Layer-By-Layer Phase Change in an In₂Se₃-Based Neuromorphic Device

Indium selenide (In₂Se₃) undergoes multiple phase changes between crystalline phases including the layered ferroelectric (alpha), paraelectric (beta) and non-layered (gamma) phase through multiple stimuli including strain, thermal excitations, and electrial excitations. Thus, In₂Se₃ is a promising candidate for a new class of phase change memory (PCM) utilizing structural changes between two or more crystalline phases rather than relying on transitions between amorphous and crystalline phases seen in conventional phase change memories. Suitable for storage class memory or neuromorphic computing applications, PCM utilizing changes between crystalline phases are expected to have faster and lower energy write operations compared to conventional PCM due to the lower entropy of the crystalline-crystalline phase change as well as addressing issues of resistance drift stemming by removing use of the amorphous phase. In this work, we present a multi-level phase change memory (PCM) based on In₂Se₃, demonstrating device resistance spanning six orders of magnitude. Local transport measurements using scanning tunneling microscopy (STM) reveal that the resistance of vertical devices changes exponentially with the number of switching steps. These findings suggest a layer-by-layer phase change within the material's structure, rather than the conventional radial expansion of low and high resistance phases seen in typical PCM devices. In contrast to planar PCM structures, which exhibit a three-orders of magnitude resistance change, the vertical structure facilitates tunneling-like transport through the active material, resulting in the observed six-orders of magnitude resistance variation. We find van-der-Waals gaps to be the limiting mechanism for this phenomenon. The structural changes between crystalline alpha-In₂Se₃ and crystalline beta-In₂Se₃ phases of In₂Se₃ provide a low-entropy switching mechanism in conjunction with multi-level switching can allow for high density data storage and neuromorphic applications.