data and advanced analytics are at the core of intelligent systems that impact our daily lives. Glancing at my phone right now, I see iTunes, YouTube, LinkedIn and my home and work email—all evidence of Cloud computing. Big Data and advanced analytics are widely used for such things as targeted advertising, insurance and credit underwriting, fraud detection, healthcare research, legal discovery, social network analysis and many other areas that impact our lives. Cities around the world, from Da Nang to Fort Lauderdale are applying technologies such as advanced data and analytical tools, cloud-based services and integrated wireless services to make life easier for everyone.

In the manufacturing industry, there is a parallel revolution also leveraging the same advanced information technologies – intelligent manufacturing. The adoption of robotics and automation in manufacturing is increasing precipitously. The use of 3D printing is exploding. Manufacturing machines are becoming more and more intelligent and warehouse automation is rapidly expanding. Intelligent manufacturing systems are dependent on data—data that is shared and acted upon at all levels.

This is leading to changes on the data side as supply chains are being automatically linked for improved tracking and coordination. Advanced analytics are enabling real-time decision making on the factory floor while tool diagnostics are often happening remotely and sometimes automatically. The semiconductor industry has led other manufacturing sectors in the adoption of highly automated, intelligent manufacturing, but is lagging in the application of new information technologies.

Out in front

The need for smaller feature sizes and more aggressive cleanliness and particle-count metrics is the very nature of the semiconductor industry. The accuracy and precision requirements of this complex micro-fabrication process has always necessitated its isolation from direct human intervention. This necessity to isolate semiconductor wafer processing from humans and the drive to adhere to Moore’s Law has pushed advanced technology into the semiconductor manufacturing process resulting in significant progress in automation and optimization of process and production. Clean processing has driven the proliferation of wafer-handling automation within process tools. Wafer-handling robot arms in vacuum and atmospheric tools are standard today. Meanwhile, Moore’s Law played the primary role in wafer size increases and the automation that is present outside of the process tools.

Starting in the 200mm generation, mini-environments (i.e., SMIF pods) as a means to isolate wafers from particles during inter-tool transport became standard. The standard carrier with twenty-five wafers, and its resulting high weight along with the increased fab throughput demands driven by Moore’s Law, led to the propa- gation of inter-bay automated material handling systems (AMHS). The movement of wafers from one processing bay to the next became automated. This trend continued in the 300mm generation with larger and heavier standard carriers (i.e., FOUPs). And with this generation came standardized intra-bay AMHS. Process tool to process tool delivery of wafers was automated as a result. Fully-automated, chamber-to-chamber automation in the semiconductor industry (at least for front-end processing) is decades ahead of other discrete manufacturing industries. In recent years, there’s been an acceleration of robotics within non-semiconductor sectors, but most of these industries are only scratching the surface compared to the semiconductor industry concerning material handling automation.

The semiconductor manufacturing process has also made major advances in data automation. The manufacture of computer chips is extremely complex requiring hundreds of process steps, each affecting change to the silicon wafers at a microscopic level. Also complicating the process is the need for producing multiple products in the same fab with overlapping, but also divergent process steps. This complexity drove the need and proliferation of manufacturing execution systems (MES) in semiconductor processing. Process tool data connections, so-called tool automation is also commonplace, enabling automatic recipe download and tool configuration, remote control and automated data collection. Advanced Process Control (APC) is widely used to improve yield.

And finally, due to the re-entrant repetitive WIP flow required by wafer processing, sophisticated WIP scheduling and dispatching systems exist to optimize, as much as possible, fab throughput and cycle time in pursuit of Moore’s Law. When it comes to data, semiconductor manufacturing is out in front of other discrete manufacturing industries – by far it seems. In the semi industry, the combination of one hundred percent of processing tools connected and automated with metrology feedback loops via APC is not something you see in other discrete manufacturing sectors.

But lagging behind

Recent actions by several large well-known companies emphasize the escalating trend toward intelligent manufacturing. Apple, moving toward fully automated production lines in the U.S., allocated $11B to robotics and automation technology. General Electric announced a $3B investment in the “Industrial Internet of Things.” Google acquired eight robotics companies in 2013. And, Amazon bought Kiva Systems, a warehouse automation company for $750M. Similar actions echoed by thousands of less well-known companies, albeit predominantly on a smaller scale, are also playing a role in the acceleration of intelligent manufacturing. The semiconductor industry is out in front relative to material handling and data automation. However, massive non-semi investment in intelligent manufacturing information technologies is leaving the semi industry lagging far behind.

The use of Big Data, coupled with advanced analytics in the manufacturing process is another area where the semiconductor industry has a long way to go. The amount of data that is needed to be tracked in semiconductor processing is exploding. As design rules shrink to below 32nm critical dimension today and 14nm in the near future, both feature density and the number of transistors per chip experience significant growth. More features per chip translate to:

• taking more measurements
• higher lithography refraction rates resulting in higher error rates
• exceptions requiring more data to resolve and lower yields meaning more excep- tions per wafer (and wafer layer)

As a result, the retention period for these measurements (e.g., to measure tool drift over time) is increased, and the volume of data to be handled by analytics (across lots and tools over time) is magnified considerably. The delayed, but looming transition to 450mm will create a geometric multiplication of the data handling needs.

The value in this massive amount of rapidly created data is in the insight and decision making that can be derived from the data. Here is where the issues lie. Semiconductor manufacturing takes advantage of APC, and in many ways, this is more advanced than a lot of other industries. However, the International Technology Roadmap for Semiconductors (ITRS 2013) emphatically states:

“…a truly comprehensive APC manufacturing strategy is not yet reality, nor is a portfolio of sensors and metrology tools to support complete factory-wide deployment, particularly given the profound changes in materials, processes, and device structures expected for future technology generations. The benefits already realized from APC are driving the development of new sensor technologies and associated control software, which will allow factory-wide comprehensive solutions to be realized in the near future.”

Integrated metrology implementation also presents difficult challenges – metrology tools included as subsystems of process tools. Usually, fabs are designed as a network of tools that each performs one specific function, not multiple functions. This assumption constrains material handling, data flow, MES, etc. Sophisticated, real-time data management and analytics are needed to take advantage of in-situ measurement data with minimal (or zero) impact to tool throughput FIGURE 1 illustrates factory scope and FIGURE 2 shows factory targets as defined by the ITRS.

FIGURE 1. Factory integration scope (Source: ITRS).

FIGURE 2. Factory integration target (Source: ITRS).

Also, in the new “Big Data” section of the ITRS, expected data volumes are shown as “TBD” which is very telling. The units are in Terabytes per day and the possibility that fabs will have to deal with multiples of Petabytes of data is very real. Beyond APC there are other significant data challenges such as traceability to lot and die, test data tracking, predictive tool mainte- nance and Fault Detection and Classification (FDC). The industry is just starting to grapple with how to effectively leverage Big Data and advanced analytics in the semiconductor manufacturing process.

There is a very complex variable interaction problem in semiconductor manufacturing. Going forward, a greater variety of data will be collected at a rapid pace. In many cases, interaction models do not exist today. This will require experimentation and experience to understand interactions in order to derive insight and value from the data. Advanced analytical techniques exist, but determining the right techniques to use for certain decision making will be extremely difficult. Infrastructure and cost are two other issues. The collection, storage and processing of large amounts of data require expensive infrastructure. Support of high data throughput process tool connectivity could require new MES and cell controller architectures. Security is also an issue. Sharing data with capital equipment suppliers and other suppliers will be necessary to derive decision-making value from the data. However this data is highly sensitive and closely guarded by the fab. Similarly, the medical industry is challenged with how to share data aggregated from patient medical records with device makers whose focus is improving patient outcomes in a way that protects patient confidentiality.

Not all the challenges fall solely on the fabs. Capital equipment suppliers have an opportunity to leverage their process and measurement tools to develop solutions to help solve the Big Data, analytics challenges of their customers – the fab operators. Understanding by these suppliers of the environment in which their tools reside will be critical. The system software that runs these tools also becomes more important. The continued development of new process controllers and add-on sensors may require an updated system design paradigm. The data acquisition and management systems of these platforms also need a fresh vision – one that can be implemented in their process and material handling control architectures. Capital equipment suppliers will need to rethink their system design and potentially their business models to leverage the value of the data that their tools can provide.

Conclusion

Semiconductor manufacturing, driven by the need for clean processing and Moore’s Law, leads most other manufacturing sectors in implementing automation and advanced process control. However, large, well-known manufacturing companies outside of semi are making huge investments to progress the use of advanced information technologies in manufacturing because they realize the advantages to be gained. Leveraging technologies capable of handling large amounts of data will provide deeper insights into their manufacturing processes.

The semiconductor industry is poised to take advantage of advanced information technologies. Yes, there is a long way to go and challenges abound. However, the potential value to each fab in addressing key operational metrics such as increased yield, reduced cycle time and increased throughput is significant. The sheer complexity of the interactions of variables in the semiconductor process and the massive amount of data to be collected, stored and analyzed are significant challenges. And, the eventual move to 450mm will compound the huge data volume and velocity issues. I believe that the solution is a collaborative approach – not only fab operators working with software solution providers deploying fab systems, but also in close collaboration with capital equipment suppliers.

Capital equipment suppliers must provide advanced analytical systems that leverage data generated by their tools to help their fab customers address the challenges of Big Data and advanced analytics. It is these companies, who understand best the process data that the equipment can track, interpret and communicate. The semiconductor industry has a long history of fab companies working with their suppliers to further the goals of the industry as a whole (e.g., SEMI standards and other consortia). This effective collaboration model can be used to leverage advanced information technol- ogies for improving the manufacturing process. Semi is lagging, but innovation, drive to success attitude, and organization of the industry will make up the ground quickly.