More Like this

1. A compute-in-memory chip based on resistive random-access memory

   https://doi.org/10.1038/s41586-022-04992-8  

   Wan, Weier; Kubendran, Rajkumar; Schaefer, Clemens; Eryilmaz, Sukru Burc; Zhang, Wenqiang; Wu, Dabin; Deiss, Stephen; Raina, Priyanka; Qian, He; Gao, Bin; et al (August 2022, Nature)

   Abstract Realizing increasingly complex artificial intelligence (AI) functionalities directly on edge devices calls for unprecedented energy efficiency of edge hardware. Compute-in-memory (CIM) based on resistive random-access memory (RRAM) 1 promises to meet such demand by storing AI model weights in dense, analogue and non-volatile RRAM devices, and by performing AI computation directly within RRAM, thus eliminating power-hungry data movement between separate compute and memory 2–5 . Although recent studies have demonstrated in-memory matrix-vector multiplication on fully integrated RRAM-CIM hardware 6–17 , it remains a goal for a RRAM-CIM chip to simultaneously deliver high energy efficiency, versatility to support diverse models and software-comparable accuracy. Although efficiency, versatility and accuracy are all indispensable for broad adoption of the technology, the inter-related trade-offs among them cannot be addressed by isolated improvements on any single abstraction level of the design. Here, by co-optimizing across all hierarchies of the design from algorithms and architecture to circuits and devices, we present NeuRRAM—a RRAM-based CIM chip that simultaneously delivers versatility in reconfiguring CIM cores for diverse model architectures, energy efficiency that is two-times better than previous state-of-the-art RRAM-CIM chips across various computational bit-precisions, and inference accuracy comparable to software models quantized to four-bit weights across various AI tasks, including accuracy of 99.0 percent on MNIST 18 and 85.7 percent on CIFAR-10 19 image classification, 84.7-percent accuracy on Google speech command recognition 20 , and a 70-percent reduction in image-reconstruction error on a Bayesian image-recovery task. 

   more » « less
2. Influence of Oxygen Vacancy and Top Electrode on Switching Behavior of InGaZnO Based Resistive Random Access Memory

   Qin, Fei; Lee, Sunghwan (June 2022, 64th Electronic Materials Conference)

   Biomimetic synaptic processes, which are imitated by functional memory devices in the computer industry, are a key focus of artificial intelligence (AI) research. It is critical to developing a memory technology that is compatible with Brain-Inspired Computing in order to eliminate the von Neumann bottleneck that restricts the efficiency of traditional computer designs. Due to restrictions such as high operation voltage, poor retention capacity, and high power consumption, silicon-based flash memory, which presently dominates the data storage devices market, is having difficulty meeting the requirements of future data storage device development. The developing resistive random-access memory (RRAM) has sparked intense investigation because of its simple two-terminal structure: two electrodes and a switching layer. RRAM has a resistive switching phenomenon which is a cycling behavior between the high resistance state and the low resistance state. This developing device type is projected to outperform traditional memory devices. Indium gallium zinc oxide (IGZO) has attracted great attention for the RRAM switching layer because of its high transparency and high atomic diffusion property of oxygen atoms. More importantly, by controlling the oxygen ratio in the sputter gas, its electrical properties can be easily tuned. The IGZO has been applied to the thin-film transistor (TFT), thus, it is very promising to integrate RRAM with TFT. In this work, we proposed IGZO-based RRAMs. ITO was chosen as the bottom electrode towards achieving a fully transparent memristor. And for the IGZO switching layer, we varied the O2/Ar ratio during the deposition to modify the oxygen vacancy of IGZO. Through the XPS measurement, we confirmed that the higher O2/Ar ratio can lower the oxygen vacancy concentration. We also chose ITO as the top electrode, for the comparison, two active metals copper and silver were tested for the top electrode materials. For our IGZO layer, the best ratio of O2/Ar is the middle value. And copper top electrode device has the most stable cycling switching and the silver one is perfect for large memory window, however, it encounters a stability issue. The optical transmission examination was performed using a UV-Vis spectrometer, and the average transmittance of the complete devices in the visible-light wavelength range was greater than 90%, indicating good transparency. 50nm, 100nm, and 150nm RS layers of IGZO RRAM were produced to explore the thickness dependency on the characteristics of the RS layer. Also, because the oxygen vacancy concentration influences the RS and RRAM performance, the oxygen partial pressure during IGZO sputtering was modified to maximize the property. Electrode selection is critical and can have a significant influence on the device's overall performance. As a result, Cu TE was chosen for our second type of device because Cu ion diffusion can aid in the development of conductive filaments (CF). Finally, between the TE and RS layers, a 5 nm SiO2 barrier layer was used to limit Cu penetration into the RS layer. Simultaneously, this SiO2 inserting layer can offer extra interfacial series resistance in the device, lowering the off current and, as a result, improving the on/off ratio and overall performance. In conclusion, transparent IGZO-based RRAMs have been created. The thickness of the RS layer and the sputtering conditions of the RS layer were modified to tailor the property of the RS layer. A series of TE materials and a barrier layer were incorporated into an IGZO-based RRAM and the performance was evaluated in order to design the TE material's diffusion capabilities to the RS layer and the BE. Our positive findings show that IGZO is a potential material for RRAM applications and overcoming the existing memory technology limitation. 

   more » « less
3. Hyb-Learn: A Framework for On-Device Self-Supervised Continual Learning with Hybrid RRAM/SRAM Memory

   https://doi.org/10.1145/3649329.3657389  

   Zhang, Fan; Yang, Li; Fan, Deliang (June 2024, ACM)

   While RRAM crossbar-based In-Memory Computing (IMC) has proven highly effective in accelerating Deep Neural Networks (DNNs) inference, RRAM-based on-device training is less explored due to its high energy consumption of weight re-programming and cells' low endurance problem. Besides, emerging trends indicate a need for on-device continual learning which sequentially acquires knowledge from multiple tasks to enhance user's experiences and eliminate data privacy concerns. However, learning on each new task leads to forgetting prior learned knowledge on prior tasks, which is known as catastrophic forgetting. To address these challenges, we are the first to propose a novel training framework, Hyb-Learn, for enabling on-device continual learning with a hybrid RRAM/SRAM IMC architecture design. Specifically, when training each new arriving task, our approach first partitions the model into two groups based on the proposed task-correlated PE-wise correlation to freeze or re-training, and correspondingly mapping to RRAM and SRAM, respectively. In practice, the RRAM stores frozen weights with strong task correlation to prior tasks to eliminate the high cost of weight reprogramming issue of RRAM, while the SRAM stores the remaining weights that will be updated. Furthermore, to maximize the freezing ratio for improving training efficiency while maintaining accuracy and mitigating catastrophic forgetting, we incorporate self-supervised learning algorithms that are initialized from a pre-trained model for training each new task. 

   more » « less
4. The Impact of Operation Conditions on Potentiation, Depression and Endurance Dynamics of Tantalum Oxide RRAMs

   https://doi.org/10.1109/IIRW59383.2023.10477696  

   Moazzeni, Alireza; Tawsif Rahman Chowdhury, Md; Tutuncuoglu, Gozde (October 2023, IEEE International Integrated Reliability Workshop)

   Resistive Random Access Memory (RRAM) devices hold promise as a key enabler technology for energy-efficient, in-memory, and brain-inspired computing paradigms, with the potential to significantly enhance high-performance computing applications. However, the widespread adoption of RRAM technology in high-performance computing applications is hindered by non-ideal device metrics and various reliability challenges. RRAM devices are reported to exhibit critical device-to-device (D2D) and cycle-to-cycle (C2C) variability. In this paper, we investigate D2D and C2C variabilities of Tantalum Oxide RRAM devices and explore potentiation, depression, and endurance dynamics under varying operation conditions. Our ultimate goal is to address performance and reliability issues associated with the oxide-based RRAM device technology and facilitate its broader implementation in future computing applications. 

   more » « less
5. Transparent InGaZnO-Based Resistive Random Access Memory

   Qin, Fei; Lee, Sunghwan (May 2022, Materials Research Society)

   The major focus of artificial intelligence (AI) research is made on biomimetic synaptic processes that are mimicked by functional memory devices in the computer industry [1]. It is urgent to find a memory technology for suiting with Brain-Inspired Computing to break the von Neumann bottleneck which limits the efficiency of conventional computer architectures [2]. Silicon-based flash memory, which currently dominates the market for data storage devices, is facing challenging issues to meet the needs of future data storage device development due to the limitations, such as high-power consumption, high operation voltage, and low retention capacity [1]. The emerging resistive random-access memory (RRAM) has elicited intense research as its simple sandwiched structure, including top electrode (TE) layer, bottom electrode (BE) layer, and an intermediate resistive switching (RS) layer, can store data using RS phenomenon between the high resistance state (HRS) and the low resistance state (LRS). This class of emerging devices is expected to outperform conventional memory devices [3]. Specifically, the advantages of RRAM include low-voltage operation, short programming time, great cyclic stability, and good scalability [4]. Among the materials for RS layer, indium gallium zinc oxide (IGZO) has attracted attention because of its abundance and high atomic diffusion property of oxygen atoms, transparency, and its easily modulated electrical properties by controlling the stoichiometric ratio of indium and gallium as well as oxygen potential in the sputter gas [5, 6]. Moreover, since the IGZO can be applied to both the thin-film transistor (TFT) channel and RS layer, the IGZO-based fully integrated transparent electronics are very promising [5]. In this work, we proposed transparent IGZO-based RRAMs. First, we chose ITO to serve as both TE and BE to achieve high transmittance in the visible regime of light. All three layers (TE, RS, BE layers) were deposited using a multi-target magnetron sputtering system on glass substrates to demonstrate fully transparent oxide-based devices. I-V characteristics were evaluated by a semiconductor parameter analyzer, and our devices showed typical butterfly curves indicating the bipolar RS property. And the IGZO-based RRAM can survive more than 50 continuous sweeping cycles. The optical transmission analysis was carried out via an UV-Vis spectrometer and the average transmittance around 80% out of entire devices in the visible-light wavelength range, implying high transparency. To investigate the thickness dependence on the properties of RS layer, 50nm, 100nm and 150nm RS layer of IGZO RRAM were fabricated. Also, the oxygen partial pressure during the sputtering of IGZO was varied to optimize the property because the oxygen vacancy concentration governs the RS and RRAM performance. Electrode selection is crucial and can impact the performance of the whole device [7]. Thus, Cu TE was chosen for our second type of device because the diffusion of Cu ions can be beneficial for the formation the conductive filament (CF). Finally, a ~5 nm SiO2 barrier layer was employed between TE and RS layers to confine the diffusion of Cu into the RS layer. At the same time, this SiO2 inserting layer can provide an additional interfacial series resistance in the device to lower the off current, consequently, improve the on/off ratio and whole performance. In conclusion, the transparent IGZO-based RRAMs were established. To tune the property of RS layer, the thickness layer and sputtering conditions of RS were adjusted. In order to engineer the diffusion capability of the TE material to the RS layer and the BE, a set of TE materials and a barrier layer were integrated in IGZO-based RRAM and the performance was compared. Our encouraging results clearly demonstrate that IGZO is a promising material in RRAM applications and overcoming the bottleneck of current memory technologies. 

   more » « less