Design of a benchmarking platform for Logic-In-Memory architectures based on ferroelectric HfO2

Silicon devices have undergone a massive down scaling process in the last decades, this brought transistors near the physical limitations introduced by the quantum mechanics. In this area the boundaries of conventional computation are stretched to their limits in order to increase performances and decrease power consumption. This path is rapidly becoming more expensive and major breakthroughs are more difficult to reach. In this scenario different technologies are being investigated to fill the holes left by silicon, especially computing technologies that can be non-volatile. Hafnium dioxide (HfO2) stands out in this landscape because of its ferroelectric properties that make of it an intrinsic memory. Moreover, its use as high K insulator in modern CMOS processes makes it perfect for integration with existing devices and production steps. At the same time evolution in computing architectures has been minimal with respect to the one of the hardware: modern systems still implement fined tuned versions of the Von Neumann architectures. This introduced a mismatch in performances, mainly created by the huge amount of time and energy spent on communications between the memory and the processor unit, the so called Von Neumann bottleneck. The task to explore the design space is impressive given the numerous degree of freedom introduced by both device and system level. This thesis proposes a benchmarking platform designed in SYSTEMC. Its aim is to speed up the process by simulating and then comparing different solutions both in the technological and in the architectural department. At the same time the estimation must remain as faithful as possible to the nature of different implementations. Furthermore, the platform could be employed to find strengths and weaknesses of the various emerging technologies with respect to each other and CMOS. This could initiate a feedback loop with device level research to help to improve the technological nodes and by reflection the overall performances. This thesis is divided in three chapters. The first one is devoted to the ferroelectric properties of HfO2, since it is the main target technology for this stage of the project. The physical phenomenon is introduced, the role of this material in CMOS processes is underlined and an overview of the main ferroelectric devices is given. The second chapter provides an exhaustive taxonomy of the ever-growing Logic in Memory (LiM) architectures field. The most important paradigms are explained and compared with each other. The last chapter focuses on the realization of the simulation platform. The design of the structure is unraveled, the different components are analyzed together with how they interact. The current state of the platform allows for LiM simulations in both CMOS and ferroelectric technology with different architectures, but the quantitative results are still incomplete due to the platform not being mature enough. However, they prove how this tool has the potentialities to rise up to the challenge.