the Blog

Run-to-Run (R2R) process control to enable next generation Semiconductor capability on existing capital equipment
Abstract
In Semiconductor manufacturing (DRAM exampled here), the trigger for investment into new capital equipment and processes hinges on observed process performance of printed feature Critical Dimensions (CD) and Layer to Layer pattern Overlay Alignment (OL).
If we consider the IRDS Lithography Roadmap below for DRAM, by 2024 required manufacturing capability for CD and OL is 1.4um and 2.8um respectfully.

Since volume manufacturing requires numerous process and metrology tools to manufacture products cost efficiently, total CD and OL (3 sigma) error must all budget within expected manufacturing variation from processes, tools and metrology.
To realize the aggressive tolerances seen in this DRAM Roadmap, operational data, and collected measurements must be exploited to keep all material on-target and within specification.
Discussion
In this process control discussion, the following "Levels" of Manufacturing Process Control will be described and illustrated using a semiconductor lithography process example. This discussion will focus on Level 0, 1 and 2.

Level 0: Statistical Process Control (SPC)
Consider a semiconductor Lithography process. Key metrics for controlling printed features include Critical Dimensions or, [CD] and Layer-to-Layer Printed Feature Overlay, or [OL]. That is, the size and position of printed features.
With high tolerance products, i.e., where inherent tool capability and process windows exceed required tolerances for a given product, the operational process starts with traditional SPC monitors to assure in control performance for CD and OL. At this level, SPC Charts can be at a high control level, e.g., by Technology, where multiple Products, Process Tools, Masks or Reticles are all lumped together.
Although such ‘lumped’ process monitoring may appear normally distributed, there is most often systematic difference across context features like process, tool and product.

At this process control monitoring level example, Process Capability (Cpk) is healthy, i.e., above 1.33 (4 sigma process). In some industries, a 5 sigma (i.e., Cpk > 1.67) or even 6 sigma (i.e., Cpk > 2.00) process is targeted.
The simplicity of process monitoring in this situation is certainly favorable, as there is only a small number of charts to manage and report.
The downside here though is that systematic components of variation are not visible and process excursions can then lead to time consuming data investigation and root-cause analysis.
The next progression in the Level 0 process control tier is to split out SPC charts into individual ‘control threads.’ For example, the most logical example is SPC by Process Tool. With this approach, process tool owners can begin monitoring the trends for individual tools, where systematic trends or excursions can be tracked down and resolved more quickly.
This transition from Process Level SPC monitoring to Process-Tool Level is most commonly instituted when Process Level Capability becomes marginal.

By applying the next generation (aka, next node) specifications, it is observed that overall Process Capability jumps from a 4 sigma healthy performance to a Not Capable (i.e., Cpk < 1.0) performance.
At this point, the granularity of process SPC monitoring increases, first to by-Tool and then even by-Tool and by-Product. In many situations, there is a trade off as partitioning the data too much can lead to sparse sampling and extended detection times of process excursions determined from statistical trends.

By simply enabling the Sentient color-by feature, separation by Process-Tool becomes clearly visible as show in Figure 5. This shows that a by-Tool level SPC approach does lead to a good progression for process control and capability by giving more transparency to Tool matching.
Level 0 Process Trends (by-Process Tool)
After splitting things out by Process Tool, systematic differences between tools becomes immediately visible. Specifically the mean difference tool to tool, i.e., +0.48, -0.82, +2.34 for ARF1,2 & 3 respectfully.

From the SPC violations here, a lot level disposition would be expected. There appears to be no obvious trend, although ARF-3 does appear to have more WECO2 SPC fails. These violating lots may get some investigation and the one sample above the UCL may be reworked. As there looks to be no major excursion, no further investigation or action may transpire.
Level 0 Analysis of Variance Analysis (ANOVA) with Sentient and in JMP
Furthermore, ANOVA of Process and Product shows there to be 40.3% contribution to variation based on Process and nearly the same magnitude for Product, i.e., 38.5%. Other contribution may also be considered, like metrology tool or gauge. We will discuss this more during Level 3 Process Control.

Improving Process Capability at this stage happens through tool matching efforts to keep each Process Tool’s trend about the same specified target.

Process Tool level SPC provides an effective approach to maintain tool matching, and also highlights ‘out of family’ tools, based on SPC violation rate or Instability (IStab).
Level 1: Run-to-Run Process Control (R2R)
This SPC strategy is effective as a “passive monitor“ to trigger tool maintenance following product failure, however it and
SPC only strategy impacts manufacturing cost due to loss of tool availability, manpower and manufacturing capacity.
An alternative strategy to improve process capability while maintaining maximum Process Tool availability is to employ real-time process control and transition over to the next manufacturing strategy of Level 1 to achieve Run-to-Run Process Control (R2R).
To illustrate the benefits of implementing a R2R strategy in the space of lithography, we explore the scenario of a tools’ degrading alignment laser. The trend in FIGURE 8 represents production data for a single Process Tool, i.e., ARF-3. The positive trend, in this situation is an artifact of the tools’ alignment laser power degradation over time.

With standard SPC monitoring, “long term” process drifts may not be apparent. Typically, process drifts may be known or suspected, however operational processes for compensating are often based on action following excessive SPC fails.
Often, operations will wait until the impact of Process Tool drift results in specification failures. The recovery, with this process control strategy is to take the Process Tool down for Unscheduled Maintenance to perform tool calibrations that will bring the tool state back in line with the rest of the Process Tools. This downtime will lead to a significant impact on manufacturing cost due to loss of tool availability and manufacturing capacity. To avoid this, we will employ R2R process control as it is perfectly suited to situations where process drifts are present.
For our OL example here, we will consider the linear trend of OL Translation Error. Instead of just waiting and reacting to SPC failures, a R2R process is applied whereby an empirical process model learns and adapts to compensate the process settings based upon the Process Tool state in real time.

The Figure above shows the process capability go from a NOT Capable (Cpk << 1.0) to a fully Capable, better than a 6 sigma (Cpk > 2.0) process with existing equipment and this R2R process control strategy significantly improves our process capability with existing equipment and extends them to next node, i.e. the 2024 Node Specifications.
To further improve the effectiveness of the R2R process, we need to consider the measurement noise coming from the metrology tools and apply smart metrology filters and matching algorithms to safeguard the process adjustments against erroneous, noisy metrology and effectively match the measured response to remove gauge specific offsets.
This is next level of manufacturing process control comprises of Level 3 and can be further extended into Level 4 where we get into Feature-Engineering to generate models using machine learning.
Other examples in Lithography include printed feature sizes, or “Critical Dimension” (CD) where CD drift is expected due to effects of Exposer Energy loss over time. Drifts of incoming film thickness would also account for CD changes over time.
Summary
Application of basic SPC process monitoring provides a clear approach to detect special cause events and trigger action to reduce process excursion and better match tools and processes.
However, systematic trends inherent in volume manufacturing (e.g., several tools and processes) is not well managed using traditional SPC Violations to trigger corrective action, as this does not account well for process drifts that may be present.
R2R, as an extension to the process control strategy toolbox, provides an automated approach to following process drift and keeping processes on target. Practical settings and filters allow real world challenges like tool, product and measurement offsets and noise to be efficiently accounted for. Process capability after application of R2R process control shows significant improvement from not capable to better than a 6-sigma process. This improvement directly impacts manufacturing efficiency, allowing for more throughput and capability without the need for capital expenditures.
References:
https://irds.ieee.org/ and https://irds.ieee.org/images/files/pdf/2021/2021IRDS_ES.pdf
Learn More...
Sentient’s “Software-as-a-Solution” offers end-to-end Knowledge, Guidance and Direction tailored to your unique needs, through Consulting, Training and Project Support. Let us help you Build Smarter and discover Value.
Abeer Singhal
Abeer has over a decade of experience in semiconductor processing and advanced manufacturing. He is co-inventor on U.S. patents in the field of semiconductor manufacturing and has extensive experience implementing advanced manufacturing systems for high volume and complexity products. Prior to his role at Sentient, Abeer has worked as an equipment technician, process engineer, advanced process control systems engineer, and manufacturing systems architect. He has a degree in Microelectronic Engineering.