Monolithic 3D integration of two-dimensional materials from processing to AI hardware

Abstract

The rapid growth of artificial intelligence is reshaping the requirements for future electronic hardware. As conventional CMOS scaling approaches fundamental limits in power density, heat dissipation, short-channel effects, and interconnect delay, system-level innovation is becoming as important as individual device scaling. Monolithic three-dimensional integration (M3D) offers a promising route to overcome these limitations by vertically integrating memory, logic, sensing, and neuromorphic functions with dense inter-tier interconnects. In this context, two-dimensional (2D) materials provide a uniquely attractive platform because of their atomic thickness, van der Waals interfaces, low-temperature process compatibility, and broad functional diversity. In this Research Highlight, we discuss the emerging transition from materials processing to system-level M3D AI hardware. We first introduce why 2D materials are promising building blocks for M3D integration, not only as field-effect-transistor channels but also as memristive, sensing, and neuromorphic layers. We then summarize three complementary integration strategies—direct growth, transfer-based integration, and solution-processed deposition—and discuss how each route must be selected according to device function, material quality, and manufacturability requirements. We further highlight hybrid CMOS–2D integration as a realistic near-term pathway, where mature CMOS provides peripheral logic and control circuitry while 2D materials add compact memory, sensing, and in-memory computing functions in the back end-of-line. Finally, we discuss all-2D monolithic integration as a longer-term vision for ultrathin, multifunctional AI systems.