Hybrid in-memory computing architecture for the training of deep neural networks

Keyphrases

• Deep Neural Network 100%
• In-memory Computing 100%
• Phase Change Random Access Memory (PCRAM) 42%
• Efficient Inference 28%
• Memory Efficiency 28%
• Model Size 28%
• Training Deep Neural Networks 14%
• Small Fraction 14%
• Novel Solutions 14%
• Hardware Accelerator 14%
• Hardware Platform 14%
• 32-bit 14%
• Endurance Limit 14%
• Extended Period 14%
• Weight Update 14%
• Low-precision 14%
• Inference Accelerator 14%
• Training Accelerator 14%
• Von Neumann Architecture 14%
• Inference Accuracy 14%
• Bit Precision 14%
• Inference Model 14%
• Efficient Deep Neural Networks 14%
• Deep Neural Network Training 14%
• Benchmark Tasks 14%
• Floating-point 14%
• U-Net 14%
• CIFAR-10 14%
• Negligible Effects 14%
• Comparable Accuracy 14%
• Training Results 14%
• Representation Technique 14%
• Memory Saving 14%
• Weight Representation 14%
• Baseline Network 14%
• Software Accuracy 14%
• Temporal Drift 14%

Computer Science

• Computer Architecture 100%
• Deep Neural Network 100%
• Trained Network 25%
• Hardware Accelerator 25%
• Hardware Platform 25%
• Von Neumann Architecture 25%
• Small Fraction 25%
• Inference Model 25%
• Benchmark Task 25%
• Floating Point 25%
• Residual Neural Network 25%
• Neural Network Training 25%