A Multi-level Memristor Based on Al-Doped HfO2 Thin Film

Wu, Lei; Liu, Hongxia; Li, Jiabin; Wang, Shulong; Wang, Xing

doi:10.1186/s11671-019-3015-x

Cited by 51 publications

(33 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…A clear state difference is observed for the CB case, where LRS values are below 100 Ω and HRS values are in the range of 1–10 kΩ. These values are similar to the ones reported for sputtered Hf oxide memristors, but here the HRS/LRS ratio is slightly higher, exceeding one order of magnitude [ 44 , 45 ].…”

Section: Resultssupporting

confidence: 81%

Electrolyte-Dependent Modification of Resistive Switching in Anodic Hafnia

et al. 2021

View full text Add to dashboard Cite

Anodic HfO2 memristors grown in phosphate, borate, or citrate electrolytes and formed on sputtered Hf with Pt top electrodes are characterized at fundamental and device levels. The incorporation of electrolyte species deep into anodic memristors concomitant with HfO2 crystalline structure conservation is demonstrated by elemental analysis and atomic scale imaging. Upon electroforming, retention and endurance tests are performed on memristors. The use of borate results in the weakest memristive performance while the citrate demonstrates clear superior memristive properties with multilevel switching capabilities and high read/write cycling in the range of 106. Low temperature heating applied to memristors shows a direct influence on their behavior mainly due to surface release of water. Citrate-based memristors show remarkable properties independent on device operation temperatures up to 100 °C. The switching dynamic of anodic HfO2 memristors is discussed by analyzing high resolution transmission electron microscope images. Full and partial conductive filaments are visualized, and apart from their modeling, a concurrency of filaments is additionally observed. This is responsible for the multilevel switching mechanism in HfO2 and is related to device failure mechanisms.

show abstract

Section: Resultssupporting

confidence: 81%

Electrolyte-Dependent Modification of Resistive Switching in Anodic Hafnia

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Multi-level resistance states depending on the SET current compliance (I cc ) were investigated to find the influence of increasing oxygen vacancy concentration inside conducting filaments (CF) [ 36 , 37 , 38 ]. Figure 5 a shows the resistive switching characteristics with variable I cc during the SET process from 100 µA to 1 mA.…”

Section: Resultsmentioning

confidence: 99%

Multi-Level Analog Resistive Switching Characteristics in Tri-Layer HfO2/Al2O3/HfO2 Based Memristor on ITO Electrode

2020

View full text Add to dashboard Cite

Atomic layer deposited (ALD) HfO2/Al2O3/HfO2 tri-layer resistive random access memory (RRAM) structure has been studied with a transparent indium tin oxide (ITO) transparent electrode. Highly stable and reliable multilevel conductance can be controlled by the set current compliance and reset stop voltage in bipolar resistive switching. Improved gradual resistive switching was achieved because of the interdiffusion in the HfO2/Al2O3 interface where tri-valent Al incorporates with HfO2 and produces HfAlO. The uniformity in bipolar resistive switching with Ion/Ioff ratio (>10) and excellent endurance up to >103 cycles was achieved. Multilevel conductance levels in potentiation/depression were realized with constant amplitude pulse train and increasing pulse amplitude. Thus, tri-layer structure-based RRAM can be a potential candidate for the synaptic device in neuromorphic computing.

show abstract

“…Several works have shown that memristor devices can exhibit stable multi-level resistance states that are captured in the device models [38], [39]. Authors in [40] fabricated a Al-doped HfO 2 -based memristor with 20-levels. Also in [41] reported on a 12-levels stable resistance states using HfO 2−x .…”

Section: Proposed Work-flow For Reram-based Neural Network Inferencementioning

confidence: 99%

Design Exploration of ReRAM-Based Crossbar for AI Inference

2021

View full text Add to dashboard Cite

ReRAM-based crossbar designs utilizing mixed-signal implementation has gained importance due to their low power, small size, low cost, and high throughput especially for multiply-and-add operations in AI-related applications. This paper provides a framework with associated code for analyzing the impact of ReRAM device variation and post-training analog conductance quantization that has not been fully explored for pre-trained network accelerators. A detailed study with end-to-end implementation is presented, ranging from mapping the pre-trained DNN weights to quantized crossbar conductance values and into final classification with the presence of variation. Monte Carlo analysis was performed to better analyze the impact of the different parameters on the final accuracy without being device-specific. The work assumes different conductance value variations, different conductance dynamic ranges, and various device quantization levels (QLs). MNIST and CIFAR-10 data sets were used in this study for ANN and CNN, respectively. Results show that for simple ANN, the accuracy drop due to quantization was ∼ 2% at 64-QLs. While for CNN, the decrease in classification accuracy was around 10% with the same number of levels. Moreover, weight variation might cause a ∼ 5% and ∼ 8% drop in classification accuracy for ANN with 5% and CNN with 3% variation, respectively. The study confirms that increasing the number of levels with small variation results in nearoptimal accuracy. However, the increase in accuracy saturates at an upper limit. The amount of distortion propagated through the layers is different in the two cases. It is dependent on the complexity of input data and network structure, such as the size of the neurons in each layer, the number of layers, and the number of channels and filters at each stage. This contribution is the first to provide the framework to explore the implication that emphasizes on device-independent post-training DNN quantization and weight variation on classification accuracy. This helps explore design trade-offs especially for edge devices in cases where there is no access to the training set to the end-user due to security or cost issues for pre-trained networks. Device variation, in-memory computing, neural network, post-training, quantization, ReRAM. INDEX TERMS

show abstract

A Multi-level Memristor Based on Al-Doped HfO2 Thin Film

Cited by 51 publications

References 38 publications

Electrolyte-Dependent Modification of Resistive Switching in Anodic Hafnia

Electrolyte-Dependent Modification of Resistive Switching in Anodic Hafnia

Multi-Level Analog Resistive Switching Characteristics in Tri-Layer HfO2/Al2O3/HfO2 Based Memristor on ITO Electrode

Design Exploration of ReRAM-Based Crossbar for AI Inference

Contact Info

Product

Resources

About