Graphs and matrices are widely used to represent maps in solutions for Simultaneous Localization and Mapping (SLAM). Space representation is crucial for map creation, finding the shortest path, planning a trajectory or route, tracking landmarks, etc. The choice of the data structure can present disadvantages such as more memory consumption when using matrices, or, the need of a transforming function to represent a map when graphs (as structure) are applied as structure. This paper propose a representing map using a storing structure that allows dynamic growth and optimization of memory consumption in SLAM. The so-called MPTE-SLAM structure, proposed here, suggests the usage of a sparse matrix of occupancy matrices to represent a map for SLAM. The goal of this approach is to reduce the use of memory (if compared with traditional metric maps also) also having a compact representation suitable for path planning and other tasks for autonomous robots; without the need of a transforming function (if compared with graph based solutions that return a map). This approach allows the addition and updating of values of the sub-matrices quickly and with dynamic growth on the overall map within the structure in the same way graphs permit; also it is suitable for a real-time implementation of solutions for SLAM. For implementation and testing, a non-holonomic mobile robot in an indoor environment was used. The final results showed that the MPTE-SLAM structure uses less memory than metric maps when optimal sub-matrices sizes are applied. Through experimentation an optimal size for submatrices (10x10) was determined; this size yielded a memory consumption less than 4.3 GB in all test experiments; also, it was observed that this approach also has dynamic growth in the structure plus the no-loss of its metric nature which gives it an advantage over graph-based representations.