Automated metrology for high-volume semiconductor manufacturing
On June 27, 2023, Thermo Fisher Scientific announced a new generation scanning transmission electron microscope—a fully automated (S)TEM metrology solution for high-volume semiconductor manufacturing.
The new Metrios 6 (S)TEM is built to increase productivity and accelerate transmission electron microscopy (TEM) data acquisition. This solution includes newly designed hardware, artificial intelligence capabilities and machine-learning-based algorithms to enable automated metrology and obtain large-volume high-quality data from complex devices and novel materials fast.
As artificial intelligence (AI) and machine learning continue to gain interest in the news, most notably ChatGPT and other AI applications, what better time to share insights into how AI and machine learning are improving tool and technical resource productivity in (S)TEMs, scanning electron microscopes (SEM), focused ion beam SEMs (FIB-SEM), and TEMs. Before we jump into this, let’s start with a brief overview of artificial intelligence and machine learning.
Artificial intelligence and machine learning applications in semiconductor manufacturing
Over the last five years, interest in deploying AI and machine learning applications in the semiconductor industry grew rapidly. With processes and tools that generate petabytes of data, AI, with its ability to mine and utilize data, offers great potential to automate labor intensive and repetitive tasks, increase productivity and optimize human and tool resources.
At a high level, AI is not a single technology. Rather, it is a collection of technologies that together allows machines to mimic human intelligence. AI systems incorporate four capabilities: 1) the ability to sense with cameras or sensors; 2) the ability to comprehend by extracting information, detecting patterns and recognizing context; 3) the ability to act and 4) the ability to learn. Of the four capabilities, learning is the most associated with AI.
While many consider AI and machine learning synonymous, the two are slightly different concepts. Machine learning is a subfield of AI and refers to automating learning. For an AI system, machine learning allows it to sense, understand, assign significance, and modify behavior in an iterative process based on past results against specified parameters to improve performance. Machine learning applications can be descriptive, predictive or prescriptive. A third area to be aware of is deep learning. Deep learning is a subset of machine learning inspired by biological neural networks. Deep learning uses artificial neural networks, which mimic biological neural networks in the human brain to process information, find connections between the data, and make intelligent decisions or take actions on their own.
Artificial intelligence and machine learning for automated microscopy
To a certain degree, the AI capabilities of today’s electron microscopes are in their infancy and mainly found in the two areas: system calibration and process automation. A less developed category is data analytics. To provide examples of some AI applications for the semiconductor industry, below is a brief description of some of the AI-enabled capabilities in Thermo Fisher’s (S)TEMs, FIB-SEMs and SEMs.
Electron microscope calibration
System calibration is primarily about keeping the electron microscope in a working state and optimizing its performance. Within system calibration, four of the most widely known applications are tool alignment, predictive maintenance and monitoring, fleet management and image optimization. Examples of these applications follow.
With tool alignment, the electron microscope utilizes computer vision and advanced algorithms to align the column and the beam. AI tracks the column’s alignment state and compares it to the stability window to keep the tool aligned and operating at specification. This ensures high quality data capture and prevents losses in productivity due to runtime errors or excursions found after data has already been collected.
Predictive maintenance and monitoring are enabled by using AI and collected sensor data to automate the identification of potential issues that may impact tool operation. Predictive maintenance and monitoring provide the ability to avoid unplanned downtime, proactively schedule maintenance as required, or be notified to intervene when a sudden failure is imminent.
An additional example of an application in this category is image optimization. For the semiconductor industry, data cleaning or denoising can be vital to producing reproducible and statistically meaningful quantitative analysis of (S)TEM data. In the example below, a machine learning network was trained on the structures to reduce the signal-to-noise ratio (SNR) and improve SEM image quality and acquisition speed. The images on the right are the denoised images.
3D NAND channel image without (left) and with (right) SNR image optimizations.
Automated electron microscopy for process automation
The goal of process automation applications is to automate sample preparation, data acquisition and semiconductor metrology tasks to increase the productivity of workforce resources. Three examples of applications that provide process automation are end-pointing, automated recipe workflows and region of interest (ROI) navigation.
End-pointing detection utilizes machine learning, sensor and metrology measurements to stop milling when the metal or via layer of interest is exposed. When a specific sensor measurement, feature or threshold is seen, the etch tool is instructed to halt the etch operation.
With automated recipe workflows, “recipes,” or scripts are written and utilized to eliminate repetitive tasks. Machine learning is a recipe component enabling the recipe to adapt to local data.
Artificial intelligence enabled navigation automates TEM analysis and image acquisition of a gate-all-around device.
The final example in this category, ROI navigation, allows detection of specific features to automatically navigate to a ROI. Through this capability, users can improve cut placement, define image acquisition areas, and enhance the quality of the end data.
New generation scanning transmission electron microscope for automated metrology
Many of the artificial intelligence capabilities discussed above are available in Thermo Fisher’ (S)TEMs, SEMs, FIB-SEMs, and TEMs. With the new Metrios 6 (S)TEM, artificial intelligence and machine learning take automation and productivity to a new level. When applied to typical workflows, and compared to previous generations, the Metrios 6 (S)TEM delivers productivity gains of approximately 20% on average.
At Thermo Fisher, we’re very excited about artificial intelligence and machine learning and the capabilities it offers the semiconductor industry. We’re also excited about our next generation Metrios 6 (S)TEM and its ability to deliver actionable, automated metrology data to help semiconductor manufacturers increase productivity, optimize resource utilization, and accelerate development cycles and time-to-yield.
Learn more about the Thermo Scientific Metrios 6 (S)TEM >>
//
Semiconductor Nanoprobing Boosts TEM Analysis Success Rates on Advanced Logic Devices
Growing need for semiconductor nanoprobing As the semiconduc...
Read More
Accelerating Semiconductor Device Analysis with the Power of the Helios 5 EXL Wafer DualBeam
Semiconductor fabrication challenges Semiconductor fabricati...
Read More
Failure analysis of wide bandgap semiconductor devices
Compound semiconductors create analytical challenges Our eve...
Read More
Enhancing High-Volume DRAM Device Manufacturing with TEM Metrology and Characterization
Dynamic random access memory (DRAM) plays a crucial role in ...
Read More
Leave a Reply