Bit Error Robustness for Energy-Efficient DNN Accelerators

2020/06/24 by David Stutz, Stutz, David, Nandhini Chandramoorthy +5 · 8 citations
Computer Science · Engineering · Mathematics · #Advanced Memory and Neural Computing #Advanced Neural Network Applications #Adversarial Robustness in Machine Learning #Algorithm #Artificial intelligence #Artificial neural network #Clipping (morphology) #Computer Vision and Pattern Recognition (cs.CV) #Computer hardware #Computer science #Cryptography and Security (cs.CR) #Efficient energy use #Electrical engineering #Energy consumption #Engineering #FOS: Computer and information sciences #Hardware Architecture (cs.AR) #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Quantization (signal processing) #Robustness (evolution) #Static random-access memory #Voltage #cs.AR #cs.CR #cs.CV #cs.LG #stat.ML

paper · pdf · doi:10.48550/arxiv.2006.13977

published in arXiv (Cornell University) (Cornell University)

openalex publication_date 2020/06/24 · arxiv created 2021/04/09 · arxiv updated 2021/04/12 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Deep neural network (DNN) accelerators received considerable attention in past years due to saved energy compared to mainstream hardware. Low-voltage operation of DNN accelerators allows to further reduce energy consumption significantly, however, causes bit-level failures in the memory storing the quantized DNN weights. In this paper, we show that a combination of robust fixed-point quantization, weight clipping, and random bit error training (RandBET) improves robustness against random bit errors in (quantized) DNN weights significantly. This leads to high energy savings from both low-voltage operation as well as low-precision quantization. Our approach generalizes across operating voltages and accelerators, as demonstrated on bit errors from profiled SRAM arrays. We also discuss why weight clipping alone is already a quite effective way to achieve robustness against bit errors. Moreover, we specifically discuss the involved trade-offs regarding accuracy, robustness and precision: Without losing more than 1% in accuracy compared to a normally trained 8-bit DNN, we can reduce energy consumption on CIFAR-10 by 20%. Higher energy savings of, e.g., 30%, are possible at the cost of 2.5% accuracy, even for 4-bit DNNs.

Citations

• Intriguing properties of neural networks
• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
• Gradient-based learning applied to document recognition
• Learning with a Strong Adversary
• Resiliency of Deep Neural Networks under Quantization
• Fixed Point Quantization of Deep Convolutional Networks
• Deep neural networks are robust to weight binarization and other non-linear distortions
• DoReFa-Net: Training Low Bitwidth Convolutional Neural Networks with Low Bitwidth Gradients
• Quantized Neural Networks: Training Neural Networks with Low Precision Weights and Activations
• Lets keep it simple, Using simple architectures to outperform deeper and\n more complex architectures
• Training Quantized Nets: A Deeper Understanding
• Improved Regularization of Convolutional Neural Networks with Cutout
• Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
• Bit Fusion: Bit-Level Dynamically Composable Architecture for Accelerating Deep Neural Networks
• Quantizing deep convolutional networks for efficient inference: A whitepaper
• A Survey on Methods and Theories of Quantized Neural Networks
• Robustness May Be at Odds with Accuracy
• On the Effectiveness of Interval Bound Propagation for Training Verifiably Robust Models
• Fixup Initialization: Residual Learning Without Normalization
• Batch Normalization is a Cause of Adversarial Vulnerability
• Improving Robustness Without Sacrificing Accuracy with Patch Gaussian Augmentation
• LQ-Nets: Learned Quantization for Highly Accurate and Compact Deep Neural Networks
• Efficient Processing of Deep Neural Networks: A Tutorial and Survey
• Provable defenses against adversarial examples via the convex outer adversarial polytope
• Learning Multiple Layers of Features from Tiny Images

Cited by

• MoRS: An Approximate Fault Modelling Framework for Reduced-Voltage SRAMs

Related

• Random and Adversarial Bit Error Robustness: Energy-Efficient and Secure DNN Accelerators
• MATIC: Learning Around Errors for Efficient Low-Voltage Neural Network Accelerators
• EDEN: Enabling Energy-Efficient, High-Performance Deep Neural Network Inference Using Approximate DRAM
• MAx-DNN: Multi-Level Arithmetic Approximation for Energy-Efficient DNN Hardware Accelerators
• Exploring Fault-Energy Trade-offs in Approximate DNN Hardware Accelerators