Scenario
Your etching process has a problem: after each batch, the etching depth of the next batch deviates by a few nanometers. While no out-of-control signals appear on the SPC control charts, the Cpk is only 1.1, indicating that the process is slowly drifting.
SPC tells you "if there's a problem," but it cannot correct it automatically. You need APC.
What is SPC
SPC (Statistical Process Control)
- Monitors process output, detects abnormalities
- After a process run, uses control charts to determine if there are out-of-control signals
- Triggers engineer intervention to manually find the cause and make adjustments
- Essence: Post-event detection (Detect and React)
SPC's strengths: Universal, easy to understand, no complex models required
SPC's limitations: Can only react after a problem occurs, cannot prevent systematic drift
What is APC
APC (Advanced Process Control)
- Uses models to predict process output, automatically adjusts parameters for correction
- Requires no manual intervention, the system corrects automatically
- Essence: Predictive adjustment (Predict and Adjust)
Two types of APC:
- Feed-Forward: Observes variations in input material and adjusts parameters before running the process
- Feedback: Observes the output results of the previous batch and adjusts parameters for the next batch
R2R Control: The Most Commonly Used APC in Semiconductor Manufacturing
R2R (Run-to-Run) Control
After each run, measurement results are sent back to the control system, which uses EWMA (Exponentially Weighted Moving Average) to calculate the optimal parameter settings for the next run.
EWMA Controller Formula:
New Setting = Old Setting + α × (Target Value - Actual Result)
α is the learning rate (0 to 1):
- α close to 1: Fast response, but prone to over-correction
- α close to 0: Slow response, stable but slow to react to drift
- Typically set α = 0.4-0.6
Complementary Relationship between APC and SPC
| Feature | SPC | APC |
|---|---|---|
| Main Purpose | Detect abnormal signals | Automatically correct systematic drift |
| Reaction Method | Manual intervention | Automatic adjustment |
| Model Requirement | Not required | Requires process model |
| Applicable Scenario | Random abnormalities | Systematic drift |
Even after implementing APC, SPC remains essential:
- SPC is used to monitor whether APC is functioning correctly
- If the APC model fails, out-of-control signals will appear on the SPC control charts
- SPC serves as APC's safety net
Golden Quote
"SPC and APC are not in competition; they are like a goalkeeper and a defensive line. APC is responsible for keeping the process centered, while SPC is responsible for detecting when APC has failed. Both are necessary; a gap will exist if one is missing."