GPU-Accelerated Quantum Transport Solver to Explore 2D Material Space for Transistor Operation

Abstract

For decades, silicon has been the mainstay of the semiconductor industry. However, to preserve the electrostatic integrity of MOSFETs (metal oxide semiconductor field effect transistors), technology downscaling necessitates that the thickness of the silicon channel be reduced to a few nanometres. Two-dimensional (2D) semiconductors with few atomic layers thickness have thus appeared as a natural replacement for bulk silicon. So far, the present candidates (transition metal dichalcogenides, black phosphorus, etc.) have not been able to outperform silicon transistors, and thus there is a quest for the ideal material. However, the 2D material space is infinite, and experimental efforts can only explore a minute fraction of it. Integrating new materials into existing process technologies is also time and capital-intensive affair in the semiconductor industry. First-principles-based device models, which enable systematic evaluation of new materials at the device and circuit level at the very early stage of technology development, thus are in great demand. Drift-diffusion model-based computer aided design tools have long been used to predict device properties prior to fabrication. Though computationally scalable, this semiclassical approach fails to capture the wave-like nature of electrons, which become dominant as channel length approaches the decananometer regime. A self-consistent solution to the single-electron Schrödinger equation with open boundary conditions is more accurate but requires an extensive computational budget. However, the emergence of graphics processing units (GPUs) in recent years represents an excellent opportunity to develop scalable quantum transport models for nanoscale transistors. In this thesis, such an efficient quantum transport modeling methodology has been introduced, employing GPU-based parallelization to accelerate the device simulation and, consequently, the discovery of suitable 2D materials from large material databases for transistor applications. First, we use this approach to investigate advanced 2D material-based transistors to address the critical challenge of high metal-semiconductor contact resistance. We propose a monolayer transistor architecture that uses the unique material chemistry of MXene to achieve low-resistive contacts. We use a high-throughput computational pipeline to perform DFT calculations and self-consistent ballistic quantum transport simulations for transistors on a large number of MXenes. Furthermore, we compare transistor performance to the International Roadmap for Devices and Systems (IRDS), indicating their potential for sub-decananometer technological scaling. Second, the scope of the proposed modeling and simulation methodology was further extended to explore the transistor performance employing a new semiconducting phase of borophene, clustered-P1, discovered using a genetic algorithm-based structure search. Clustered-P1 exhibits a bulk silicon-like band gap yet lower and symmetric effective masses, along with excellent dynamic and structural stability. The performance of clustered-P1 borophene based transistors is compared to IRDS standards and other 2D material-based transistors to illustrate their suitability as high-performance transistor applications. Finally, the methodology was applied to dissipative quantum transport modeling in order to explore the potential transistor application of boron compound BX (X= P, As, and Sb), which exhibits high room-temperature mobility because of their low effective mass and high optical phonon energy. Through ab initio calculations and dissipative quantum transport modeling, we demonstrate that these BX-based transistors are capable of providing near-coherent transport properties along with high ballisticity. The proposed first-principles-based device modeling framework, integrated with GPU acceleration, creates a connection between ab initio material modeling tools and quantum transport solver. This approach holds the potential to efficiently predict device performance prior to fabrication, thereby promising solutions to the critical challenges of design technology co-optimization associated with new materials.