The synthesis of separation networks has posed computational challenges in process design. One reason is that the space of all possible configurations for a given multicomponent feed is combinatorially large. The search for the best configurations is often hindered by the fact that a large fraction of these configurations use more separation units than necessary, without compensating with better energy consumption. Prior formulations in the literature turn out to either be incomplete or to also include such configurations, with no easy way of removing them before the search begins. Secondly, energy consumption is generally a non-linear and non-convex function of the composition of intermediate streams in the separation, even if the discrete configuration itself remains unchanged. This would hinder the search by trapping the optimizer in local minima and by misreporting a potentially good configuration as being suboptimal. In this work, we present a novel framework to consider only those separation configurations that have the minimum number of separation units. We accomplish this by introducing the concept of a “supernetwork”, which considers only configurations with the minimum number of separation units. We provide evidence that additional configurations with more units than necessary do not provide better energy consumption for distillation networks, and are hence not necessary to consider when searching for the optimal configuration. We show how the new search space formulation allows for easy integration with external constraints generated by process or market forces. We also evaluate different optimization methods to search the configuration space and optimize the energy consumption with respect to composition and flow rates.