The rising demand for high-performance integrated circuits for artificial intelligence (AI) is pushing conventional silicon transistors to their physical limits, prompting the search for alternative semiconductor materials. Oxide semiconductors (OS), with wide band gap, ultralow leakage, high mobility, and low thermal budget, offer appealing advantages for logic, dynamic random-access memory (DRAM), and monolithic 3D (M3D) integration in the post-Moore era. While tremendous progress has been made in the past decade in this field, several key challenges in terms of materials, devices, and process integration remain to be addressed. In this work, we discuss the critical issues in mobility, threshold voltage, reliability, and scalability and evaluate different strategies such as composition tuning, defect and interface engineering, and doping to enhance the device performance. We further discuss the potential and limitations of the OS in their applications in logic, DRAM, and M3D integration. Looking forward, a holistic cooptimization across materials, devices, and processes, combined with AI computing architectures, is expected in order to realize reliable, low power, high-mobility integrated circuits for next-generation intelligent chips.
Wang et al. (Fri,) studied this question.