Etching ML Predictor

Neural network-based process model for predicting etch rate, selectivity, and CD. 7-input → 20-node → 10-node → 3-output architecture trained on historical process data. Real-time predictions with confidence intervals, feature importance analysis, and interactive response surface exploration.

Neural Network Architecture

Input Layer: 7 features (Pressure, Power, CF₄ Flow, O₂ Flow, Bias, Temperature, Time)
Hidden Layers: 20 nodes → 10 nodes (ReLU activation)
Output Layer: 3 predictions (Etch Rate, Selectivity, CD)

Model Accuracy

0.97

Training Samples

500

RMSE

3.2

MAE

2.1

Input Parameters

Adjust process parameters to get real-time predictions for etch rate, selectivity, and CD.

Chamber Pressure (mTorr)

20 mTorr

RF Power (W)

1000 W

CF₄ Flow (sccm)

100 sccm

O₂ Flow (sccm)

20 sccm

Bias Voltage (V)

100 V

Temperature (°C)

25 °C

Etch Time (sec)

120 sec

Calculating predictions...

Neural Network Architecture Visualization

Training Loss History

Predicted vs Actual (Validation Set)

Residuals Analysis

Feature Importance Analysis

3D Response Surface (Power vs Pressure)

Sensitivity Analysis

Real-time Predictions

Prediction Ready! Neural network model has analyzed your inputs.

Predicted Etch Rate

145.3 nm/min

Confidence: ±4.2 nm/min (95% CI)

Predicted Selectivity

18.7:1

Confidence: ±1.3 (95% CI)

Predicted CD

352.1 nm

Confidence: ±3.8 nm (95% CI)