Back-End CMOS Compatible and Flexible Ferroelectric Memories for Neuromorphic Computing and Adaptive Sensing

Development of unconventional computing architectures, including neuromorphic computing, relies heavily on novel devices with properly engineered properties. This requires exploration of new functional materials and their designed interfaces. Ferroelectric memories including two-terminal ferroelectric tunnel junctions and three-terminal ferroelectric field-effect transistors have shown promising performances in recent years as analog, multibit memory components with ultralow power consumption. However, for ferroelectric memory technology to become a mainstream technology, CMOS integration of these components is of major importance. For further diversifying their application to edge computing and smart sensing industry, a vast unchartered territory of low-temperature processable and CMOS back-end-of-line (BEOL) compatible materials needs to be researched. In recent years, doped HfO2-based memory devices and in-memory computing architectures have gathered huge momentum as one of the beyond von Neumann computing alternatives. In comparison, molecular ferroelectric-based systems are still in their early exploratory phase. This review discusses the potential for doped HfO2 and molecular ferroelectrics as CMOS BEOL and flexible and wearable platform compatible neuromorphic devices and circuits and the challenges that need to be overcome for turning the opportunities to a technological reality.

Automated summary

Unconventional computing architectures, such as neuromorphic computing, rely on novel devices with properly engineered properties, including ferroelectric memories. To make ferroelectric memory technology mainstream, CMOS integration and further research into materials for edge computing are crucial. Doped HfO2 and molecular ferroelectrics show potential as CMOS BEOL compatible neuromorphic devices, but challenges remain in turning these opportunities into technological reality.