US 12,136,031 B2
System and method for increasing utilization of dot-product based neural network accelerator
Ali Shafiee Ardestani, Santa Clara, CA (US); and Joseph Hassoun, Los Gatos, CA (US)
Assigned to Samsung Electronics Co., Ltd., Yongin-si (KR)
Filed by Samsung Electronics Co., Ltd., Suwon-si (KR)
Filed on May 18, 2023, as Appl. No. 18/320,133.
Application 18/320,133 is a continuation of application No. 16/900,852, filed on Jun. 12, 2020, granted, now 11,687,764.
Claims priority of provisional application 63/011,599, filed on Apr. 17, 2020.
Prior Publication US 2023/0289584 A1, Sep. 14, 2023
This patent is subject to a terminal disclaimer.
Int. Cl. G06F 7/22 (2006.01); G06F 7/24 (2006.01); G06F 7/544 (2006.01); G06F 9/50 (2006.01); G06N 3/063 (2023.01); G06N 5/04 (2023.01)
CPC G06N 3/063 (2013.01) [G06F 7/24 (2013.01); G06F 7/5443 (2013.01); G06F 9/5027 (2013.01); G06N 5/04 (2013.01)] 19 Claims
OG exemplary drawing
 
1. A method of flattening channel data of an input feature map in an inference system, the method comprising:
extracting first values from a plurality of pixel values of the input feature map;
rearranging second values corresponding to an overlap region of a kernel of the inference system from among the first values into a single one-dimensional vector;
identifying a first group of consecutive values from among the rearranged second values of the single one-dimensional vector for supplying to a first dot-product circuit of the inference system; and
identifying a second group of consecutive values from among the rearranged second values of the single one-dimensional vector for supplying to a second dot-product circuit of the inference system.