Brain function is substantially linked to the highly organized structure of neuronal networks. Emerging three-dimensional (3D) neuronal cell culture technologies attempt to mimic the complexity of brain circuits as in vitro microphysiological systems. Nevertheless, structures of in vitro assembled neuronal circuits often varies between samples and changes over time that makes it challenging to reliably record network functional output and link it to the network structure. Hence, engineering neuronal structures with pre-defined geometry and reproducible functional features are essential to model in vivo neuronal circuits in a robust way. Here, we engineered thin microchannel devices to assemble 2D and 3D modular networks. Microchannel devices were coupled with multi-electrode array (MEA) electrophysiology system to enable long-term electrophysiology recordings from microengineered circuits. Each network was composed of 64 micromodules which were connected through micron size channels to their adjacent modules. Microstructures physically confined neurons to the recording electrodes that considerably enhanced the electrophysiology readout efficiency. In addition, microstructures preserved modular network structure over weeks. Modular circuits within microfluidic devices showed consistent spatial patterns of activity over weeks, which was missing in the randomly formed circuits. Number of physical connections per module was shown to be influencing the measured activity and functional connectivity parameters, that represents the impact of network structure on its functional output. We show that microengineered 3D modular networks with a profound activity and higher number of functional connections recapitulate key functional features of developing cortex. Structurally and functionally stable 2D and 3D network mimic the modular architecture of brain circuits and offers a robust and reproducible in vitro microphysiolopgical system to serve basic and translational neuroscience research.