Today’s most advanced semiconductor fabs rely on sophisticated monitoring techniques to ensure wafers result in high yields for memory and logic devices. These methods evolved from periodic sampling of partially processed wafers within the production sequence and blank silicon or “short-loop” test wafers to check for particles or defects generated from a specific process tool.
Some sampling can be both costly and destructive, e.g. scanning-electron-micrograph (SEM) cross-sectioning to verify process quality and non-destructive methods such as optical microscopy to check for defects. The cost can be simply the price of the inspection tool and the added time in the manufacturing sequence that impacts productivity, or the loss of some output due to destructive sampling.
To catch production problems and improve yield, a fab must make many tradeoffs such as with the sampling interval (how many places in the production sequence that a wafer is sampled) and sampling rate (how often in that measurement step a wafer is measured) to determine the most cost-effective system.
Detecting the problems
Responding to a fab’s need to make these tradeoffs, tool suppliers are outfitting their process equipment with sensors and monitors that reduce the interval and rate of sampling and measuring wafers. These in-line monitors are effectively surrogates for on-wafer measurements.
With improved understanding of a tool’s sensor information and the resultant effects on the wafer, the surrogate provides the quality assurance needed for the respective step in the process.
Leading-edge fabs leveraged this process-tool information to build a fab-wide fault-detection and classification (FDC) system that gathers real-time equipment data. It’s turned into actionable information through process troubleshooting and controls. Poor quality is detected, and the system automatically stops processing wafers.
This methodology augments the in-line wafer sampling, potentially reducing the sample rate and intervals based on the strength of the correlation between tool data and wafer performance. As the fab knowledge base further develops, lower cost and higher yield can be attained.
Following this trend, photolithography tool suppliers are providing tool monitoring through sophisticated sensors and software algorithms that compensate for suboptimal performance to achieve more predictable results. Real-time information on exposure dose and focus are used for feed-forward and feedback loop systems to improve critical-dimension (CD) uniformity and thus high yield.
Characterizing the light source
Since the light source for photolithography plays a critical role in the wafer’s quality, Cymer introduced a new data and information product known as SmartPulse™. Excimer-laser light-source data is packaged to match a wafer exposure sequence, enabling users to link the laser parametric data to a specific wafer.
Conventional excimer-laser sources reported data for energy, wavelength, and bandwidth, but this information was not filtered for wafer-exposure events (vs. calibration-related events) that the lithography tool performs with every wafer. As a result, the controls (limits and alarms) around this data accommodated all of these events, even if the process requirements for wafer exposure were more stringent.
By filtering for exposure-only data, the information can be directly actionable in real time and usable by the fab-wide FDC system to further detect faults and prevent excursions.
While energy, wavelength, and bandwidth have been key light-source performance indicators, other parameters, including beam pointing, divergence, and polarization, come into play in the lithography tool’s performance.
The latest-generation argon fluoride (ArF) (193nm) immersion scanners include freeform or programmable illumination schemes that are increasingly sensitive to the light source beam parameters. Time-consuming calibration routines are required whenever the light source is perturbed, such as during a maintenance event. In developing a solution that provides a comprehensive characterization of the light source, new metrology captured and packaged these additional parameters on a wafer level.
SmartPulse includes this comprehensive data stream that fully characterizes the light source. The more extensive coverage in the lithography tool set reduces dependence on wafer monitoring and minimizes the risk of excursions (Fig. 1).
Figure 1: Light-source data for multiple tools, allowing performance comparisons (top). Drill-down information for one tool (bottom).
Taking corrective action
As ArF immersion lithography processes for critical-layer patterning have advanced through multiple technology generations, each lithographic imaging solution has become highly optimized for specific patterns. Imaging solutions for distinct device patterns drive different control levels for process variables.
For example, highly optimized source-mask-optimization (SMO) imaging solutions require tighter control of the illumination pupil than simple spacer double-patterning (SDP) with dipole illumination. Very high throughput lithography patterning processes were implemented to reduce the cost of multiple patterning steps, which are commonly used for advanced device production. Process parameter monitoring and control for the process tools reduce errors and improve control while minimizing metrology capital costs.
In the case of SmartPulse, the additional metrology for beam pointing, divergence, and polarization establishes correlations to the scanner illuminator downstream, highlighting the importance of monitoring these additional parameters. For example, beam pointing and divergence changes were quantified following laser service, and the impact on the scanner illuminator pupil uniformity was captured (Fig. 2).
Figure 2: Two examples of illumination pupil change after laser service events led to measurable beam changes using the light source on-board metrology. (Difference map after laser service shown over green background.)
The absence of this level and granularity of data process troubleshooting in the litho cell requires significant time to identify sources of deviation for corrective action. For example, loss of energy in the system could arise from optics degradation in the exposure lens, illuminator, beam-delivery unit, or light source. Corrective action requires physical inspection of the various components.
With real-time monitoring of the light source, the problem can be quickly narrowed down before a service action. Additionally, integrating this comprehensive real-time characterization of the light source into the fab FDC infrastructure improves process control. Excursion prevention helps accelerate the factory ramping to high volume with lower cost while reaching the yield goals.
–Theodore Cacouris is OnPulse product marketing director at Cymer and regularly participates in SPIE Advanced Lithography. He has doctorate, master’s, and bachelor’s degrees in electrical engineering from Columbia University (USA).
Have a question or comment about this article? Write to us at email@example.com.
To receive a print copy of SPIE Professional, the SPIE member magazine, become an SPIE member.