The globalization of integrated circuit (ICs) design and manufacturing processes has increased the risk of malicious modifications, commonly known as hardware Trojans (HTs), which can disrupt the normal functioning of ICs. Traditional logic testing methods often struggle to effectively detect these threats, a limitation that can be overcome through side-channel analysis techniques. Side-channel detection techniques capitalize on the subtle variations in bypass information, caused by the additional circuitry introduced by HTs, to identify abnormal activities. This paper introduces an innovative approach that leverages the capabilities of deep neural networks, specifically combining Long Short-Term Memory (LSTM) networks with Temporal Convolutional Networks (TCN), to extract features from multivariate side-channel signals for Trojan detection. LSTM demonstrates superior capability in modeling long-term dependencies, while TCN enhances training efficiency and ensures stable gradient propagation in signal processing tasks. This hybrid model improves the detection accuracy of abnormal features in circuits compromised by HTs, significantly enhancing the reliability of multivariate side-channel Trojan detection.