Servicenavigation

Project Details

Back

Projekt Print View

MRAM Based Design, Test and Reliability for ultra Low Power SoC

Subject Area Computer Architecture, Embedded and Massively Parallel Systems
Electronic Semiconductors, Components and Circuits, Integrated Systems, Sensor Technology, Theoretical Electrical Engineering
Term from 2015 to 2022
Project identifier Deutsche Forschungsgemeinschaft (DFG) - Project number 284013114
 
Final Report Year 2022

Final Report Abstract

In any computing system, data storage and data movement are crucial steps. While technology scaling has provided benefits in terms of area reduction and performance improvements, both of dominant SRAM and DRAM memory technologies are facing various challenges, making them very hard to scale any further. Additionally, these memories are volatile, meaning that they lose their contents when they are powered off. While the amount of static (aka leakage) energy is increasing for these technologies, power gating strategies are not very straightforward for such memories as they lose their content and the valid data in such memories need to be saved somewhere else before they can be powered off. Due to these challenges, non-volatile memory technologies are very desirable for the replacement of on-chip SRAM and DRAM memories, particularly for normally-off and instanton computing to save leakage energy. Among several emerging non-volatile technologies, sprintronic memories, such as spin transfer torque (STT), have gain lots of interest in both industry and research communities due to several attractive features such as non-volatility, high density, high retention, high endurance and CMOS compatibility. However, there are several reliability and testing issue of this emerging technology which should be resolved, before widespread deployment. While the switching behavior of spintronic memories is inherently stochastic, the variability in the read and write operations of these memories should be modeled and accounted for at the design level. Additionally, the source of yield detractions in this technology should be understood and proper yield improvement techniques should be devised. At the same time the manifestation of manufacturing defects in this technology is very different than conventional CMOS-based memories, due to additional magnetic layers. Therefore, proper fault modeling for spintronic memories should be performed and accordingly, specific memory test patterns need to be developed for MRAM memories. The objective of this research project was to investigate, analyze, design, and optimize the use of non-volatility offered by STT-MRAM in a hybrid CMOS/MRAM normally-off computing architectures, and more especially in the context of ultra-low-power devices. The goal was to evaluate the full potential of this technology combined with an advanced CMOS technology by covering various aspects of cell, array, memory hierarchy, and architecture design as well as fault modeling, design for test, reliability, and robustness. We also developed a MRAM-based holistic design methodology, from cell to architecture-level, by designing a set of modeling, design, and simulation tools from circuit to system levels to design and evaluate hybrid memory hierarchy and processor architecture. We particularly explored reliability and test challenges in spintronic memories and provide proper models and techniques to analyze the variation effects in this technology, perform yield modeling and design for yield explorations, and finally perform reliability evaluation, design for reliability as well as defect characterization, fault modeling and test pattern generation for STT-MRAM memories.

Publications

 
 

Additional Information

© 2024 DFG Disclaimer / Copyright / Privacy Policy

Textvergrößerung und Kontrastanpassung