Data-driven modeling of etching processes allows for the accurate prediction of etch profiles significantly enhancing the efficiency of IC research. In this study, a convolutional Transformer based etching network (CTransE) is developed to extract multi-scale features from etching profiles, enabling accurate modeling and prediction under variable process conditions. Recognizing that etching profile evolution is a physically constrained, long-sequence task, the model exploits the standard Transformer architecture to effectively capture cross-timestep dependencies in physical image sequences through self-attention mechanism. We specifically employed the CTransE model for the etching of SiO2 using Ar/CF4 gas mixtures using a simulated dataset of 18,360 2D profiles and a few-shot experimental dataset. Compared to conventional RNN or LSTM, CTransE model reduces the mean squared error (MSE) by 93.15% and improves the structural similarity index measure (SSIM) by 29.02%, enabling accurate predictions of real-world experimental profiles. The inference time for a single profile is 0.13 s, significantly speedup the prediction efficiency of temporal etching processes. The proposed method delivers a fast, robust, and low-cost strategy for monitoring plasma etching in small-node IC fabrication.

Plasma etching is a critical step in semiconductor manufacturing, yet existing approaches are either computationally expensive or limited to predicting scalar etching metrics rather than full profile evolution. We propose EPreNet, a condition-guided spatio-temporal network for pixel-level prediction of plasma etching profile evolution from historical profile frames and process parameters. To support this task, we construct a benchmark dataset of 18,360 images spanning 918 process conditions simulated via TCAD, sampled with Latin Hypercube Sampling to ensure uniform parameter-space coverage, and further establish an evaluation framework combining image-level and geometry-based metrics for etching-profile prediction. Experiments demonstrate that EPreNet reduces MSE by 16% and achieves SSIM of 0.992 and PSNR of 30.323 dB, while achieving manufacturing-relevant geometric accuracy with $1.4°$ sidewall angle error and 1.6% depth error rate. Inference requires only 38.92 ms per frame faster than TCAD simulation 1300 s, supporting rapid surrogate-based evaluation and accelerated TCAD-assisted process exploration. The model also shows strong generalization to unseen initial critical dimensions and encouraging initial transferability to preprocessed experimental SEM images, suggesting its potential as an efficient surrogate for TCAD-assisted process development while maintaining high geometric fidelity.