An Energy-Efficient 3D Cross-Ring Accelerator with 3D-SRAM Cubes for Hybrid Deep Neural Networks

Keyphrases

• Energy Efficient 100%
• 3D SRAM 100%
• Hybrid Deep Neural Network 100%
• Deep Neural Network Accelerator 60%
• Router 40%
• CNN Accelerator 40%
• On chip 20%
• Deep Neural Network 20%
• Interconnection Architectures 20%
• Promising Solutions 20%
• Computation-intensive 20%
• Resource Utilization 20%
• VGG16 20%
• 3D Stacking 20%
• Edge Devices 20%
• Embedded Applications 20%
• Stacking Technology 20%
• Overall Energy Efficiency 20%
• Convolutional Layer 20%
• Intensive Workloads 20%
• DRAM Access 20%
• Fully Connected Layer 20%
• Flexible Connection 20%
• Dynamic Workload 20%
• Data Movement 20%
• Recurrent Layer 20%
• On-device 20%
• PE Utilization 20%
• Workload Distribution 20%

Computer Science

• Energy Efficient 100%
• Deep Neural Network 100%
• Convolutional Neural Network 28%
• Energy Efficiency 14%
• Neural Network Model 14%
• Resource Utilisation 14%
• Intensive Computation 14%
• Embedded Application 14%
• Dynamic Workload 14%
• Convolution Layer 14%
• VGG-16 14%
• Workload Distribution 14%
• Fully Connected Layer 14%