Development of High-Performance Hafnium Oxide based Non-Volatile Memory Devices on 300mm Wafer Platform for Data Storage and Neuromorphic Applications
Final Dissertation Submission
Dissertation Approval Form
Cast your vote
You can rate an item by clicking the amount of stars they wish to award to this item.
When enough users have cast their vote on this item, the average rating will also be shown.
Your vote was cast
Thank you for your feedback
Thank you for your feedback
Keywordnon-volatile memory (NVM)
resistive random-access memory (RRAM)
von Neumann computing architectures
Dynamic random-access memory (DRAM)
high performance embedded memory
mass data storage
MetadataShow full item record
AbstractFundamental limitations associated with scaling and modern von Neumann computing architectures illustrates the need for emerging memory solutions in the semiconductor industry. One such promising non-volatile memory (NVM) solution is resistive random access memory (RRAM), which is seen as a potential candidate that can meet the performance needs of DRAM and the density of NAND Flash in terms of scalability, reliability and switching performance. However, reliable operation of RRAM devices requires further development to remedy device- to-device and cycle-to-cycle uniformity variation, increase the conductance window, and to improve retention, yield and endurance properties. This research work primarily focuses on improving RRAM performance metrics through optimization of processing conditions and programming algorithms for CMOS-integrated nanoscale HfO2 RRAM devices on a full scale 300mm wafer platform. It was observed that tuning of ALD parameters during RRAM switching layer HfO2 deposition had a significant impact on device switching performance. An excellent memory window of >30 with switching yield ~90%, along with low cycle-to-cycle (σ <0.5) and cell-to-cell variability (σ <0.4) were achieved for tested 1 Transistor 1 RRAM (1T1R) cells across full 300mm wafers. The devices demonstrated excellent endurance (>1010 switching cycles) and data retention performance at elevated temperature (105 s at 373K). The fabricated RRAM cells were also optimized for multi-level-cell switching behavior and ~10 distinct resistance levels were obtained through a combined current- and voltage-control based programming approach. An incremental pulse write technique combined with read verification algorithm enabled accurate resistance states programming within a large resistance window along with linear and symmetric potentiation-depression characteristics yielding superior analog synaptic functionality of fabricated RRAM devices. In addition to RRAM devices, hafnium zirconium oxide (HZO) based nanoscale ferroelectric tunnel junction (FTJ) devices were successfully implemented on a 300 mm wafer platform. Current measurement, as a function of voltage for both up and down polarization states, yielded a tunneling electroresistance (TER) ratio of ~5 and switching endurance up to 106 cycles in TiN/ Al2O3/ Hf0.5Zr0.5O2/ TiN FTJ devices distributed across full 300 mm wafer. Investigation of current transport mechanisms showed that the conduction in these FTJ devices is dominated by direct tunneling (DT) at low electric field and by Fowler-Nordheim (F-N) tunneling at high electric field. The realization of CMOS-compatible nanoscale RRAM and FTJ devices on 300mm wafers demonstrates the promising potential of these devices in large scale high-yield NVM manufacturing for high performance embedded memory and mass data storage applications.
Showing items related by title, author, creator and subject.
An Examination of Clinical Measurements of Verbal Working MemoryHungerford, Suzanne; Roberts, Katharine (2011)Research indicates that verbal working memory is an important predictor of oral and written language processing, social problem solving, and academic success. This study examines working memory scores derived from common clinical tools in speech-language pathology, the relationships among these variables, and their ability to predict language and related skills.
Application of Resistive Random Access Memory (RRAM) For Non-Von Neumann ComputingRafiq, Sarah (2022-05)The movement of data between physically separated memory and processing units in conventional computing systems (the so-called von Neumann architecture) incurs significant costs in energy and latency. This is known as the von Neumann bottleneck. With the advent of the Internet of Things (IoT) and edge computing, computing systems are also becoming significantly power limited. In this work, hafnium oxide resistive random access memory (ReRAM) integrated with 65nm CMOS technology on a 300 mm wafer platform was assessed to carry out two novel non-von Neumann computing applications that processes data within memory and avoid excessive data movement. These computing applications are based on regulating the flow of sneak path currents in memory arrays to perform computation, called flow-based computing, and detecting degree of association (correlation) between binary processes in an unsupervised manner using the ReRAM non-volatile accumulative behavior, termed as temporal correlation detection. Electrical characterization of hafnium oxide ReRAM arrays was conducted for multi-level resistance states for flow-based computing, which was then investigated for two functions, approximate edge detection and XOR Boolean logic, through both experiments and simulation. The effect of device non-idealities was also evaluated. A trade-off between the flow-based output resistance ratio and the variability of flow-based outputs was found for different patterned binary resistance Roff/Ron ratios. For the second non-von Neumann application, the feasibility of ReRAM as a non-volatile candidate device was investigated with an empirical ReRAM model through simulation. Experimental ReRAM analog incremental switching data, from both SET and RESET regimes, was also evaluated on the modified temporal correlation detection algorithm, where the RESET regime resulted in better performance. The ReRAM based implementation yielded 36,000-53,000 vi times lower energy consumption than similar implementation with phase change memory for 25 binary processes, and a speed-up of computation time by 1,600-2,100 times than that of a CPU-based implementation using 1xPOWER8 CPU. 1xPOWER8 CPU is a CPU available on the IBM* Power* System S822LC system, the POWER8 system series, where the CPU was run for 1 thread. In summary, hafnium oxide ReRAM based on 65nm CMOS technology has been evaluated for two non-von Neumann computing applications, and the effect of device non-idealities has also been assessed. These ReRAM in-memory computing applications show the promising potential of ReRAM in overcoming the von-Neumann bottleneck.
RRAM Device Optimization and Circuit Design for Low Power Low Latency Domain Specific Edge ApplicationsAbedin, Minhaz Ibna (2023-05)Conventional von Neumann computing architecture has little parallelism and physically divides memory and logic units. In-memory computation offers further opportunity for power and efficiency improvements as data transport between memory and logic units introduces substantial power consumption and latency. RRAM memory devices are a non-volatile, highly scalable alternative that is compatible with advanced logic CMOS processes. Because of its in-memory computing capabilities, it is a more appealing solution for data-intensive applications. Depending on the needs of the application, RRAM's different switching types such as binary switching, multilayer switching, and analog switching can be employed. These devices can be utilized in applications where they must continuously update their condition (e.g neural network training). The RRAM memory system can be of great use in applications where the device must continuously maintain the resistance state as well, either with or without in-memory processing. Moreover, RRAM devices are currently being researched for in-sensor or near-sensor applications to speed up AI inference. The diversity of RRAM switching schemes and application range creates vast research opportunities to examine the viability of these devices. Hence, leveraging RRAM’s unique switching properties analog (continuous) or binary/multilevel switching, in-memory computation scheme (MAC or bit-wise) as well as integration location (in-pixel, near pixel or memory) there is still a wide range of different applications yet to be explored. In this work, RRAM analog switching properties of based on different materials stacks, inference-like applications such as error correcting code implementation, genome alignment, and in-sensor AI inference acceleration have been investigated.