Detail View

A High-Resistance FeFET based Time-domain CiM Macro with MAC Distribution-Aware Readout

Citations

WEB OF SCIENCE

Citations

SCOPUS

Metadata Downloads

DC Field Value Language
dc.contributor.advisor 윤종혁 -
dc.contributor.author 김태우 -
dc.date.accessioned 2026-09-01T19:32:34Z -
dc.date.available 2026-09-02T06:00:35Z -
dc.date.issued 2026 -
dc.identifier.uri https://scholar.dgist.ac.kr/handle/20.500.11750/60802 -
dc.identifier.uri http://dgist.dcollection.net/common/orgView/200001007035 -
dc.description Compute-in-Memory, Non-volatile Memory, FeFET, MAC operation, delay time, MAC distribution -
dc.description.abstract Compute-in-Memory (CIM) has emerged as a promising solution to overcome the memory bottleneck in conventional von-Neumann architectures, particularly for data-intensive applications such as machine learning (ML) and edge computing.
This thesis proposes a time-domain FeFET-based compute-in-memory (CIM) macro that performs multiply-and-accumulate (MAC) operations using delay cells. The proposed architecture alleviates the limited sensing-margin problem of conventional voltage- and current-domain CIM by representing MAC results as accumulated propagation delays. To mitigate the long bitline-discharge latency caused by the high resistance of FeFET devices, the resistance-dependent voltage transition of the FeFET RC network is converted into a delay. In addition, a MAC-distribution-aware hybrid TDC applies linear quantization to the low-MAC region and a limited number of nonlinear references to the high-MAC region, thereby reducing hardware overhead while maintaining computational accuracy.
The proposed design provides a compact, low-power, and reliable CIM optimized for next-generation edge
devices, effectively demonstrating the potential of time-domain computing using FeFET technology.| 인-메모리 연산은 기존 폰 노이만 구조에서 발생하는 메모리 병목 문제를 극복하기 위한 유망한 해결책으로 주목받고 있으며, 특히 머신러닝 및 엣지 컴퓨팅과 같은 데이터 집약적 응용 분야에서 그 중요성이 커지고 있다.
본 학위논문에서는 지연 셀을 이용하여 곱셈-누산 연산을 수행하는 시간 영역 FeFET 기반 Compute-in-Memory 매크로를 제안한다. 제안하는 구조는 MAC 결과를 누적된 전파 지연으로 표현함으로써 기존 전압 및 전류 영역 CIM 구조의 제한된 센싱 마진 문제를 완화한다. 또한 FeFET 소자의 높은 저항으로 인해 발생하는 긴 비트라인 방전 지연을 개선하기 위해, FeFET RC 네트워크의 저항 의존적 전압 천이를 지연 시간으로 변환한다. 이와 함께 MAC 분포 인지형 hybrid TDC는 낮은 MAC 영역에 선형 양자화를 적용하고 높은 MAC 영역에 제한된 수의 비선형 기준을 배치함으로써, 연산 정확도를 유지하면서 하드웨어 오버헤드를 줄인다.
시뮬레이션 및 분석을 통해 제안된 매크로가 소형, 저전력, 신뢰성을 갖춘 엣지 지향 인메모리 연산 솔루션임을 확인하였으며, FeFET 기반 시간 영역 컴퓨팅의 실용 가능성을 입증하였다.
-
dc.description.tableofcontents Abstract i
List of contents ii
List of figures iii
List of tables ⅳ

Ⅰ. Introduction 1

Ⅱ. Background 3
2.1 Prior non-volatile CIM works 3
2.1.1 PCRAM-based CIM macro 4
2.1.2 RRAM-based CIM macro 5
2.2 Characteristic of FeFET 5
2.2.1 Electrical properties of FeFET 6
2.2.2 128x128 FeFET array 7
2.3 Challenges and solutions in FeFET based CIM 8

Ⅲ. Proposed time domain based FeFET CIM 11
3.1 Overall architecture of the proposed FeFET CIM macro 11
3.2 Resistance-dependent delay chain 12
3.3 MAC distribution aware TDC 14

Ⅳ. Simulation results 18

Ⅴ. Conclusion 23
-
dc.format.extent 25 -
dc.language eng -
dc.publisher DGIST -
dc.title A High-Resistance FeFET based Time-domain CiM Macro with MAC Distribution-Aware Readout -
dc.type Thesis -
dc.identifier.doi 10.22677/THESIS.200001007035 -
dc.description.degree Master -
dc.contributor.department Department of Electrical Engineering and Computer Science -
dc.contributor.coadvisor Minyoung Song -
dc.date.awarded 2026-08-01 -
dc.publisher.location Daegu -
dc.description.database dCollection -
dc.citation XT.IM 김88 202608 -
dc.date.accepted 2026-07-21 -
dc.contributor.alternativeDepartment 전기전자컴퓨터공학과 -
dc.subject.keyword Compute-in-Memory, Non-volatile Memory, FeFET, MAC operation, delay time, MAC distribution -
dc.contributor.affiliatedAuthor 김태우 -
dc.contributor.affiliatedAuthor Jong-Hyeok Yoon -
dc.contributor.affiliatedAuthor Minyoung Song -
dc.contributor.alternativeName 김태우 -
dc.contributor.alternativeName Jong-Hyeok Yoon -
dc.contributor.alternativeName 송민영 -
Show Simple Item Record

File Downloads

  • There are no files associated with this item.

공유

qrcode
공유하기

Total Views & Downloads

???jsp.display-item.statistics.view???: , ???jsp.display-item.statistics.download???: