US 11,989,638 B2
Convolutional neural network accelerating device and method with input data conversion
Mingrun Liu, Shanghai (CN); Liang Chen, Shanghai (CN); and Xiaopeng Li, Shanghai (CN)
Assigned to THINKFORCE ELECTRONIC TECHNOLOGY CO., LTD, Shanghai (CN)
Appl. No. 16/977,098
Filed by THINKFORCE ELECTRONIC TECHNOLOGY CO., LTD, Shanghai (CN)
PCT Filed Mar. 4, 2019, PCT No. PCT/CN2019/076839
§ 371(c)(1), (2) Date Sep. 1, 2020,
PCT Pub. No. WO2019/170049, PCT Pub. Date Sep. 12, 2019.
Claims priority of application No. 201810181302.1 (CN), filed on Mar. 6, 2018.
Prior Publication US 2021/0019594 A1, Jan. 21, 2021
Int. Cl. G06N 3/04 (2023.01); G06F 9/48 (2006.01); G06F 13/40 (2006.01)
CPC G06N 3/04 (2013.01) [G06F 9/4881 (2013.01); G06F 13/4018 (2013.01)] 13 Claims
OG exemplary drawing
 
1. An input data conversion device, comprising:
a scheduling unit, to generate a control command according to a size/dimension of an input data and a size/stride of a filter, and to control actions of a data moving unit and a row data expansion unit;
the data moving unit, to actively read an input data from a system storage space according to the control command of the scheduling unit;
a row data caching unit, to store the input data that is read in; and
the row data expansion unit, to read out one row of the input data from the row data caching unit each time, and then to expand the row of the input data into one row of data in different filter windows according to the sizes of the filter windows, wherein the row data expansion unit is a multi-channel row data expansion unit, and the multi-channel row data expansion unit expands and generates a plurality of channels of row data in the filter windows, simultaneously.