Are EPD and OES
Enough?

Finding the endpoint and keeping final quality consistent are two different problems.

CMP and etch are two semiconductor processes where timing matters. In CMP, EPD is used to identify when the target film has been removed or the underlying layer has begun to appear. In plasma etch, OES tracks changes in the plasma emission spectrum that can indicate a transition between the layer being etched and the layer beneath it.

These technologies provide a major improvement over running every wafer for a fixed amount of time. They allow the process to respond to what is actually happening inside the tool and determine when to stop or move to the next step.

But even a well-detected endpoint does not guarantee the same final result on every wafer. In production, important processing still takes place after endpoint detection.

CMP EPD identifies endpoint from a change in friction, acoustic, or optical signal; etch OES identifies it from a change in the emission spectrum. In both processes an over-polish or over-etch step follows the endpoint, and material can still remain on the wafer at the moment the endpoint is detected.

CMP continues after the endpoint

CMP endpoint detection typically identifies a signal change associated with the target film clearing or the underlying layer becoming exposed. The entire wafer, however, does not reach that state at exactly the same time. Within a single wafer, the following can vary by location:

• Film thickness and pattern density
• Pressure distribution
• Slurry delivery
• Pad contact condition

At the moment the endpoint is detected, some areas may already be clear while others still have material remaining.

For this reason, CMP recipes often include an over-polish step after endpoint detection. This additional polishing time helps clear the remaining material across the wafer.

• Too little over-polish: residual film or partially cleared areas remain.
• Too much over-polish: dishing, erosion, and loss of the underlying layer increase.

Final quality therefore depends not only on detecting endpoint accurately, but also on what happens after it.

Over-etch matters for the same reason

In plasma etch, OES measures changes in the wavelengths and intensity of light emitted by the plasma. As the material being etched and the reaction products change, the emission spectrum changes as well. These changes can be used to identify when the target layer is clearing and the underlying layer is beginning to appear.

That endpoint may represent the first part of the wafer reaching clear, rather than every feature being fully etched. Etch rate can vary with:

• Wafer position
• Pattern size and density
• Loading
• Chamber condition

Some features may still contain material even after an endpoint signal has been detected.

Many etch recipes therefore include an over-etch (or finish-etch) step after the main etch. Its job is to remove residual material and stringers while keeping underlying-layer loss, CD shift, and profile damage within an acceptable range.

• Too little over-etch: incomplete features or residues remain.
• Too much over-etch: the underlying layer or surrounding structures can be damaged.

Here too, endpoint is not the final result — it marks the beginning of another quality-critical part of the process.

Fixed over-polish and over-etch have limits

The simplest approach is to run the same over-polish or over-etch time after every endpoint. This can work when the process is stable and has enough margin.

Production conditions rarely stay completely fixed, though. Several things change continuously:

• Incoming film thickness and pattern mix
• CMP pad and slurry aging
• Etch chamber and focus ring condition
• Removal rate/etch rate shifts from chamber cleaning, maintenance, and consumable replacement

A change in time-to-endpoint can be a sign that the incoming wafer or process rate has changed. If the post-endpoint step always uses the same time and settings, those changes can show up as variation in final quality.

Improving EPD or OES accuracy alone does not solve this. Endpoint detection tells the tool when a transition has occurred in the current process. It does not automatically adjust the next wafer so that the final result remains on target.

APC controls what happens after endpoint

APC (Advanced Process Control) uses process data and quality results to adjust the conditions applied to the next run. In CMP, this may include polishing time or other recipe settings. In etch, it may include over-etch time or selected recipe setpoints.

• If metrology shows residual material after processing → the next run is adjusted to provide more clearing.
• If the result shows excessive removal or underlying-layer loss → the post-endpoint step is reduced.

The controller can also take into account equipment condition, consumable age, recent endpoint behavior, and incoming wafer information — giving the process a way to respond to change rather than relying on one fixed post-endpoint setting.

Those adjustments still need guardrails:

• Limits on recipe changes
• Maximum processing time
• Fallback recipes
• Approval rules

EPD and OES do not compete with APC. They serve different roles: EPD and OES detect changes during the current run, while APC uses those signals together with process and metrology data to keep results consistent across runs.

Endpoint is not the end of process control

CMP EPD and etch OES are already valuable forms of process automation. They reflect the actual process state more closely than fixed-time operation and can reduce both under-processing and unnecessary over-processing.

Production control, however, requires more than finding endpoint repeatedly. It also requires managing over-polish and over-etch so that final quality remains consistent from wafer to wafer.