Semiconductor Yield Improvement with High-Quality FIB SEM Sample Prep

Cutting-edge semiconductor devices aren’t just getting smaller than their predecessors, they’re also becoming more complex. Standard analysis and characterization are no longer sufficient to drive development and production. Instead, manufacturers are increasingly turning to high-resolution transmission electron microscopy (TEM) to support their semiconductor yield improvement efforts.

However, TEM requires focused ion beam (FIB) sample preparation that extracts portions of the wafer (lamella) for analysis, a destructive process which turns the wafer into scrap. Scrap is an expensive reality for a fab, costing manufacturers thousands of dollars.

The value of TEM analysis can offset these costs by driving process improvements to enable faster volume production ramp-up and semiconductor yield improvement.

TEM samples for semiconductor yield improvement

Finished TEM samples awaiting extraction on a film-stack wafer

3 Sample Prep Requirements for Semiconductor Yield Improvement

To further maximize the value of this workflow, it is advantageous that labs extract as much high-quality information as possible from each wafer and each lamella. This means that not only is throughput a critical factor in the TEM preparation efforts, but that each lamella meets the quality requirements to ensure successful TEM analysis at the end of the workflow.

With that in mind, there are several requirements for high-quality TEM sample preparation of silicon wafers:

Accurate end pointing: TEM characterization is only successful if the area of interest is correctly located and extracted in the lamella. With shrinking dimensions and increasing complexity, capturing the target device or feature within that lamella means hitting an ever-smaller target–a process that requires high precision operator skills. To support the demand for TEM analysis, further automation and ease-of-use improvements need to be utilized to minimize the painstaking preparation steps that are performed by skilled operators today.
Optimized sample thickness: Nanoscale devices require increasingly thin lamella to capture only a single feature or device of interest. For some critical analyses, convolution of multiple features in the TEM image makes the metrology challenging, if not impossible. On top of that, EELS is needed for advanced materials and lighter elemental analysis – a TEM technique that requires very thin lamella (30nm thick or less) which pushes current TEM preparation tool capability, repeatability, and operator skills to their limits. For these types of measurements, leveraging automation takes on greater importance to support the increasing need for these analytical results and semiconductor yield improvement.
High surface quality: Curtaining and surface damage during lamella preparation can significantly hamper analysis. Sample extraction must balance speed vs. surface quality to quickly produce the best possible samples with minimal damage. Final finishing can be by its nature, a slow process, but is unavoidable to achieve the necessary end-result quality. As lamella get thinner and features get smaller, there is less leeway in damage layer thickness. For instance, a 10-nm-thick lamella must have a damage layer less than 2 nm-thick for accurate TEM analysis. Minimizing the surface damage requires precise FIB-SEM control at low accelerating voltages.

FIB-SEM Sample Prep with Automation and Machine Learning

Luckily, modern FIB-SEM sample-preparation technology is actively addressing these problems through the integration of automation and machine learning.

Not only can automation assist in maintaining the instrument by automating cumbersome and time-consuming alignment, but it can also optimize system utilization by enabling unattended performance of routine tasks, minimizing the number of touchpoints needed by the operator. Some advanced systems can even automate sample polishing, ensuring that the surface is ideally suited for TEM analysis.

Overall, these innovations allow more samples to be prepared per operator, increasing productivity, and enhancing manufacturers’ yield learning processes, therefore enabling semiconductor yield improvement.

Learn more about our new Helios 5 EXL Wafer DualBeam, which offers automation in TEM sample prep to help accelerate time-to-yield.

Sean Zumwalt is a Senior Marketing Manager for semiconductors at Thermo Fisher Scientific. This blog was written in collaboration with Alex Ilitchev, Science Writer at Thermo Fisher Scientific.