The use of mushroom cell PCM as an NVM memristive weight element for analog computing has been a focus of recent progress towards hardware designed for artificial intelligence. A major detriment of standard Ge-Sb-Te based materials (e.g. GST225) typically used in binary memories, is its relatively low resistance and a low crystallization temperature that leads to high power consumption when implemented in an array such as a neural network, and poor resistance-state retention, respectively. By doping PCM with insoluble non-conducting material to form a homogenous mixture (dPCM), the PCM grain sizes are confined, increasing the crystallization temperature, improving retention through increased crystallization temperature. The resistance of the dPCM is also significantly increased, decreasing programming current requirements. There is, however, a limit to the level of doping that can be used for dPCM as high amounts of dopant can create an inability for the device to SET to its crystalline state. In this work, we show how insertion of a thin, undoped or low-doped interfacial PCM layer between the confined bottom electrode heater (BEH) and the bulk dPCM allows for the use of higher doping and more resistive dPCMs to further reduce programming current requirements.