We use a "top-down" approach to investigate the prospects for quantum information and quantum simulations in large scale systems consisting of up to 100000 qubits. These qubits are represented by single atoms stored on individual lattice sites of an optical lattice. Such a quantum register of the atoms is formed by employing the Mott insulator transition from an atomic Bose-Einstein condensate. We plan to investigate the initialization of such a quantum register and will try to further enhance present experimental techniques to detect and minimize defects of such a quantum register. We will use controlled coherent collisions, as first demonstrated by our group, to create 1D and 2D cluster states and investigate their lifetime and decoherence properties. Furthermore, we are planning to investigate novel, more robust entanglement schemes for ultracold atoms in optical lattices, which are based on coherent collisional spin changing collisions in lattices, only recently demonstrated by our group. Based on these gates, we will try to enhance and study the fidelity of individual collisional quantum gates acting between two atoms only and then apply these gate operations to larger 1D and 2D arrays of trapped atoms.