Data write-back system

Abstract

The invention provides a data write-back system. The system comprises at least one write-back module; the write-back module comprises two volatile storage units, a logic control unit, a data transmission unit and a cache control unit; the cache control unit is connected with an activator; the write-back system can receive a plurality of data at the same time, so that parallelism can be greatly improved, and a high data outgoing rate can be realized; the logic control unit is connected with the two volatile storage units; the two volatile storage units are connected with the data transmission unit; when the number of the write-back modules is larger than 1, the data transmission units are connected in series; the cache control unit is used for receiving data from the activator; and the logic control unit is used for controlling the two volatile storage units to read data from the cache control unit and write data into the data transmission unit in a ping-pong read-write mode; and sincethe data can be read and written at the same time, the continuous output of the data is ensured, and the write-back efficiency of the data is improved.

Description

technical field [0001] The invention relates to the technical field of deep learning, in particular to a data write-back system. Background technique [0002] In addition to intensive operations in the current convolutional neural network, frequent reading and writing of memory is also a prominent feature of the convolutional neural network. In order to ensure that the computing resources in the accelerator can perform powerful calculations, the input data must be written back to the memory in a timely and accurate manner for the next layer of convolution operations. [0003] The calculation speed of the neural network accelerator is very fast. The data output by the activator cannot be accurately and quickly written back into the memory, and the calculation unit of the neural network accelerator will enter an idle state because it cannot read the calculation data from the memory, thus affecting the overall performance. [0004] Therefore, it is necessary to provide a nove...

AI Technical Summary

Problems solved by technology

[0003] The calculation speed of the neural network accelerator is very fast, and the data output by the activator cannot be accurately and quickly written back into the memory, and the computing unit of the neural network accelerator will enter an idle state because it cannot read the calculation data from the memory, thus affecting the overall performance

Images