In this paper, disruptive research using modern embedding techniques and a deep learning (DL) model based on a relational graph convolutional network (R-GCN) encoder that automates the placement task of analog layout synthesis is conducted. The proposed methodology introduces structure in the input data, drastically reducing the total number of trainable parameters, leading to a smaller and more effective regression model. Moreover, its unsupervised training does not rely on expensive legacy layout data but only on sizing solutions. Experimental results show that the proposed R-GCN deep model generates placement solutions at push-button speed for multiple technology nodes and generalizes to circuit topologies not used in training. Moreover, the model outperforms other dense DL models while being 3000x smaller and producing solutions that compete with highly optimized analog designs.