A Reconfigurable Streaming Deep Convolutional Neural Network Accelerator for Internet of Things
This provides an energy-efficient hardware solution for enabling intelligent image detection on resource-constrained IoT devices, though it appears incremental as it builds on existing accelerator designs with specific optimizations.
The paper tackles the challenge of implementing convolutional neural networks (CNNs) for image detection on Internet of Things (IoT) devices by proposing a streaming hardware accelerator that optimizes energy efficiency and supports arbitrary convolution window sizes, achieving 152 GOPS peak throughput and 434 GOPS/W energy efficiency at 350mW in a 65nm prototype.
Convolutional neural network (CNN) offers significant accuracy in image detection. To implement image detection using CNN in the internet of things (IoT) devices, a streaming hardware accelerator is proposed. The proposed accelerator optimizes the energy efficiency by avoiding unnecessary data movement. With unique filter decomposition technique, the accelerator can support arbitrary convolution window size. In addition, max pooling function can be computed in parallel with convolution by using separate pooling unit, thus achieving throughput improvement. A prototype accelerator was implemented in TSMC 65nm technology with a core size of 5mm2. The accelerator can support major CNNs and achieve 152GOPS peak throughput and 434GOPS/W energy efficiency at 350mW, making it a promising hardware accelerator for intelligent IoT devices.