Circuit structure for accelerating convolutional layer and fully connected layer of neural network

Abstract

The invention belongs to the technical field of integrated circuit designs and particular relates to a circuit structure capable of simultaneously accelerating a convolutional layer and a fully connected layer. The circuit structure comprises five parts: a characteristic / weight pre-fetching module for data reading, a local cache for improving a data reuse rate, a matrix operation unit for realizing matrix multiplication, a temporary data accumulation module for accumulating temporary output results and an output control module in responsible for data rewriting. According to the circuit, a special mapping method is adopted, and operations of the convolutional layer and the fully connected layer are mapped to the matrix operation unit with a fixed size. In the circuit, the memory layout modes of characteristics and weights are adjusted so as to greatly improve the access efficiency of the circuit. Meanwhile, the assembly line mechanism is adopted for scheduling of a circuit module so that all hardware units in each clock period are in a working state, the utilization rate of the hardware units is improved and the working efficiency of the circuit is improved.

Description

technical field [0001] The invention belongs to the technical field of integrated circuit design, and in particular relates to a circuit structure for accelerating convolutional layers and fully connected layers of neural networks. Background technique [0002] In the 1960s, Hubel and others proposed the concept of receptive field through the study of cat visual cortical cells. In the 1980s, Fukushima proposed the concept of neurocognitive machine based on the concept of receptive field, which can be regarded as The first implementation of the convolutional neural network, the neurocognitive machine decomposes a visual pattern into many sub-patterns (features), and then enters the hierarchically connected feature plane for processing. It attempts to model the visual system. It enables it to complete the recognition even when the object is displaced or slightly deformed. [0003] Convolutional Neural Networks are a variant of Multilayer Perceptrons. It was developed from th...

AI Technical Summary

Problems solved by technology

However, more than 90% of the calculations of the convolutional neural network are in the calculation of the convolutional layer and the fully connected layer. Usually, people design a separate acceleration circuit for these two operations, thus introducing redundant chip area.

Images