Intracranial electroencephalogram (IEEG) is an invasive procedure widely used for preoperative assessment of drug-resistant epilepsy (DRE). In recent years, methods based on machine learning achieved high accuracy in automatic EEG recognition. However, these models have also grown in complexity, requiring a large amount of time for various feature extraction or signal transformation, which makes it difficult to efficiently process the IEEG recordings that span days to weeks. In this study, we propose IEEG-TCN, a concise temporal convolutional network that performs nicely on different IEEG classification tasks while costing less time. According to experimental results on the Bern-Barcelona EEG dataset, our method reaches an impressive accuracy of 93.76% in detecting focal signals. The latency for processing each IEEG segment is reduced by 19 seconds. We also verify the robustness of IEEG-TCN based on the Multicenter IEEG dataset that contains four categories of IEEG segments (physiological activity, pathological activity, artifacts, and noise). The results show that the model can be successfully generalized to multi-class problems.