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 201–247 of 247
Raha, A., Thirumala, S. K., Gupta, S. K., & Raghunathan, V. (2025) · ACM Transactions on Design Automation of Electronic Systems · doi:10.1145/3771845 · Cited by 2
“This involves the requirement of two storage elements (2 bit-cells / weight) [67, 78].”
InfluentialFault-tolerant Neuromorphic Computing with Functional ATPG for Post-manufacturing Re-calibration
Ahmed, S. T., & Tahoori, M. B. (2022) · IEEE VLSI Test Symposium (VTS) · doi:10.1109/vts52500.2021.9794185 · Cited by 2
MVSTT: A Multi-Value Computation-in-Memory based on Spin-Transfer Torque Memories
Jafari, A., Mayahinia, M., Ahmed, S. T., Münch, C., & Tahoori, M. B. (2022) · Euromicro Conference on Digital System Design (DSD) · doi:10.1109/dsd57027.2022.00052 · Cited by 2
“We considered the XNOR implementation proposed in [33] for the 1-bit binarized NN.”
MethodFlash: A “Forgotten” Technology in VLSI Design
Khatri, S. P., Vrudhula, S. B. K., Abusultan, M., Bharathi, K., Chu, S.-W., Lee, C., Scott, K. R., Singh, G., & Wagle, A. (2022) · Springer eBooks · doi:10.1007/978-3-031-16344-9_3 · Cited by 1
Enabling Neuromorphic Computing for Artificial Intelligence with Hardware-Software Co-Design
Li, B., Zhong, D., Chen, X., & Liu, C. (2023) · Artificial intelligence · doi:10.5772/intechopen.111963 · Cited by 1
Exploring Bit-Level Sparsity for Partial Sum Quantization in Computing-In-Memory Accelerator
Bai, J., Sun, S., & Kang, W. (2023) · IEEE Non-Volatile Memory Systems and Applications Symposium (NVMSA) · doi:10.1109/nvmsa58981.2023.00021 · Cited by 2
“PSQ is a simple and effective approach to reduce the resolution but most of previous reports only apply PSQ to binary neural networks and bring down the resolution to 3- or 4-bit [14].”
MethodLow-Latency, Energy-Efficient In-DRAM CNN Acceleration with Bit-Parallel Unary Computing
Thakkar, I. G., Shivanandamurthy, S. M., & Salehi, S. A. (2023) · Springer eBooks · doi:10.1007/978-3-031-19568-6_14 · Cited by 1
A Variation-Aware Binary Neural Network Framework for Process Resilient In-Memory Computations
Le, M.-S., Pham, T.-N., Nguyen, T.-D., & Chang, I. (2024) · Electronics · doi:10.3390/electronics13193847 · Cited by 8
“SOTA works showed that the precision of the partial sums significantly affects the accuracy of the computed BNNs [2], [10], [11], [16].”
PUF-based Lightweight Authentication for Binarized Neural Networks
Rajendran, G., Deb, S., & Chattopadhyay, A. (2024) · IEEE Asia Pacific Conference on Circuits and Systems (APCCAS) · doi:10.1109/apccas62602.2024.10808901 · Cited by 1
Fast Simulation Method for Analog Deep Binarized Neural Networks
Lee, C., Kim, J., Kim, J., Kim, J.-H., Hwang, C. S., & Choi, K. (2019) · International SoC Design Conference · doi:10.1109/isocc47750.2019.9078516 · Cited by 1
“The work in [5-7] uses eNVM (embedded Non-Volatile Memory) to implement multiplications in analog and drastically enhance energy-efficiency.”
MethodEmbracing the Unreliability of Memory Devices for Neuromorphic Computing
Bocquet, M., Hirtzlin, T., Klein, J., Nowak, E., Vianello, E., Portal, J., & Querlioz, D. (2020) · IEEE International Reliability Physics Symposium proceedings · doi:10.1109/irps45951.2020.9128346 · Cited by 2
Exploiting Process Variations to Protect Machine Learning Inference Engine from Chip Cloning
Huang, S., Peng, X., Jiang, H., Luo, Y., & Yu, S. (2021) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas51556.2021.9401659 · Cited by 1
Self-citationNeuromorphic applications using MOx-based memristors
Brivio, S., & Vianello, E. (2022) · Elsevier eBooks · doi:10.1016/b978-0-12-814629-3.00014-3 · Cited by 1
Outstanding Bit Error Tolerance of Resistive RAM-Based Binarized Neural NetworksPreprint
Hirtzlin, T., Bocquet, M., Klein, J., Nowak, E., Vianello, E., Portal, J., & Querlioz, D. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1904.03652 · Cited by 1
Shivanandamurthy, S. M., Thakkar, I. G., & Salehi, S. A. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.03953 · Cited by 3
“Some of these accelerators use memory in the conventional volatile RAM configuration (e.g., SRAM [15], DRAM [5] [12]-[14] based ANN accelerators), whereas some others use the emerging non-volatile memory in the crossbar configuration (e.g., STT-MRAM [18]- [19], ReRAM [2][16][17][20]).”
BackgroundStochastic Computing for Hardware Implementation of Binarized Neural NetworksPreprint
Hirtzlin, T., Penkovsky, B., Bocquet, M., Klein, J., Portal, J., & Querlioz, D. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1906.00915
“The low memory requirements of BNNs (one bit by synapse), as well as the fact that they do not require any multiplication, makes them extremely adapted for inference hardware [2], [16], [22], [23].”
BackgroundPyle, S. D. (2019) · Journal of International Crisis and Risk Communication Research
Embracing the Unreliability of Memory Devices for Neuromorphic ComputingPreprint
Bocquet, M., Hirtzlin, T., Klein, J., Nowak, E., Vianello, E., Portal, J., & Querlioz, D. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2007.06238
Merged Logic and Memory Fabrics for AI Workloads
Crafton, B., Spetalnick, S. D., & Raychowdhury, A. (2021) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1145/3394885.3431615
“Recent designs have encoded various numbers of bits in each cell ranging from 8-bit [4], down to 2-bit [15] or 1-bit [16].”
BackgroundTowards efficient deep neural network inference and training for ubiquitous AI
Yuan, G. (2023) · Digital Repository Service (Northeastern University) · doi:10.17760/d20581919
A low power time-domain computing-in-memory micro for binary neural networks
He, Y. (2023) · Digital Repository Service (Northeastern University) · doi:10.17760/d20487159
Hardware acceleration of DNA pattern matching using analog resistive CAMs
Bazzi, J., Sweidan, J., Fouda, M., Kanj, R., & Eltawil, A. M. (2024) · Frontiers in Electronics · doi:10.3389/felec.2023.1343612 · Cited by 1
Chang, C.-Y., Huang, C.-T., Chuang, Y.-C., Chou, K.-C., & Wu, A.-Y. A. (2024) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1109/asp-dac58780.2024.10473797
“Partial-sum quantization-aware training [15]-[16] and nonlinear quantization [13][17] are also explored to overcome the ADC overhead in CIM accelerators with minimal accuracy loss.”
BackgroundMethodRaha, A., Thirumala, S. K., Gupta, S. K., & Raghunathan, V. (2024) · SSRN Electronic Journal · doi:10.2139/ssrn.4782177
Harnessing Valley-Spin Hall Effect in WSe2 for Energy-Efficient Logic-Memory Integration
Gupta, S. K., & Cho, K. (2024) · Spintronics · doi:10.1117/12.3028292
“The first technique utilizes two bit-cells storing complementary bits to encode a single weight [53-56].”
NavCim: Comprehensive Design Space Exploration for Analog Computing-in-Memory Architectures
Park, J., Kim, B. J., & Sung, H. (2024) · International Conference on Parallel Architectures and Compilation Techniques (PACT) · doi:10.1145/3656019.3676946
“Research has been conducted to implement ACiM using different types of memory cells, including SRAM [11, 45], DRAM [23], non-volatile Resistive RAM (ReRAM) [7, 30, 32, 35, 37], and Magnetic RAM (MRAM) [1], focusing on accelerating deep neural network (DNN) models.”
Nonvolatile logic gate and full adder based on tri-terminal oxide resistive switching devices
Cao, J., Ye, J., Wang, T., Ding, Y. S., Cheng, R., Liu, D., & Chen, B. (2024) · Microelectronic Engineering · doi:10.1016/j.mee.2024.112280
FPGA and ASIC accelerators for genome data analysis
Haghi, A. (2024) · UPCommons institutional repository (Universitat Politècnica de Catalunya) · doi:10.5821/dissertation-2117-421134
HorNets: Learning from Discrete and Continuous Signals with Routing Neural NetworksPreprint
Koloski, B., Lavrač, N., & Škrlj, B. (2025) · Qeios · doi:10.32388/22i1po
HorNets: learning from discrete and continuous signals with routing neural networks
Koloski, B., Lavrač, N., & Škrlj, B. (2025) · Machine Learning · doi:10.1007/s10994-024-06673-1
“Binary neural networks were also shown as a promising variant of models that can be computed on specialized hardware such as FPGAs (Moss et al., 2017; Sun et al., 2018).”
Malhotra, A., & Gupta, S. K. (2024) · doi:10.1145/3676536.3697134
“An alternative method, particularly for emerging memories, involves storing each weight across two memory elements [22].”
Kim, D., Jeong, H., Park, K., Kim, S., An, S., & Lee, K. (2025) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas56072.2025.11043653
Ahmed, I., Malhotra, A., & Gupta, S. K. (2025) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1109/dac63849.2025.11133247
“In fact, CiM-driven optimizations sacrificing memory density for robustness have been utilized for other technologies as well [3].”
InfluentialError-Corrected RRAM-Based Compute-in-Memory Accelerator Using Voltage-Clamp Current Mirror
Liu, Y., Zhang, W., Song, X., Zhou, J., Li, J., Feng, H., Lu, W., Peng, C., Zhao, Q., Dai, C., et al. (2025) · IEEE Transactions on Device and Materials Reliability · doi:10.1109/tdmr.2025.3612983
“For example, Sun et al. [24] highlighted this issue, indicating that the pass rate of CSA is only approximately 75% when the BL voltage approaches the decision level, subsequently leading to a reduction in classification accuracy.”
Digital Implementation of Neuromorphic systems using Emerging Memory devices
Hirtzlin, T. (2020) · theses.fr (ABES) · doi:10.70675/7c42d7b9z4001z4424z8a0az6c7cb2079ec8 · Cited by 7
Modeling and Computing-in-Memory Design Based on New Spintronic Devices
Wang, G. (2023) · theses.fr (ABES) · doi:10.70675/51754389zff0fz4195z96c7zc9466a59a7a6
Gautier, A. (2024) · theses.fr (ABES) · doi:10.70675/5827871fz5213z492ez9180z641a7bd710e5
Design of mixed system (RRAM memory & digital) targeting binary neural network hardware accelerator
Jebali, F. (2023) · theses.fr (ABES) · doi:10.70675/b7632edfz3807z4b8fzb462z31a6dcf54161
Diao, Z., Tsukamoto, T., Tohei, T., & Sakai, A. (2026) · IEEE Electron Devices Technology & Manufacturing Conference (EDTM) · doi:10.1109/edtm65772.2026.11496861
“Previous studies have proposed bitwise neural networks that replace analog memristors with digital memristors using binarized resistance states, thereby achieving lower energy consumption without significant loss of computational accuracy [4].”
Energy-Efficient FA-Based Up/Down Clipping Counter for MRAM-Based Digital IMC Systems
Krishna, D. N. M., & Reddy, P. A. K. (2026) · International Conference Trends Material Science and Inventive Materials · doi:10.1109/ictmim68190.2026.11507866
“Sun et al. [12] proposed a resistive XNOR-RRAM synaptic architecture of BNNs, which can be scaled to parallel computation and use less energy.”
Wu, Z.-Y., Ramesh, A. K., Bhukya, V. N., Wei, J., Sheu, S.-S., Hsin, Y.-C., & Tseng, Y. (2026) · IEEE Transactions on Electron Devices · doi:10.1109/ted.2026.3692177
“While prior studies have focused either on device-level behavior [19], [20], [21] or system-level performance [12], [13], [22], [23], [24], [25] a systematic evaluation of how these competing e ff ects translate into variation-aware CIM performance remains limited.”
Park, G. S., Lee, J. W., Baek, S. H., Shin, M. S., Kim, S. H., Yeom, J., & Choi, W. Y. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.6946180
Energy‐Aware Perturbation Optimization for Memristor‐Array Convolutional Neural Networks
Xu, A., Lu, L., Wu, Y., Guo, Y., Li, Y., Wang, Y., & Li, Q. (2026) · Advanced Intelligent Systems · doi:10.1002/aisy.70477 · Cited by 1
PiC-NDM: A 65nm processing-in-CAM-based binary CNN Accelerator for natural disaster management
Harary, Y., Rechef, E., & Yavits, L. (2026) · Future Generation Computer Systems · doi:10.1016/j.future.2026.108823
Bit-sliced quantization for robust memristor-based compute-in-memory
Yu, Y., Yang, J., Ye, Z., Han, S., Wang, B., Wang, S., Wang, H., Qi, X., & Wang, Z. (2026) · npj Unconventional Computing · doi:10.1038/s44335-026-00099-9
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.