The NBER network of U.S. patents from 1963 to 1999 (Hall, Jaffe, Tratjenberg 2001, USPTO) is an example of a very large citation network (3774768 vertices and 16522438 arcs). Using islands algorithm (Zaveršnik, Batagelj, 2004) for the Search Path Count (SPC) weights (Hummon and Doreian 1989; Batagelj 2003) the most powerful theme in the entire network was determined. From this we selected a group of companies and categories that appeared and split the entire network into subnetworks according to selected companies and technological categories. We study the general trends and features of the subnetworks over the past thirty-seven years. We propose another approach for studying patents' network as a temporal network. Vertices from the same category in the same time slice are shrunk and then the obtained smaller networks over time are studied. By studying development patterns of the network over time we are trying to determine the general trends in the research and development for the selected companies and categories over the past three decades.