ABSTRACT
One main fault-tolerant method for a neural network accelerator based on resistive random access memory crossbars is the programming-based method, which is also known as write-and-verify (W-V). In the basic W-V scheme, all devices in crossbars are programmed repeatedly until they are close enough to their targets, which costs huge overhead. To reduce the cost, we optimize the W-V scheme by proposing a probabilistic termination criterion on a single device and a systematic optimization method on multiple devices. Furthermore, we propose a joint algorithm that assists the novel W-V scheme by incremental retraining, which further reduces the W-V cost. Compared to the basic W-V scheme, our proposed method improves the accuracy by 0.23% for ResNet18 on CIFAR10 with only 9.7% W-V cost under variation with σ = 1.2.
- S. Yu. Resistive Random Access Memory (RRAM). Morgan & Claypool, 2016.Google ScholarCross Ref
- A. Shafiee et al. ISAAC: A convolutional neural network accelerator with in-situ analog arithmetic in crossbars. In ISCA, pages 14--26, 2016.Google Scholar
- P. Chi et al. PRIME: A novel processing-in-memory architecture for neural network computation in ReRAM-based main memory. In ISCA, pages 27--39, 2016.Google Scholar
- A. Grossi et al. Fundamental variability limits of filament-based RRAM. In IEDM, pages 4.7.1--4.7.4, 2016.Google ScholarCross Ref
- L. Chen et al. Accelerator-friendly neural-network training: Learning variations and defects in RRAM crossbar. In DATE, pages 19--24, 2017.Google ScholarDigital Library
- G. Charan et al. Accurate inference with inaccurate rram devices: Statistical data, model transfer, and on-line adaptation. In DAC, pages 1--6, 2020.Google ScholarCross Ref
- Z. Meng et al. Digital offset for RRAM-based neuromorphic computing: A novel solution to conquer cycle-to-cycle variation. In DATE, pages 1078--1083, 2021.Google Scholar
- B. Liu et al. Vortex: Variation-aware training for memristor X-bar. In DAC, pages 1--6, 2015.Google Scholar
- Y. Zhu et al. Statistical training for neuromorphic computing using memristor-based crossbars considering process variations and noise. In DATE, pages 1590--1593, 2020.Google Scholar
- Y. Long et al. Design of reliable DNN accelerator with un-reliable ReRAM. In DATE, pages 1769--1774, 2019.Google Scholar
- Y. Geng et al. An on-chip layer-wise training method for RRAM based computing-in-memory chips. In DATE, pages 248--251, 2021.Google Scholar
- C. Ma et al. Go unary: A novel synapse coding and mapping scheme for reliable ReRAM-based neuromorphic computing. In DATE, pages 1432--1437, 2020.Google Scholar
- K. Higuchi et al. 50nm hfo2 ReRAM with 50-times endurance enhancement by set/reset turnback pulse & verify scheme. In SSDM, pages 1011--1012, 09 2011.Google Scholar
- J. K. Yoon et al. A 40nm 64Kb 56.67 TOPS/W read-disturb-tolerant compute-in-memory/digital RRAM macro with active-feedback-based read and in-situ write verification. In ISSCC, pages 404--406, 2021.Google Scholar
- P. Yao et al. Fully hardware-implemented memristor convolutional neural network. Nature, 577(7792):641--646, 2020.Google ScholarCross Ref
- S. Yu et al. Investigating the switching dynamics and multilevel capability of bipolar metal oxide resistive switching memory. APL, 98(10):103514, 2011.Google ScholarCross Ref
- G. L. Zhang et al. An efficient programming framework for memristor-based neuromorphic computing. In DATE, pages 1068--1073, 2021.Google Scholar
- Y. Cai et al. Long live TIME: Improving lifetime for training-in-memory engines by structured gradient sparsification. In DAC, pages 1--6, 2018.Google ScholarDigital Library
Recommendations
Fast-Write Resistive RAM (RRAM) for Embedded Applications
Editor's note:Especially for microcontroller and mobile applications, embedded nonvolatile memory is an important technology offering to reduce power and provide local persistent storage. This article describes a new resistive RAM device with fast write ...
Impact of oxygen composition of ZnO metal-oxide on unipolar resistive switching characteristics of Al/ZnO/Al resistive RAM (RRAM)
Display Omitted Effect of oxygen content on electrical performance of ZnO RRAM.Demonstrate the optimal oxygen content in ZnO film for RRAM application.Propose the resistance switching model of ZnO RRAM at different oxygen content. This study ...
Comparison of oxygen vacancy defects in crystalline and amorphous Ta2O5
Display Omitted In the crystalline phase, the 2-fold coordinated intra-layer vacancy is the lowest cost vacancy.The 3-fold intra-layer vacancy and the 2-fold inter-layer vacancy are higher cost defects, and form shallower levels.In the amorphous phase, ...
Comments