Primate cortex is a layered computational system characterized by different cell clusterings and connectivities. Their details can be simplified by implementing two-dimensional sheets of visuo-topically arranged model columns defined as excitatory-inhibitory (E-I) pairs of single-voltage compartments (point-like entities). Such E-I pairs are driven by feedforward input, lateral recurrent interaction and modulating feedback from contextual signals. Here, we develop a principled modular approach to map the elemental computational processes of excitatory and inhibitory conductance-based neural elements onto spiking neuromorphic hardware. We focus on the braininspired IBM TrueNorth Neurosynaptic System to realize a realtime canonical columnar cortical model circuit that is also energy-efficient. The coding precision of the model is demonstrated in several experiments with populations of such model units.