Novel Strategies Toward High-Performance FeFET for Computing In Memory Application

Ferroelectric Field-Effect Transistors (FeFETs) attract the interest of researchers due to their capability of low power and high-speed operation for computing in memory (CIM). However, challenges with the poor memory window (MW) and reliability of FeFETs have triggered problems of density and reliability for CIM. Our study addresses these obstacles through a comprehensive approach, including materials, devices, and array architecture. We successfully demonstrated robust ferroelectric properties for relatively thick (~30 nm) ferroelectric (FE) materials. Furthermore, we adopted a metal-FE-metal-insulator-silicon (MFMIS) stack, enabling a controlling voltage distribution between FE and MIS components. This resulted in a wide MW ($\geq$ 11 V), improved switching speed ($< 50$ ns), and excellent endurance ($\geq 10^{10}$ cycles). Additionally, we demonstrated vector-matrix multiplication (VMM) operation using a fabricated FeFET-CIM array. We expect our research to offer a viable foundation for future research in CIM.

1. Introduction

FeFETs are viable candidates for CIM applications due to their multi-level operation, high on/off ratio, non-volatility, and compatibility with CMOS technology [1]. However, their narrow MW and bad endurance critically degrade the reliability of CIM. To overcome these challenges, we propose comprehensive strategies spanning material design, device structure, and array architecture. As a crucial facilitator for CIM, excellent ferroelectricity in relatively thick layers was proposed from a thermodynamics point of view. Furthermore, the MFMIS FeFET device with an outstanding FE layer displayed notable device properties with a wide-range MW V), high switching speed and outstanding endurance ( cycles), which can be attained by combining the effects from both material and device. In addition, the FeFET-PIM array, comprising developed FeFETs, could achieve high linearity in VMM operation. With the developed FeFET, which is capable of saving eight states in a single cell, our PIM array exhibited good area efficiency. The results in this paper can be a possible research platform for the research of future advanced computing devices.