Device Stack Optimization for Protonic Non-Volatile Programmable Resistors
Analog computing could alleviate computational bottlenecks in digital deep learning systems by utilizing local information processing through the physical properties of devices, such as electrochemical ion-intercalation in three-terminal devices where channel resistance is modulated by ionic exchang...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Published: |
Massachusetts Institute of Technology
2024
|
| Online Access: | https://hdl.handle.net/1721.1/156286 https://orcid.org/0009-0004-9904-8318 |
| Summary: | Analog computing could alleviate computational bottlenecks in digital deep learning systems by utilizing local information processing through the physical properties of devices, such as electrochemical ion-intercalation in three-terminal devices where channel resistance is modulated by ionic exchange via an electrolyte. Previous work has demonstrated such ionic programmable resistors featuring WO₃ as the channel, phosphorous-doped SiO₂ (PSG) as the electrolyte, Pd as the gate reservoir, and protons as the ions. This thesis aimed to optimize the device stack in four directions and demonstrated a symmetric WO₃-PSG-WO₃ structure in a CMOS-compatible process, with the help of circular transfer length model (CTLM), which efficiently examines the resistance properties of WO₃. We have explored: (a) device protonation as part of the fabrication process, (b) encapsulation preventing proton depletion during device fabrication and operation, (c) contact metal optimization to replace gold with a CMOS-compatible material, (d) PSG evaluation vehicle for device performance optimization. The symmetric device combining all the stack optimizations features non-volatile and repeatable conductance modulation with voltage pulses. |
|---|