CNN accelerator data access method and system
A data access and accelerator technology, applied in the field of calculation and computing, can solve the problems of limited on-chip resources and memory bandwidth acceleration, and achieve the effect of reducing high transmission bandwidth pressure, reducing memory bandwidth and power consumption, and reducing local storage overhead.
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0047] FPGA has powerful parallel processing capability and abundant logic resources. For example, DSP can perform multiply-accumulate operations with higher computing precision, which is very suitable as a CNN implementation platform.
[0048] like figure 1 As shown, a CNN accelerator data access method provided by an embodiment of the present invention includes steps S1 to S8 as described below.
[0049] S1. Preset training parameters, obtain an input feature map and store it in a buffer area.
[0050] It should be noted that the training parameters include the size and weight of the input feature map, the weight can be obtained by training the neural network, and the input feature maps of different channels are stored in the buffer.
[0051] S2. Perform judgment processing on the input feature map in the buffer area, obtain prefetched data through register array processing, and perform cyclic processing on the prefetched data to obtain post-sequence data
[0052] The pro...
Embodiment 2
[0082] A CNN accelerator data access system provided by an embodiment of the present invention includes:
[0083] Data cache module: used to preset training parameters, obtain input feature maps and store them in the cache area;
[0084] Data preprocessing module: used to judge and process the input feature map in the buffer area, obtain prefetched data through register array processing, and cyclically process the prefetched data to obtain the sequential data;
[0085] Convolution calculation module: used for convolution calculation processing of sequential data to obtain convolution data;
[0086] Data post-processing module: used to perform batch normalization, RELU activation function activation, quantization, pooling and FIFO buffer processing on the convolution data in turn to obtain the output results;
[0087] Judgment output module: used to judge the number of processing layers of the output result to obtain the target coordinates.
[0088] As will be appreciated by ...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More