Georgia Institute of Technology
Ferroelectric and Monolithic 3D (M3D) Memory Technologies for High-Performance, Energy-Efficient Computing
Abstract
dc:description.abstractPresent-day hardware with limited on-chip memory capacity falls short of meeting the speed and energy-efficiency requirement for data-intensive applications such as Artificial Intelligence due to the need for expensive data transfer between processor and off-chip memory. The objective of this research is to mitigate the energy-inefficiency and latency by developing ultra-high density cache memory and computational memory for enabling near-memory computing and in-memory computing. First, a back-end-of-line (BEOL)-compatible dual-gate (DG) W-doped In2O3 (IWO) field-effect transistor (FET) with high-performance, low-leakage and improved reliability is demonstrated, which is promising as access/read transistors for realizing M3D embedded-DRAM. The proposed DG IWO FET synergistically improves the device performance and VT stability, thereby breaking the performance-stability trade-off typically observed in BEOL FETs. Furthermore, a comprehensive reliability model is developed to elucidate the VT shift mechanisms. Second, we study and develop ferroelectric (FE) FET based computational memory for accelerating in-situ training and edge inference in deep neural networks (DNNs). On that front, we develop a scaled Si-channel FE-FinFET process and explores its potential application as a 2bit/cell weight cell for feature extraction and DNN inference. Lastly, this thesis proposes a novel idea of engineering ferroelectric gate stack using superlattice (SL) structure which enhances multi-state programmability of FEFET and thus can enable denser memory and novel functionality such as analog weight cell to accelerate in-situ training and inference of DNN. We experimentally demonstrate a BEOL SL FEFET analog weight cell with a record-high 1000 analog conductance states exhibiting a high degree of linearity and symmetry. We also demonstrate 8 non-overlapping programmed states for the SL FEFET which enables enhanced multi-level-cell operation (3bit/cell) over conventional FEFETs.
Degree
thesis:*- Level thesis:degree_level
- Doctoral
- Department dc:contributor.department
- Electrical and Computer Engineering
- Grantor dc:publisher
- Georgia Institute of Technology
- Year dc:date.issued
- 2024
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Aabrar, Khandker Akif
- Advisor dc:contributor.advisor
-
- Datta, Suman
- Committee members dc:contributor.committeemember
-
- Mukhopadhyay, Saibal
- Yu, Shimeng
- Khan, Asif Islam
- Ni, Kai
Subjects
dc:subject × 5Rights
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/80067