XNOR-RRAM: A scalable and parallel resistive synaptic architecture for binary neural networks
The DOI matches its Crossref record; 18 of the 20 references it cites were found in a registry or an index.
Abstract
Recent advances in deep learning have shown that Binary Neural Networks (BNNs) are capable of providing a satisfying accuracy on various image datasets with significant reduction in computation and memory cost. With both weights and activations binarized to +1 or -1 in BNNs, the high-precision multiply-and-accumulate (MAC) operations can be replaced by XNOR and bit-counting operations. In this work, we propose a RRAM synaptic architecture (XNOR-RRAM) with a bit-cell design of complementary word lines that implements equivalent XNOR and bit-counting operation in a parallel fashion. For large-scale matrices in fully connected layers or when the convolution kernels are unrolled in multiple channels, the array partition is necessary. Multi-level sense amplifiers (MLSAs) are employed as the intermediate interface for accumulating partial weighted sum. However, a low bit-level MLSA and intrinsic offset of MLSA may degrade the classification accuracy. We investigate the impact of sensing offsets on classification accuracy and analyze various design options with different sub-array sizes and sensing bit-levels. Experimental results with RRAM models and 65nm CMOS PDK show that the system with 128×128 sub-array size and 3-bit MLSA can achieve accuracies of 98.43% for MLP on MNIST and 86.08% for CNN on CIFAR-10, showing 0.34% and 2.39% degradation respectively compared to the accuracies of ideal BNN algorithms. The projected energy-efficiency of XNOR-RRAM is 141.18 TOPS/W, showing ~33X improvement compared to the conventional RRAM synaptic architecture with sequential row-by-row read-out.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“In binarized networks, ADC can be replaced by a 1-bit comparator96 or ADC-like multi-level sense amplifier158.”
From Hardware implementation of memristor-based artificial neural networks (2024) · cites this paper as [158]
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474
XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks
Rastegari, M., Ordonez, V., Redmon, J., & Farhadi, A. (2016) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-46493-0_32 · Cited by 4,408
Eyeriss: An Energy-Efficient Reconfigurable Accelerator for Deep Convolutional Neural Networks
Chen, Y.-H., Krishna, T., Emer, J. S., & Sze, V. (2017) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2016.2616357 · Cited by 3,247
Courbariaux, M., Hubara, I., Soudry, D., El-Yaniv, R., & Bengio, Y. (2016) · arXiv · doi:10.48550/arxiv.1602.02830 · Cited by 2,106
Quantizing for minimum distortion
Max, J. (1960) · IEEE Transactions on Information Theory · doi:10.1109/tit.1960.1057548 · Cited by 2,038
14.5 Eyeriss: An energy-efficient reconfigurable accelerator for deep convolutional neural networks
Chen, Y.-H., Krishna, T., Emer, J., & Sze, V. (2016) · 2016 IEEE International Solid-State Circuits Conference (ISSCC) · doi:10.1109/isscc.2016.7418007 · Cited by 713
Chen, P.-Y., Peng, X., & Yu, S. (2017) · 2017 IEEE International Electron Devices Meeting (IEDM) · doi:10.1109/iedm.2017.8268337 · Cited by 435
Self-citationMemristor Crossbar-Based Neuromorphic Computing System: A Case Study
Hu, M., Li, H., Chen, Y., Wu, Q., Rose, G. S., & Linderman, R. W. (2014) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2013.2296777 · Cited by 414
Mitigating effects of non-ideal synaptic device characteristics for on-chip learning
Chen, P.-Y., Lin, B., Wang, I.-T., Hou, T.-H., Ye, J., Vrudhula, S., Seo, J.-S., Cao, Y., & Yu, S. (2015) · 2015 IEEE/ACM International Conference on Computer-Aided Design (ICCAD) · doi:10.1109/iccad.2015.7372570 · Cited by 232
Self-citationBinary neural network with 16 Mb RRAM macro chip for classification and online training
Yu, S., Li, Z., Chen, P.-Y., Wu, H., Gao, B., Wang, D., Wu, W., & Qian, H. (2016) · 2016 IEEE International Electron Devices Meeting (IEDM) · doi:10.1109/iedm.2016.7838429 · Cited by 219
Self-citation
Cited by
Items 101–200 of 247
A Mini Tutorial of Processing in Memory: From Principles, Devices to Prototypes
Pan, B., Wang, G., Zhang, H., Kang, W., & Zhao, W. S. (2022) · IEEE Transactions on Circuits & Systems II Express Briefs · doi:10.1109/tcsii.2022.3172494 · Cited by 15
Hardware-aware Quantization/Mapping Strategies for Compute-in-Memory Accelerators
Huang, S., Jiang, H., & Yu, S. (2022) · ACM Transactions on Design Automation of Electronic Systems · doi:10.1145/3569940 · Cited by 17
Self-citationLogic-in-Memory Computation: Is It Worth It? A Binary Neural Network Case Study
Coluccio, A., Vacca, M., & Turvani, G. (2020) · Journal of Low Power Electronics and Applications · doi:10.3390/jlpea10010007 · Cited by 14
“In [26,27,30,31] several Binary Convolutional Neural Networks (BCNNs) implementations are discussed: they achieve very good results in terms of energy and power, thanks to the intrinsic low power nature of the MTJ and RRAM devices.”
Saha, G., Jiang, Z., Parihar, S., Cao, X., Higman, J. M., & Karim, M. A. U. (2020) · IEEE Access · doi:10.1109/access.2020.2993989 · Cited by 12
“Moreover, in our scheme, we have eliminated the need for area and energy hungry ADCs [19] and Multi-level Sense Amplifiers (MLSAs) [20] for pop-counting operation.”
BackgroundFully Binarized, Parallel, RRAM-Based Computing Primitive for In-Memory Similarity Search
Kingra, S. K., Parmar, V., Verma, D., Bricalli, A., Piccolboni, G., Molas, G., Regev, A., & Suri, M. (2022) · IEEE Transactions on Circuits & Systems II Express Briefs · doi:10.1109/tcsii.2022.3207378 · Cited by 12
“3207378 circuits have performance limitations that can be potentially addressed by using emerging NVM (Non-Volatile Memory) devices [10], [11], [12].”
BackgroundGraphene-based 3D XNOR-VRRAM with ternary precision for neuromorphic computing
Alimkhanuly, B., Sohn, J., Chang, I., & Lee, S. (2021) · npj 2D Materials and Applications · doi:10.1038/s41699-021-00236-x · Cited by 15
BNN An Ideal Architecture for Acceleration With Resistive in Memory Computation
Ding, A., Qiao, Y., & Bagherzadeh, N. (2023) · IEEE Transactions on Emerging Topics in Computing · doi:10.1109/tetc.2023.3237778 · Cited by 12
Chang, C.-Y., Huang, C.-T., Chuang, Y.-C., Chou, K.-C., & Wu, A.-Y. A. (2023) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2023.3338378 · Cited by 12
“Compared with prior works that calibrate ADC references for preserving inference accuracy [13], [14], [16], [17], proposed BFP-CIM only requires linear ADCs without complex reference calibration.”
BackgroundMethodMohseni, A., Moaiyeri, M. H., Amirany, A., & Rezayati, M. H. (2024) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2024.3397925 · Cited by 11
“In [32], a parallel XNOR-Resistive RAM (RRAM) architecture with a custom bit-cell design was proposed for efficient BNN inference.”
MethodInference engine benchmarking across technological platforms from CMOS to RRAM
Peng, X., Kim, M., Sun, X., Yin, S., Rakshit, T., Hatcher, R., Kittl, J. A., Seo, J.-S., & Yu, S. (2019) · International Symposium on Memory Systems (MEMSYS) · doi:10.1145/3357526.3357566 · Cited by 13
“To minimize the ADC truncation effects on the partial-sums, we utilize the nonlinear quantization with various quantization edges (corresponding to different ADC precision), where the edges are determined according to the distribution of partial-sums, as proposed in [17].”
MethodResultInfluentialSelf-citationLow Power In-Memory Implementation of Ternary Neural Networks with Resistive RAM-Based Synapse
Laborieux, A., Bocquet, M., Hirtzlin, T., Klein, J., Diez, L. H., Nowak, E., Vianello, E., Portal, J. M., & Querlioz, D. (2020) · IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS) · doi:10.1109/aicas48895.2020.9073877 · Cited by 14
A 0.02% Accuracy Loss Voltage-Mode Parallel Sensing Scheme for RRAM-Based XNOR-Net Application
Zhao, Y., Yu, J., Zhang, D., Hu, Q., Liu, X., Jiang, H., Ding, Q., Han, Z., Cheng, J., Zhang, W., et al. (2022) · IEEE Transactions on Circuits & Systems II Express Briefs · doi:10.1109/tcsii.2022.3157767 · Cited by 12
“Among them, RRAM has the advantages of a simple process and a high HRS/LRS ratio, and it could perform multi-bit MAC calculation through Ohm’s law and Kirchhoff’s law [4]–[6].”
MethodSTAR: Synthesis of Stateful Logic in RRAM Targeting High Area Utilization
Wang, F., Luo, G., Sun, G., Zhang, J., Kang, J., Wang, Y., Niu, D., & Zheng, H. (2020) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2020.3015465 · Cited by 13
“Besides digital fashion, RRAM also supports matrix–vector multiplication in the analog fashion, which has been exploited to accelerate convolutional neural networks (CNNs) [30]–[32] and binary neural networks (BNNs) [33].”
MethodVCAM: Variation Compensation through Activation Matching for Analog Binarized Neural Networks
Kim, J., Lee, C., Kim, J., Kim, Y., Hwang, C. S., & Choi, K. (2019) · International Symposium on Low Power Electronics and Design (ISLPED) · doi:10.1109/islped.2019.8824902 · Cited by 11
“Due to the advantages, BNN has been implemented on many different hardware platforms such as GPU [4], FPGA [5], and ASIC with digital and analog processing elements [6]–[11].”
BackgroundMethodResultInfluentialA Flash-based Current-mode IC to Realize Quantized Neural Networks
Scott, K. R., Lee, C.-Y., Khatri, S. P., & Vrudhula, S. B. K. (2022) · Design, Automation, and Test in Europe · doi:10.23919/date54114.2022.9774539 · Cited by 11
Lee, S., Woo, S. Y., & Lee, J. (2020) · IEEE Access · doi:10.1109/access.2020.3018226 · Cited by 10
“However, it may cause considerable inference accuracy degradation because the threshold of the binary neuron circuit can be different from the threshold value in the algorithms due to the process variation [8], [20].”
BackgroundMethodResultInfluential8T XNOR-SRAM based Parallel Compute-in-Memory for Deep Neural Network Accelerator
Jiang, H., Liu, R., & Yu, S. (2020) · IEEE International Midwest Symposium on Circuits and Systems (MWSCAS) · doi:10.1109/mwscas48704.2020.9184455 · Cited by 14
“There are emerging non-volatile memory based designs reported (e.g. XNOR-RRAM [7]), however, emerging NVMs are generally premature for large-scale manufacturing at this moment.”
BackgroundSelf-citationPartial Sum Quantization for Reducing ADC Size in ReRAM-Based Neural Network Accelerators
Azamat, A., Asim, F., Kim, J., & Lee, J. (2023) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2023.3294461 · Cited by 10
“There were previous attempts to reduce the precision of ADC output [14], [19].”
MethodResultInfluentialIn-memory computing: characteristics, spintronics, and neural network applications insights
Jangra, P., & Duhan, M. (2024) · Multiscale and Multidisciplinary Modeling Experiments and Design · doi:10.1007/s41939-024-00517-0 · Cited by 9
Lou, Q., & Jiang, L. (2018) · IEEE Computer Architecture Letters · doi:10.1109/lca.2018.2882384 · Cited by 15
“We compared BRAWL against the latest binarized CNN accelerator Envision [9], a DRAM-based binarized CNN PIM XNOR-POP [10], a ReRAM-based binarized CNN PIM ReRAM-XNOR [12], a SOTMRAM-based binarized CNN PIM IMCE [11], the GTX1080 GPU, an Intel Xeon D-2141I CPU and a mobile SoC platform Jetson KT1 including a 4-core ARM…”
BackgroundMethodInfluentialQiao, W., Zhang, D., Zhao, Y., Liu, C., Hu, Q., Liu, X., Yang, J., & Lv, H. (2021) · Micromachines · doi:10.3390/mi12040385 · Cited by 9
“[26]. read out, charge-sharing Copyright @ 2019 IEEE. function is realized Moreover, Xiaoyu Sun [20] and our previous study [27] proposed the in-memory X(N)OR logic operations based on novel nonvolatile memory (NVM).”
BackgroundATRIA: A Bit-Parallel Stochastic Arithmetic Based Accelerator for In-DRAM CNN Processing
Shivanandamurthy, S. M., Thakkar, I. G., & Salehi, S. A. (2021) · IEEE Computer Society Annual Symposium on VLSI · doi:10.1109/isvlsi51109.2021.00045 · Cited by 9
“To address this problem, several prior works have explored processing-in-memory (PIM) designs based on the emerging non-volatile memory (NVM) crossbar technologies (e.g., ISAAC [10], PRIME [22], XNOR-RRAM [23]) as well as the traditional DRAM technology (e.g., DRISA [1], SCOPE [2], DRACC [21], LACC [3]).”
MethodPhysical Synthesis for Advanced Neural Network Processors
He, Z., Liao, P., Liu, S., Ma, Y., Lin, Y., & Yu, B. (2021) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1145/3394885.3431625 · Cited by 12
“Recent work have demonstrated that neural network computation can be implemented in various emerging non-volatile memories (NVM), such as RRAM [38, 39], STT-MRAM [40, 41], PCM [42], and memristor [43].”
BackgroundLow-Power Adiabatic/MTJ LIM-Based XNOR/XOR Synapse and Neuron for Binarized Neural Networks
Nasab, M. T., & Thapliyal, H. (2023) · IEEE International Conference on Nanotechnology (IEEE-NANO) · doi:10.1109/nano58406.2023.10231249 · Cited by 8
“In the literature, much research has been conducted and several designs have been proposed [7-14].”
BackgroundMethodProgrammable Non-Volatile Memory Design Featuring Reconfigurable In-Memory Operations
Jao, N., Ramanathan, A. K., Sengupta, A., Sampson, J., & Narayanan, V. (2019) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas.2019.8702534 · Cited by 8
“However, for associative support, the search key needs to access the memory by column and requires a complementary input to perform an XNOR-based search as shown in [6][14].”
MethodTernary Compute-Enabled Memory using Ferroelectric Transistors for Accelerating Deep Neural Networks
Thirumala, S. K., Jain, S., Gupta, S. K., & Raghunathan, A. (2020) · Design, Automation, and Test in Europe · doi:10.23919/date48585.2020.9116495 · Cited by 11
“To address this issue, various works have proposed in-memory computing, wherein computations are performed within the memory array, eliminating the memory access overheads associated with traditional von-Neumann architectures [6-15].”
BackgroundDT2CAM: A Decision Tree to Content Addressable Memory Framework
Rakka, M., Fouda, M. E., Kanj, R. N., & Kurdahi, F. J. (2023) · IEEE Transactions on Emerging Topics in Computing · doi:10.1109/tetc.2023.3261748 · Cited by 14
Computing in-memory with cascaded spintronic devices for AI edge
Bian, Z., Liu, B., & Cai, H. (2023) · Computers & Electrical Engineering · doi:10.1016/j.compeleceng.2023.108767 · Cited by 8
Zanotti, T., Puglisi, F. M., & Pavan, P. (2021) · IEEE International Reliability Physics Symposium proceedings · doi:10.1109/irps46558.2021.9405103 · Cited by 8
“…in place of the ADC and the accumulator, in practice when considering sufficiently large models the size of a single layer could be smaller than the size of the crossbar array, and thus a higher precision is still required to accumulate the results among the different crossbars [24], [25].”
BackgroundHuang, S., Peng, X., Jiang, H., Luo, Y., & Yu, S. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2003.09739 · Cited by 17
“Prior work has suggested that even with fine-tuned linear/nonlinear reference levels, ADC quantization (less than full precision of the partial sum of VMM) will introduce a slight accuracy loss [9].”
BackgroundSelf-citationXu, W., Luo, J., Fu, B., Fu, Z., Wang, K., Su, C., Huang, Q., & Huang, R. (2024) · IEEE Electron Devices Technology & Manufacturing Conference (EDTM) · doi:10.1109/edtm58488.2024.10512190 · Cited by 7
“However, the existing CIM-based QNN schemes are generally resistive type, which suffer from inevitable static DC power consumption during computing [7], and require complementary two storage elements for each signed weight of QNN, limiting the improvement of area-and energy-efficiency [3-6].”
BackgroundMethodNasab, M. T., Amirany, A., Moaiyeri, M. H., & Jafari, K. (2024) · AEU - International Journal of Electronics and Communications · doi:10.1016/j.aeue.2024.155284 · Cited by 7
A Parallel RRAM Synaptic Array Architecture for Energy-Efficient Recurrent Neural Networks
Yin, S., Sun, X., Yu, S., Seo, J.-S., & Chakrabarti, C. (2018) · SiPS ... design and implementation - IEEE Workshop on Signal Processing Systems · doi:10.1109/sips.2018.8598445 · Cited by 8
“The customized bit-cell design for implementing XNOR function in RRAM array [23].”
BackgroundMethodInfluentialSelf-citationSingle RRAM Cell-based In-Memory Accelerator Architecture for Binary Neural Networks
Oh, H., Kim, H., Kang, N., Kim, Y., Park, J., & Kim, J. (2021) · IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS) · doi:10.1109/aicas51828.2021.9458444 · Cited by 7
“For example, XNOR-RRAM [1] used two vertically adjacent cells to represent a binary weight and Sun et al. [2] used two horizontally adjacent cells to represent a binary weight.”
BackgroundMethodSynaptic Tunnel Field-Effect Transistors for Extremely-Low-Power Operation
Lee, J. W., Woo, J. S., & Choi, W. Y. (2022) · IEEE Electron Device Letters · doi:10.1109/led.2022.3179173 · Cited by 7
“Offline-training-based binary neural network (BNN) synaptic architectures possess good potential to achieve these goals owing to their low computing overhead and simple bit-counting [2].”
BackgroundMethodTowards Efficient RRAM-based Quantized Neural Networks Hardware: State-of-the-art and Open Issues
Krestinskaya, O., Zhang, L., & Salama, K. N. (2022) · IEEE International Conference on Nanotechnology (IEEE-NANO) · doi:10.1109/nano54668.2022.9928590 · Cited by 7
“Work CMOS, RRAM Tiles RRAM bits Bits W,I,O∗1 Network Training, bits Binary Neural Networks [3] Ti/HfO2 - 1 1b, 1b, 1b 4 FC QAT, 32b [4] 45nm 128x128 1 1b, 1b, 1b 6 Conv, 3 FC PTQ, 32b [5] 65nm - 1 1b, -, 4 Conv, 3FC QAT, 6b Higher precision networks (mostly fixed-point arithmetics) [6] 65nm 128x128 2 up to 10b, 8b (s-…”
MethodVertical AND-Type Flash Synaptic Cell Stack for High-Density and Reliable Binary Neural Networks
Kim, J., Im, J., Oh, S., Shin, W., Jung, G., Lee, S., & Lee, J. (2024) · IEEE Electron Device Letters · doi:10.1109/led.2024.3401399 · Cited by 8
“These advantages have led to active research on various synaptic devices, such as resistive random-access memory (RRAM) [6], [7], [8], magnetic RAM (MRAM) [9], and flash memory [10], [11], [12] for hardware-based BNNs.”
BackgroundCho, K., Malhotra, A., & Gupta, S. K. (2023) · IEEE Journal on Exploratory Solid-State Computational Devices and Circuits · doi:10.1109/jxcdc.2023.3320677 · Cited by 7
“Similar to [7], the analog summation of I BL can be expressed using the number of bit-cells with I H ( = a ) and I L ( = N − a ) Now, by replacing I H and I L with the corresponding bit values of + 1 and − 1, respectively, the MAC output ( OUT M ) of the column is obtained as The I BL of the column…”
BackgroundMethodFlash Memory for Synaptic Plasticity in Neuromorphic Computing: A Review
Im, J., Pak, S., Woo, S. Y., Shin, W., & Lee, S. (2025) · Biomimetics · doi:10.3390/biomimetics10020121 · Cited by 8
“…the overall count of input transistors relative to the previous design in [82]. feature of this design is that every synapses within a single synaptic string share the four input transistors, thereby decreasing the overall count of input transistors relative to the previous design in [82].”
Securing Binarized Neural Networks via PUF-Based Key Management in Memristive Crossbar Arrays
Rajendran, G., Basak, D., Deb, S., & Chattopadhyay, A. (2024) · IEEE Embedded Systems Letters · doi:10.1109/les.2024.3422294 · Cited by 6
“Implementing BNN in RRAM crossbars has the advantage that a current comparator can replace the resource-hungry analog-to-digital converter [1], [2].”
InfluentialMaximizing Parallel Activation of Word-Lines in MRAM-Based Binary Neural Network Accelerators
Ahn, D., Oh, H., Kim, H., Kim, Y., & Kim, J. (2021) · IEEE Access · doi:10.1109/access.2021.3121011 · Cited by 7
“3a) [9].”
Ahmed, S. T., Hefenbrock, M., Münch, C., & Tahoori, M. B. (2022) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2022.3205872 · Cited by 6
“The weight matrix splitting and current accumulation as proposed in [49] is used in this article.”
MethodADC-Free ReRAM-Based In-Situ Accelerator for Energy-Efficient Binary Neural Networks
Kim, H., Jung, Y., & Kim, L. (2022) · IEEE Transactions on Computers · doi:10.1109/tc.2022.3224800 · Cited by 6
“Therefore, previous works [19], [21] adopted a complementary cell and word-line structure and assigned two adjacent cells in the bit-line direction for each weight element.”
BackgroundMethodInfluentialHigh Speed Binary Neural Network Hardware Accelerator Relied on Optical NEMS
Gholami, Y., Marvi, F., Ghorbanloo, R., Eslami, M. R., & Jafari, K. (2023) · IEEE Transactions on Nanotechnology · doi:10.1109/tnano.2023.3343618 · Cited by 6
Malhotra, A., Wang, C., & Gupta, S. K. (2024) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1109/asp-dac58780.2024.10473947 · Cited by 6
“To alleviate this problem, computing-in-memory (CiM) has been explored, where a part of the DNN computations (such as vector-matrix multiplications) are performed within the memory array itself, overcoming the von-Neumann bottleneck [1].”
MethodAn All-digital Compute-in-memory FPGA Architecture for Deep Learning Acceleration
Li, Y., Li, X., Shen, H., Fan, J., Xu, Y., & Huang, K. (2024) · ACM Transactions on Reconfigurable Technology and Systems · doi:10.1145/3640469 · Cited by 5
“References [ 8 , 30 , 34 , 35 , 39 , 40 ] adopt the 1T1R RRAM structure, which is composed of a transistor and a resistor.”
MethodPark, D., Park, H., Yeo, I., Kim, J., Lee, S., Lee, S., Shin, H., Pae, S.-I., Fan, D., Kung, J., et al. (2025) · IEEE Internet of Things Journal · doi:10.1109/jiot.2025.3633487 · Cited by 11
“…optimized for binary and ternary CIM operations, where each bit-cell stores either 1 or 1.58 bit, providing a solution tailored to low-precision computing for BNNs and TNNs. Sun et al. [10] use two 1T1R bit-cells to represent each weight in XNOR -based MAC operations, enabling parallel processing.”
InfluentialMitigating Adversarial Attack for Compute-in-Memory Accelerator Utilizing On-chip Finetune
Huang, S., Jiang, H., & Yu, S. (2021) · IEEE Non-Volatile Memory Systems and Applications Symposium (NVMSA) · doi:10.1109/nvmsa53655.2021.9628739 · Cited by 5
Self-citationLee, S., Kim, H., Yoo, H.-N., Kwon, D., & Lee, J. (2022) · Neurocomputing · doi:10.1016/j.neucom.2022.05.030 · Cited by 5
RIBoNN: Designing Robust In-Memory Binary Neural Network Accelerators
Kundu, S., Malhotra, A., Raha, A., Gupta, S. K., & Basu, K. (2022) · IEEE International Test Conference (ITC) · doi:10.1109/itc50671.2022.00061 · Cited by 5
“Table I summarizes the improvement statistics furnished by RiBoNN, which are normalized over the state-of-the-art XNOR-RRAM [5].”
MethodInfluentialMemory-Based Computing for Energy-Efficient AI: Grand Challenges
Karimzadeh, F., Imani, M., Asgari, B., Cao, N., Lin, Y., & Fang, Y. (2023) · IFIP/IEEE International Conference on Very Large Scale Integration · doi:10.1109/vlsi-soc57769.2023.10321880 · Cited by 11
“As an example, Flash-ADC could be used for binary neural network [37].”
MethodBenchmark of RRAM based Architectures for Dot-Product Computation
Peng, X., & Yu, S. (2018) · IEEE Asia Pacific Conference on Circuits and Systems (APCCAS) · doi:10.1109/apccas.2018.8605606 · Cited by 6
“In section III, we describe the conventional 8-bit RRAM and XNOR-RRAM architectures in sequential row-by-row and parallel read-out fashion.”
MethodInfluentialSelf-citationXNOR Operation of Binary Neural Networks Using Nanoelectromechanical Memory Switches
Park, G. T., Lee, J. W., Woo, J. S., & Choi, W. Y. (2024) · IEEE Transactions on Electron Devices · doi:10.1109/ted.2024.3486267 · Cited by 4
“However, owing to the low ON / OFF resistance ratio of memristor-based eNVMs, both the inference accuracy and energy efficiency degrade during XNOR operations [15], [16], [17], [18], [19], [20].”
Kim, D., Jeong, H., Kim, S., Yoon, H., & Lee, K. J. (2025) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2025.3603076 · Cited by 4
“To address the energy and throughput limitations of A-to-D conversion, several quantization techniques [16], [32], [33], [34] have been introduced.”
InfluentialA Fault-Tolerant Framework for Stuck-at Fault Mitigation in Memristor-Based Ternary Neural Networks
You, K., & Li, C. (2025) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2025.3609785 · Cited by 6
“To overcome these challenges, mem-ristors emerged as promising non-volatile storage elements [5], o ff er unique adjustable resistance properties that make them suitable for implementing neural network circuits [6], [7], [8].”
Sparse Vector-Matrix Multiplication Acceleration in Diode-Selected Crossbars
Jao, N., Ramanathan, A. K., Sampson, J., & Narayanan, V. (2021) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2021.3114186 · Cited by 4
Implementing binary neural networks in memory with approximate accumulation
Gupta, S., Imani, M., Zhao, H., Wu, F., Zhao, J., & Simunic, T. (2020) · International Symposium on Low Power Electronics and Design (ISLPED) · doi:10.1145/3370748.3406562 · Cited by 4
“In the end, we also compare BitNAP with highly optimized in-memory binary vector-matrix multipliers, XNOR-RRAM [20] and XIMA [15].”
MethodInfluentialData and Computation Reuse in CNNs Using Memristor TCAMs
de Moura, R. F., de Lima, J. P. C., & Carro, L. (2022) · ACM Transactions on Reconfigurable Technology and Systems · doi:10.1145/3549536 · Cited by 5
“PRIME [8] Conv-RAM [6] XNOR-RRAM [54] ✓ ✗ ✓ n.”
BackgroundLu, S. K., & Sengupta, A. (2022) · IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS) · doi:10.1109/aicas54282.2022.9869925 · Cited by 4
“For instance, leveraging insights from reduced precision implementation of neural networks (which has, in turn, resulted in the recent wave of specialized “In-Memory” hardware accelerators using CMOS [23, 24] and post-CMOS technologies [25] that are highly optimized for single-bit state representations), a recent pape…”
BackgroundPossibilities and Limitations of Memristor Crossbars for Neuromorphic Computing
Telminov, O. A., & Gornev, E. (2022) · Scientific School Dynamics of Complex Networks and their Applications (DCNA) · doi:10.1109/dcna56428.2022.9923302 · Cited by 4
“Thus, in [45] values 1 are used for them, 3-bit sense amplifiers are used to calculate the partial sum of the weights; the improvement in parameters is shown compared to the conventional RRAM synaptic architecture with sequential row-by-row read-out.”
MethodADC-Less Reprogrammable RRAM Array Architecture for In-Memory Computing
Dongre, A., Boro, B., & Trivedi, G. (2023) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2023.3319578 · Cited by 4
“To deal with the drawbacks of the Von Neumann architecture-based systems, various in-memory computing architectures have been proposed to alleviate the memory bottleneck problem [3], [4], [5], [6], [7].”
BackgroundXBM: A Crossbar Column-wise Binary Mask Learning Method for Efficient Multiple Task Adaption
Zhang, F., Li, Y., Meng, J., Cao, Y. K., Seo, J.-S., & Fan, D. (2022) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1109/asp-dac52403.2022.9712508 · Cited by 5
Self-citationPower and Area-Efficient XNOR-AND Hybrid Binary Neural Networks Using TFT-Type Synaptic Device
Lee, I.-S., Kim, H., Park, M., Hwang, J., Koo, R., Kim, J., & Lee, J. (2022) · IEEE Electron Device Letters · doi:10.1109/led.2022.3229321 · Cited by 4
“Since standard BNN utilizes only − 1 and 1 weights and activations, the vector-matrix multiplication (VMM) operation can be replaced with the XNOR pop-count action [13].”
BackgroundMethodAnalysis of Low Area Digital Up/Down Clipping Counter for Digital In-Memory Computing
Lee, S., Lee, G., Ahn, S., & Na, T. (2025) · IEEE Access · doi:10.1109/access.2025.3531982 · Cited by 3
“Most state-of-the-art approaches utilize two bit-cells per synapse to store one weight value, enabling the BNN’s XNOR operation [11], [12], [13], [14].”
Implementation of Ternary Neural Networks Based on Ferroelectric Capacitors
Yeom, J., Woo, J. S., Lee, J. W., Park, C., Park, J., Ryu, M., & Choi, W. Y. (2025) · IEEE Electron Device Letters · doi:10.1109/led.2025.3634334 · Cited by 3
“3c and 3d show the sensing pass rate of 20 and 40 synapses obtained through Monte Carlo (MC) simulations, which represents the percentage of accurate sensing results in the voltage sense amplifier (VSA) [17].”
Analysis of VMM computation strategies to implement BNN applications on RRAM arrays
Parmar, V., Kingra, S. K., Negi, S., & Suri, M. (2023) · APL Machine Learning · doi:10.1063/5.0139583 · Cited by 3
A Mixed-Signal Quantized Neural Network Accelerator Using Flash Transistors
Scott, K. R., Lee, C.-Y., Khatri, S. P., & Vrudhula, S. B. K. (2023) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2023.3336248 · Cited by 3
“The design in [22] achieves an energy efficiency of 141.2 TOPS/W (Tera-OPerations per Second per Watt) when performing BNN classification of Cifar-10, which is 2 .”
BackgroundMethodInfluentialHuang, S., Sun, X., Peng, X., Jiang, H., & Yu, S. (2022) · ACM Transactions on Design Automation of Electronic Systems · doi:10.1145/3500929 · Cited by 4
Self-citationPower-Efficient Implementation of Ternary Neural Networks in Edge Devices
Molina, M., Mendez, J., Morales, D. P., Castillo, E., López‐Vallejo, M., & Cuéllar, M. P. (2022) · IEEE Internet of Things Journal · doi:10.1109/jiot.2022.3172843 · Cited by 6
“Other researchers have focused on creating efficient hardware (HW) devices for the implementation of ANNs [9] or the adaptation of general devices, such as graphics processing units (GPUs) and field-programmable gate arrays (FPGAs) with this purpose [8].”
BackgroundSpin Orbit Torque-based Crossbar Array for Error Resilient Binary Convolutional Neural Network
Danouchi, K., Prenat, G., & Anghel, L. (2022) · IEEE Latin American Test Symposium (LATS) · doi:10.1109/lats57337.2022.9936951 · Cited by 7
Ahmed, S. T., & Tahoori, M. B. (2023) · IEEE Design and Test · doi:10.1109/mdat.2023.3270126 · Cited by 4
“The fault model simulates the active wordlines of the complementary exclusive-NOR (XNOR) bitcell [12], as inactive wordlines do not impact the overall current sum.”
BackgroundAGNI: In-Situ, Iso-Latency Stochastic-to-Binary Number Conversion for In-DRAM Deep Learning
Shivanandamurthy, S. M., Vatsavai, S. S., Thakkar, I. G., & Salehi, S. A. (2023) · International Symposium on Quality Electronic Design (ISQED) · doi:10.1109/isqed57927.2023.10129301 · Cited by 3
“Among these, some PIM designs are based on the emerging non-volatile memory (NVM) crossbar technologies (e.g., ISAAC [2], PRIME [6], XNOR-RRAM [7]), some are based on the traditional DRAM technology (e.g., DRISA [8], SCOPE [9], DRACC [10], LACC [11]), and some are based on the SRAM technology (e.g., [12] [13] [14]).”
Self-Writing xnor Magnetic Random Access Memory for Binary Neural Network Acceleration
Park, J. (2023) · IEEE Transactions on Magnetics · doi:10.1109/tmag.2023.3315283 · Cited by 5
“Partial sums from sub-arrays are ultimately summed up to generate the final sum for binary activation [41], [42], [43].”
BackgroundParallel Stateful Logic in RRAM: Theoretical Analysis and Arithmetic Design
Wang, F., Luo, G., Sun, G., Zhang, J., Huang, P. C., & Kang, J. (2019) · IEEE International Conference on Application-specific Systems, Architectures and Processors (ASAP) · doi:10.1109/asap.2019.000-8 · Cited by 3
“Besides digital fashion, some works [19], [20] accelerate neural networks acceleration in the analog fashion.”
Lee, S., Yeom, G., Hwang, J., Kim, H., Yoo, H.-N., Park, B., & Lee, J. (2021) · IEEE Journal of the Electron Devices Society · doi:10.1109/jeds.2021.3123632 · Cited by 4
SSR: A Skeleton-based Synthesis Flow for Hybrid Processing-in-RRAM Modes
Wang, F., Sun, G., & Luo, G. (2021) · doi:10.1109/iccad51958.2021.9643493 · Cited by 3
VPU-CIM: A 130nm, 33.98 TOPS/W RRAM based Compute-In-Memory Vector Co-Processor
J, C. M., Rayapati, V., Rao, N., & Suri, M. (2024) · 2024 IEEE International Symposium on Circuits and Systems (ISCAS) · doi:10.1109/iscas58744.2024.10558155 · Cited by 2
“Although [13] [15] achieve better energy efficiency, these implementations are 1,1 (I, W) configuration (1-bit width).”
BackgroundAn Extremely Low-voltage Floating Gate Artificial Neuron
Scott, K. R., & Khatri, S. P. (2023) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas46773.2023.10182158 · Cited by 2
“The work of [8] proposed an analog PIM inference accelerator that uses a resistive RAM (RRAM) crossbar array which operates in current mode.”
MethodHardware Platform-Aware Binarized Neural Network Model Optimization
Vo, Q. H., Asim, F., Alimkhanuly, B., Lee, S., & Kim, L.-W. (2022) · Applied Sciences · doi:10.3390/app12031296 · Cited by 5
“Moreover, RRAM was also utilized in the paper [19] with a synaptic architecture (XNOR-RRAM) that can implement equivalent XNOR and bit-counting operations.”
BackgroundMethodOpen-RIMC: Open-Source RRAM-based IMC Co-Processor with Reconfigurable Logic Mapping
Parmar, V., Ray, A., Chithambara Moorthii, J., Mishra, R., Verma, D., Pandey, D., & Suri, M. (2023) · IEEE International Conference on Nanotechnology (IEEE-NANO) · doi:10.1109/nano58406.2023.10231247 · Cited by 3
“Parameters This work DATE 2018 [18] FRAI 2021 [19] ISSCC 2020 [20] Technology Node (nm) 130 65 40 130 Array size 8x16 64x128 128x128 784x100,100x10 In,W,O 1,1,4 1,1,3 1,1,3 1,T,1 Energy Efficiency (TOPS/W) 5.”
RRAM IMC based Efficient Analog Carry Propagation and Multi-bit MVM
Chithambara Moorthii, J., Mourya, M. V., Bansal, H., Verma, D., & Suri, M. (2024) · IEEE Electron Devices Technology & Manufacturing Conference (EDTM) · doi:10.1109/edtm58488.2024.10511694 · Cited by 2
Jeong, I., & Park, J.-E. (2024) · Design, Automation, and Test in Europe · doi:10.23919/date58400.2024.10546655 · Cited by 2
Binary Weight Multibit Activation Quantization for Compute-in-Memory CNN Accelerators
Zhou, W., Liu, Z., Ren, Y., & Wong, N. (2025) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2025.3595830 · Cited by 4
“The first strategy, such as CIM-BNN [14] and XNOR-RRAM [13], focuses on extremely quantized networks with binary weights and binary activations [binary weight binary activation (BWBA), see Fig.”
InfluentialA Novel Computing Paradigm for MobileNetV3 using Memristor
Li, J., Liu, Z., Ma, L., Sham, C., & Fu, C. (2025) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn64981.2025.11228482 · Cited by 2
1S-1R array: Pure-memristor circuit for binary neural networks
Zhang, Z., Ge, Z., Wei, Y., Cheng, X., Xie, G., & Liu, G. (2021) · Microelectronic Engineering · doi:10.1016/j.mee.2021.111697 · Cited by 2
Thakuria, N., Elangovan, R., Thirumala, S. K., Raghunathan, A., & Gupta, S. K. (2022) · Frontiers in Nanotechnology · doi:10.3389/fnano.2022.905407 · Cited by 3
“For example, binary networks such as XNOR-RRAM (Sun et al., 2018) and XNOR-SRAM (Yin et al., 2020) feature large parallel vector-matrix multiplication capability, but suffer from low accuracies due to aggressive quantization of weights and inputs to binary values.”
A cross-layer approach to cognitive computing
Saha, G., Wang, C., Raghunathan, A., & Roy, K. (2022) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1145/3489517.3530642 · Cited by 2
“Efficient perception and navigation with obstacle detection for drones have been demonstrated based on neuromorphic computing models with the fusion of event-based and optical sensing [16].”
Accelerating Inference on Binary Neural Networks with Digital RRAM Processing
Vieira, J., Giacomin, E., Qureshi, Y. M., Zapater, M., Tang, X., Kvatinsky, S., Atienza, D., & Gaillardon, P. (2020) · IFIP advances in information and communication technology · doi:10.1007/978-3-030-53273-4_12 · Cited by 2
“Xiaoyu et al. [29] proposed a parallel XNOR-RRAM array using CSAs.”
MethodArea Efficient Pattern Representation of Binary Neural Networks on RRAM
Wang, F., Luo, G., Sun, G., Wang, Y., Niu, D., & Zheng, H. (2021) · Journal of Computer Science and Technology · doi:10.1007/s11390-021-0906-y · Cited by 2
“To cope with this problem, previous studies propose two mainstream representations, the pos-neg representation [8,9] and the XNOR representation [10,11] .”
MethodInfluentialA Reliability-Concerned Compute-in-Memory Behavior Model for Convolutional Neural Network
Cheng, K., Song, J., Zhang, X., He, Y., Wang, R., & Wang, Y. (2021) · IEEE International Symposium on the Physical and Failure Analysis of Integrated Circuits (IPFA) · doi:10.1109/ipfa53173.2021.9617344 · Cited by 3
Optimizing for In-Memory Deep Learning With Emerging Memory Technology
Wang, Z., Luo, T., Goh, R. S. M., Zhang, W., & Wong, W. (2023) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2023.3285488 · Cited by 3
“Sun et al. [25] used the single-bit cell to execute the XNOR operations in XNOR - Net.”
MethodA Study on Adversarial Attacks and Defense Method on Binarized Neural Network
Dinh, V.-N., Bui, N.-M., Nguyen, V.-T., Nguyen, K.-S., Duong, Q., & Trinh, Q. (2022) · International Conference on Advanced Technologies for Communications · doi:10.1109/atc55345.2022.9943040 · Cited by 2
“Several BNN macro implementation has been proposed recently, such as ReRAM BNN macro in [20], [21], and STT-MRAM BNN macro in [22].”
MethodLeveraging Stochasticity for In Situ Learning in Binarized Deep Neural Networks
Pyle, S. D., Sapp, J. D., & DeMara, R. F. (2019) · Computer · doi:10.1109/mc.2019.2906133 · Cited by 3
“…sign activation function) and parallel operation (using two input lines per input bit and two single-transistor/single-resistive-element (1T1R) cells per weighted input to compute parallel bitwise XNOR operations and then using a Sense Amplifier output to realize the sign activation function) [6].”
BackgroundMethodInfluentialTraining of quantized deep neural networks using a magnetic tunnel junction-based synapse
Greenberg-Toledo, T., Perach, B., Hubara, I., Soudry, D., & Kvatinsky, S. (2021) · Semiconductor Science and Technology · doi:10.1088/1361-6641/ac251b · Cited by 5
MTJ-Based Hardware Synapse Design for Quantized Deep Neural Networks.Preprint
Greenberg-Toledo, T., Perach, B., Soudry, D., & Kvatinsky, S. (2019) · arXiv (Cornell University) · Cited by 2
Le, M.-S., Pham, T.-N., Nguyen, T.-D., & Chang, I. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.09962 · Cited by 2
“SOTA works showed that the precision of the partial sums significantly affects the accuracy of the computed BNNs [2], [10], [11], [16].”
An In-Memory Power Efficient Computing Architecture with Emerging VGSOT MRAM Device
Sarkar, M. R., Chowdhury, S. S., Walling, J. S., & Yi, Y. (2024) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas58744.2024.10557835 · Cited by 1
A Novel Current Comparator Enabling Large RRAM Crossbars for BNNs and PUFs
Rajendran, G., Basak, D., Deb, S., Wang, S., & Chattopadhyay, A. (2024) · IFIP/IEEE International Conference on Very Large Scale Integration · doi:10.1109/vlsi-soc62099.2024.10767833 · Cited by 1
“The generated I k is digitized with a comparator or analog-to-digital converter (ADC) [1], [8] as shown in Fig.1.”
A Compute-in-Memory Macro Based on Complementary 2T2C FeRAM Cell for BNNs
Li, J., Xie, M., & Xiong, S. (2024) · IEEE International Conference on Solid-State and Integrated Circuit Technology · doi:10.1109/icsict62049.2024.10831829 · Cited by 1
Farias, M., & Kung, H. T. (2025) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas56072.2025.11043969 · Cited by 1
“Past works address digital partial sum accumulation to increase accuracy [5], [17], [18].”
Metadata
- Authors
- Sun, X., Yin, S., Peng, X., Liu, R., Seo, J.-S., & Yu, S.
- Affiliations
- Arizona State University (US)
- Venue
- 2018 Design, Automation & Test in Europe Conference & Exhibition (DATE)
- Publisher
- IEEE
- Published
- 2018-03
- Pages
- 1423-1428
- OpenAlex
- W2798554798
- Cited by
- 246
- References
- 20
- Citations by year
- 2026: 7 · 2025: 17 · 2024: 33 · 2023: 34 · 2022: 40 · 2021: 35 · 2020: 42 · 2019: 30 · 2018: 8
- FWCI
- 45.74
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- Design, Automation, and Test in Europe (publishedVersion)
- Funders
- Canadian Institute for Advanced Research
- Topics
- Physical Sciences › Engineering › Electrical and Electronic Engineering › Advanced Memory and Neural Computing
- Physical Sciences › Engineering › Electrical and Electronic Engineering › Low-power high-performance VLSI design
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Keywords
- resistive random access memory, binary neural networks, synaptic structure, array partitioning, energy-efficient neuromorphic hardware, convolutional neural networks, fully connected layers, MNIST, CIFAR-10
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.