2021 · 61 citations · 10 references
ComputingComputer EngineeringComputer ScienceEdge ComputingMachine LearningSystem-on-chip DesignDeep LearningEdge ArchitectureEdge DevicesHardware AccelerationMulticore Processor DesignNetwork-on-chip DesignActivation FunctionsConvolutional Neural NetworksDeep Learning AcceleratorsMany-core ProcessorsNeural Network QuantizationOff-chip Memory Interface DesignOff-chip Memory SystemsOn-chip Memory Design
To deploy convolutional neural networks (CNNs) on edge devices efficiently, most existing CNN processors were built on quantized CNNs to optimize the inference operations. However, three issues (Fig. 9.2.1) have not been well addressed: 1) Duplicate weights in each kernel after quantization yielding repetitive multiplications; 2) a huge number of unnecessary MACs caused by ReLU activation functions; 3) frequent off-chip memory access in residual blocks.
10