Future nodes require new defect reduction strategies
Existing defect reduction protocols have proven less than ideal for reducing waste and speeding production as transistor geometries shrink and process complexity increases. Brewer Science offers insights into tools and analytical techniques that can improve defect elimination in devices below 10 nm. By Darin Collins, Director of Metrology, Brewer Science
As the semiconductor industry migrates to advanced lithography at the 10-nm node and beyond, standard best practices for defect reduction will be insufficient. Contamination levels will need to be measured in parts per trillion (ppt). Defect reduction at this level requires improvements in analytical tools, quality control (QC) and quality engineering (QE). Semiconductor manufacturers looking to reduce defect levels for advanced nodes must carefully control and characterize their entire supply chain, starting with raw materials. Raw materials suppliers typically provide data on the purity levels of their chemicals, but purity is insufficient to give semiconductor manufacturers confidence that the material will enable them to meet their yield requirements. They need data on the detailed impurity profiles at the level of parts per billion (ppb) or ppt. This requires either going to their materials suppliers’ sub-suppliers or conducting extensive testing on all materials received.
What can raw materials suppliers do to differentiate themselves and gain the confidence of potential customers in the semiconductor industry? Quite a bit. At least three primary opportunities exist for companies developing materials and processes to support semiconductor and microelectronic device fabrication: data collection and analysis; factory automation; and a cultural emphasis on quality.
Although it is not yet standard industry practice, some suppliers provide mass spectrometry data on their chemicals. Such data describe levels of multiple metal impurities at the ppb or ppt level, and can track the history of these impurity levels. Brewer Science, for example, monitors levels of at least 10 common metal contaminants, with detection levels ranging from 4 to 13 ppt, and provides the resulting data to customers. Semiconductor manufacturers can work with a certain level of impurities, so long as values are consistent and manufacturers understand how different impurities are affecting their yields. By continuously tracking data on metal impurities, chemical precursors and other contaminants, any deviation from the baseline shows up. The better manufacturers understand how these deviations affect device yield, the better they can optimize their process to keep yield as high as possible. The approach must be collaborative in order for these slight deviations to be understood. Suppliers can incorporate several techniques to decrease the level of impurities in their products and improve consistency in impurity profiles. These practices include factory automation, closed-loop systems and failure mode effects analysis (FMEA).
Figure 1: Brewer Science’s scrap cost as a percentage of revenue, showing reduction each year.
Factory automation improves production quality in several ways. In the world of manufacturing, there should ideally be no such thing as employee error, and factory automation is a necessary step toward decreasing it. Factory automation that improves product consistency can dramatically reduce the volume of product scrapped, even as production quantities increase (Figure 1).
An optimized factory automation setup (Figure 2) will allow employees to see every part of the manufacturing process on a single screen, enabling them to easily monitor those and act immediately if anything is out of specification. Alarms and alerts in real time can be sent via text and email, allowing employees to monitor systems remotely via their smartphones.
Figure 2: Factory automation portal. In this example, tank 1 has an issue that needs to be addressed
An optimized factory automation setup (Figure 2) will allow employees to see every part of the manufacturing process on a single screen, enabling them to easily monitor those and act immediately if anything is out of specification. Alarms and alerts in real time can be sent via text and email, allowing employees to monitor systems remotely via their smartphones.
An automated factory creates a massive stream of data. Quality engineers can analyze these data to continuously improve their processes. These results allow employees to understand the impact of quality improvements, and empower them to focus on product quality. Having a history of the impurity profile of a product makes it easier to know where to tighten processing specifications to produce a more consistent result.
Factory automation also reduces the possibility of contamination. At Brewer Science, the blending and bottling of chemicals occurs in a closed-loop system (Figure 3). What is today a competitive advantage will soon be necessary to attract customers. As their own defect requirements tighten, semiconductor manufacturers will begin demanding higher levels of cleanliness, automation and characterization from their raw materials suppliers. The traditional advanced on-wafer defect instrumentation isn’t able to detect the impact of raw materials in the supply chain. Typical solutions are found in the sub-supplier detection capability and process stabilization.
QE plays an important role in achieving continuous improvement. While FMEA is a standard tool in many manufacturing environments, it isn’t always used optimally. The ideal approach to FMEA is to evaluate all possible failure modes together and incorporate lessons learned to avoid repeating mistakes. FMEA is useful at many stages of product development: when a product or process is first being designed; when it is being applied in a new way; or when analyzing failures or planning improvements for an existing product or process.
Figure 3: Closed-loop system from feed stream in preparation for blending materials at Brewer Science’s facility in Vichy, Missouri.
In semiconductor manufacturing, keeping yields high is critical to achieving cost control. Incoming raw materials may pass the manufacturer’s specifications, but the yield on a 300-mm wafer might be too low to be cost-competitive. In that case, semiconductor manufacturers need to re-evaluate their entire process, starting with raw materials.
If the purity of incoming materials is partially responsible for yield loss, the material supplier needs to work with their customer, the semiconductor manufacturer, to help fix the problem. Choosing suppliers that are on the leading edge of automation and characterization will help avoid potential problems. Such suppliers should also be responsive in the case of yield issues linked to their materials.
Computational chemistry stands as the next step in further improving raw materials purity. Molecular modeling can point the way to improvements at the molecular and atomic levels that will be needed to design the next generation of raw materials for semiconductor manufacturing. Modeling can predict the interaction of specific metal contaminants with a polymer base and model flow during spin coating to predict defect distribution. Computational fluid dynamics can model fluid flow at a microscopic level to ensure consistent mixing during material production and to further drive down defect density. Brewer Science is already using such modeling to improve product performance.
Last but not least, a cultural focus on quality and people plays a vastly important role in these processes and techniques. It requires people’s commitment to quality from the top down, going beyond product/process to become a mindset for success in innovation. Empowering the technicians to make meaningful changes and understand the effects on the customer, and the customer’s customer, will emphasize the individual contribution to technology advancement.
Tomorrow’s semiconductor devices will demand a new level of defect reduction. A process that incorporates a focus on quality throughout the supply chain should enable manufacturers to meet ever-more-stringent requirements and produce high-yielding devices.
