Data-Driven Approach to Plasma Electron Temperature Estimation Using VI Probe Harmonics

This study presents a machine learning approach to predict plasma electron temperature (𝑇ₑ) using harmonic data from a VI-probe, typically used for RF power measurements. Experiments were conducted in an Inductively Coupled Plasma (ICP) reactor with argon gas, under RF power from 50 to 1000 W and pressure between 5 and 100 mTorr. Voltage, current, and phase of the 13.56 MHz RF signal were measured up to the 15th harmonic using a VI-probe, while 𝑇ₑ was measured simultaneously with a Langmuir probe. The VI-probe data served as input features, and 𝑇ₑ as the target variable, for training regression models including linear regression and random forest. Performance was evaluated using RMSE, MSE, and 𝑅². Feature importance analysis identified key contributors to prediction. The models achieved high accuracy with 𝑅² exceeding 0.9, though overfitting was observed due to limited training data. These findings demonstrate the potential of using VI-probe data for reliable, real-time, and automated plasma diagnostics, without the need for complex sensor systems