Monolithic 3D integration as a pathway to energy-efficient computing and beyond: From materials and devices to architectures and chips

Abstract

As emerging technologies like artificial intelligence (AI) and big data continue to evolve, the demand for high-performance computing (HPC) has been increasing, driving the development of computing chips towards greater energy efficiency and multifunctionality. Monolithic 3D integration (M3D) is poised to be a key enabling technology, by vertically stacking multiple functional layers made of backend-of-the-line (BEOL)-compatible devices on top of Si circuits and interconnecting them with high-density interlayer vias (ILVs). Currently, contenders for functional materials and devices in M3D include carbon nanotubes, two-dimensional (2D) materials, oxide semiconductors and a variety of emerging memories, such as resistive random-access memory (RRAM). This article first discusses the key properties and latest research developments of those materials and their device applications. As a representative example, we then review the recent progress on RRAM-based M3D architectures that integrate memory, computing, and other functional elements to facilitate computing-in-memory (CIM). Finally, we further discuss the opportunities and challenges of M3D as a promising pathway to energy-efficient computing.