While machine learning is rapidly revolutionizing countless disciplines, training neural networks requires large memory and consumes enormous amounts of power. Thus, fundamentally new computing paradigms are needed to greatly improve the energy efficiency of computing. Electrochemical random-access memory (ECRAM) is an emerging technology for energy-efficient neuromorphic computing. ECRAM devices allow for low-energy, analog tuning of a channel resistance by electrochemically controlled ion insertion (extraction) into (from) the channel material, which is typically a semiconducting metal oxide. To preserve electroneutrality, cation insertion usually leads to changing electronic charge carrier concentrations in the bulk, and it is generally assumed that this change is at the origin of the resistance modulation in ECRAM devices. However, there is evidence in the literature that the resistance of polycrystalline WO₃, a common channel material for ECRAM devices, can be dominated not by the bulk resistivity, but by grain boundary resistance^[1], which is associated with space charge effects resulting in the depletion of electron carriers in the vicinity of the interface. We use electrochemical impedance spectroscopy (EIS) combined with careful choice of source/drain electrode geometry to distinguish between bulk and space charge resistances such as those arising at grain boundaries or electrode/WO₃ interfaces in WO₃ channel films. EIS of the channel in a Mg-ECRAM device during programming reveals that in the low ion concentration regime, modulation of space charge resistance dominates, while for higher ion concentration levels, bulk doping is the more prominent effect. We will discuss possible mechanisms for this modulation of space charge resistance, including changing charge carrier concentrations in the space charge region upon ion intercalation and direct reduction of the interfacial core charge density. This novel understanding of additional resistance modulation mechanisms in Mg-ECRAM opens new pathways for exploration, such as the role, if any, that space charge modulation plays when H⁺ is used as the working ion. These results also call for an informed design of channel microstructure and geometry in ECRAM devices, which is required to aid in meeting the demanding speed and energy efficiency requirements for neuromorphic hardware.