Fault-Tolerant Circuit Design for the Dynamic Operation Unit in RRAM CIM Architecture
Authors:
- 16 July 2025
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 291-292
Abstract
Resistive random-access memory (RRAM) offers high speed, low power consumption, and long durability, making it highly promising for computing-in-memories (CIM) design. However, the crossbar array design is constrained by the limited number of word lines and bit lines that can be activated simultaneously. To address this limitation, the concept of a dynamic operation unit has been introduced, but previous studies have not considered fault tolerance in its design. To enhance chip yield and prevent the entire chip from becoming unusable due to a few faulty memory cells, this paper proposes a fault-tolerance mechanism that incorporates spare rows for the dynamic operation unit. Experimental results consistently show that the proposed approach incurs minimal area and power overhead while introducing no timing overhead.Keywords
Funding Information
- National Science and Technology Council
This publication has 5 references indexed in Scilit:
- PIM-Prune: Fine-Grain DCNN Pruning for Crossbar-Based Process-In-Memory Architecture2020
- Sparse ReRAM engine2019
- Eyeriss v2: A Flexible Accelerator for Emerging Deep Neural Networks on Mobile DevicesIEEE Journal on Emerging and Selected Topics in Circuits and Systems, 2019
- Accelerator-friendly neural-network training: Learning variations and defects in RRAM crossbar2017
- Cambricon-X: An accelerator for sparse neural networks2016