Artificial intelligence (AI) has proven to be effective in solving complex problems in fields such as computer vision, natural language processing and medicine. However, in terms of energy efficiency, deep learning approaches are still outmatched by their biological counterparts. Yet, the underlying fundamental mechanisms of information storage, processing and computation of biological neuronal networks (BNNs) remain poorly understood.In bottom-up neuroscience, we tackle this challenge by analyzing low-density, engineered BNNs in vitro. Primary hippocampal rat neurons or human induced pluripotent stem cell (hiPSC) derived ngn2 neurons are seeded on top of high-density microelectrode arrays (HD-MEAs), where their growth is confined by polydimethylsiloxane (PDMS) microstructures. This approach allows us to record and stimulate BNNs to study the propagation of information on a subcellular level.In this work, we use PDMS microstructures with nanochannels which can generate isolated feed-forward circuits consisting of two input and one output node. Artificial inputs are encoded by stimulation patterns and the response of the middle node is captured with rate or temporal coding. By implementing different stimulation protocols, we successfully achieve three fundamental logic operations: NOT, AND, and OR. Furthermore, these operations are robust to stochasticity in the input and reproducible across different circuits. In addition, we compare stimulation-induced and spontaneous responses. Initial results show similarities not only in terms of spike timing but also in the principal components of the recorded signal shape. Looking ahead, this observation would allow interconnecting small logic gates to do more complex operations.