Fifth generation (5G) wireless networks require massive connectivity with a large number of devices and ask for non-orthogonal multiple access which has been proposed as a paradigm shift of physical layer technologies. Among all the existing non-orthogonal technologies where the number of users can be larger than the spreading factor, recently proposed sparse code multiple access (SCMA) is shown to achieve a better link performance. However, the message passing algorithm (MPA) is used in SCMA decoding. And the decoding complexity of SCMA is very high and is increasing exponentially as the dimensions of the codebook increase. From this point of view, implementing SCMA long codewords in a real hardware environment with the high complexity MPA is intractable. A scheme to reduce the decoding complexity of SCMA is essential. In this paper, we propose a parallel SCMA system which can be called SCMA hierarchical coding to come through SCMA long codewords. We try to break the process of the encoding and decoding of a long SCMA codeword into several processes of encoding and decoding of short codewords. We also analyze the tradeoff between the performance and the decoding complexity of the parallel SCMA system. The simulation results show that the parallel SCMA system can greatly reduce the decoding complexity of SCMA long codewords while still keeping higher capacity and lower block error rate compared to long term evolution (LTE) and low density signature (LDS) respectively.