Enabling reliable two-terminal memristor network by exploiting the dynamic reverse recovery in a diode selector

The transistor is broadly used to address memristor networks, but its three-terminal structure can impose limitations on fully exploiting the potential of efficient integration that a two-terminal memristor can offer. While a two-terminal selector is desirable for unlocking this potential, no existing device has attained a similar level of functional maturity. The diode, despite its technological maturity, is still limited by its unipolarity in addressing mainstream bipolar memristors. Here, we demonstrate that a diode can be implemented as a bidirectional selector for constructing two-terminal memristor architecture by exploiting its reverse recovery dynamics. This is demonstrated by the construction of one-diode-one-memristor (1D1R) programmable arrays, which are implemented for in situ neural training and classification. Furthermore, a crossbar array made from stacking 1D1R cells is fabricated to demonstrate scalable integration. This dynamic paradigm combines the advantages of functional maturity and structural simplicity of diode selectors to improve the development of memristor integration.