Georgia Institute of Technology
Characterization and mitigation of ADC noise by reference tuning in RRAM based CiM
Abstract
dc:description.abstractIn response to the escalating demand for low-power and high-compute capabilities, the exploration of Compute-in-Memory architecture for non-volatile memory has emerged as a promising avenue. However, the integration of circuits sensitive to reference voltage in this architecture necessitates addressing inherent challenges to uphold precision. Without altering the fundamental architecture, we systematically assess the accuracy of Multiply-Accumulate operations across diverse parameters, including reference scope, varied BL voltage, and different accumulation settings. To quantify inaccuracies, we modeled noise through effective bits, obtained by fitting actual response. Additionally, we affirm the absence of read disturb in the configuration. Subsequently, our model is applied to various neural networks encompassing supervised and reinforcement learning. Results indicate that intricate networks exhibit heightened susceptibility to noise, necessitating reference tuning at the per-ADC level, as for other simple networks, the per-module referencing will be sufficient. Furthermore, our findings highlight that reinforcement learning is more susceptible to noise compared to supervised learning.
Degree
thesis:*- Level thesis:degree_level
- Masters
- 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
-
- Wei, Ying-Hao
- Advisor dc:contributor.advisor
-
- Raychowdhury, Arijit
- Committee members dc:contributor.committeemember
-
- Yu, Shimeng
- Hao, Cong
Subjects
dc:subject × 5Rights
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/75179