InsightFab
Knowledge Base/Endpoint Detection and Uniformity Control in Etch Process (Etch)
Semiconductor Process6 min read

Endpoint Detection and Uniformity Control in Etch Process (Etch)

The author recounts a past experience dealing with an etch tool malfunction where OES waveforms exhibited erratic behavior, leading to significant yield degradation. This article delves into the critical role of precise endpoint detection and uniformity control, essential for achieving the exact desired etch depth in semiconductor manufacturing processes.

That Day, the OES Waveform Looked Drunk, and Everyone Was Stunned

I still remember it was many years ago. It was past eleven at night, and I was just about to leave work when my phone suddenly rang. An assistant engineer from the etch tool called, his voice a bit frantic: "Senior, for a batch of advanced process wafers, the OES waveform is suddenly erratic, nothing like a normal endpoint. It's stuck on the tool now, and we dare not move it!" I thought, "This is bad," and rushed back to the factory from home. When I arrived, sure enough, the OES (Optical Emission Spectroscopy) waveform on the screen was like a drunken mess, fluctuating up and down, completely missing that characteristic "inflection point." Wouldn't you know it, the yield for this batch dropped directly to 88% the very next day.

What Was the Problem? Etch Endpoint and Uniformity, in Essence, Are About "Control"

You might think, isn't etching just removing unwanted parts? What's so difficult? Honestly, the difficulty lies in etching "just right." One second too long, it's too deep; one second too short, it's not enough. This is the importance of "endpoint detection." When etch gases react with the wafer surface, they produce specific spectral signals. We can detect changes in the intensity of these signals using OES. When the target etch layer is completely penetrated and the underlying stop layer is exposed, the spectrum of the gas reacting with the stop layer will differ from before. This "moment of change" is the endpoint we are looking for.

However, just finding the endpoint is not enough; the etch must also be "uniform." A single wafer might have millions or even hundreds of millions of components etched at once. If the etch depth at the center differs from the edge by 50 angstroms, the yield will be completely compromised. This is what we call "uniformity" control.

How Is It Actually Done? Data, Data, and More Data

OES is probably the most commonly used method for endpoint detection. We typically set a "detection window," for example, near specific wavelengths like 480 nm or 750 nm, to monitor the rate of change in spectral intensity. When the rate of intensity change exceeds a preset threshold (e.g., changing from -0.05 to +0.02), or when the signal intensity reaches an absolute value (e.g., dropping from 1000 counts to 500 counts), we determine that the etch is complete.

Uniformity control is a bit more complex. We regularly sample and measure multiple points on the wafer (e.g., 5 points, 9 points, or even 49 points) to calculate the difference between center and edge points, or to calculate the standard deviation across the entire wafer. Ideally, your uniformity Cpk should be maintained above 1.33. If it drops to 1.08, the DPMO (Defects Per Million Opportunities) will surge to 6210, a number that would make any boss panic. Therefore, after every consumables replacement or chamber cleaning, new uniformity verification must be performed to ensure the etch profile is consistent with previous runs.

The Most Common Pitfalls: Consumables, Cleaning, and Parameter Drift

The most common issues I've encountered are consumables not being replaced when their lifespan is up, or being replaced without verification. Consumables inside the reaction chamber, such as quartz rings and electrode plates, will be eroded by plasma over time, changing their shape and leading to uneven plasma distribution, which directly ruins etch uniformity. Furthermore, incomplete cleaning of the reaction chamber, with residues accumulating on the chamber walls, can also affect plasma stability.

On one occasion, a new assistant engineer mistakenly adjusted a parameter in the cleaning process, shortening a cleaning step. As a result, the endpoint for two consecutive batches was premature by 5 seconds. However, because the OES waveform changes were not obvious, this was only caught when yield anomalies were discovered in subsequent processes. The cost of the wafers lost that time, if converted to money, could probably buy two imported cars. Therefore, any parameter change must undergo strict Change Control and be thoroughly verified.

One Thing You Can Do Today

Before leaving work each day, check the tool's OES waveform to ensure there are no abnormal trends.

Want to try it yourself?

Every tool mentioned in this article is available on InsightFab — just upload a CSV to analyze.

Go to Tools