US 12,288,153 B2
Schedule-aware tensor distribution module
Gautham Chinya, Sunnyvale, CA (US); Huichu Liu, Santa Clara, CA (US); Arnab Raha, San Jose, CA (US); Debabrata Mohapatra, San Jose, CA (US); Cormac Brick, San Francisco, CA (US); and Lance Hacking, Spanish Fork, UT (US)
Assigned to Intel Corporation, Santa Clara, CA (US)
Filed by Intel Corporation, Santa Clara, CA (US)
Filed on Jan. 10, 2024, as Appl. No. 18/408,716.
Application 18/408,716 is a continuation of application No. 16/456,707, filed on Jun. 28, 2019, granted, now 11,907,827.
Prior Publication US 2024/0220785 A1, Jul. 4, 2024
Int. Cl. G06N 3/063 (2023.01); G06F 9/38 (2018.01); G06F 9/448 (2018.01); G06F 9/50 (2006.01); G06N 5/04 (2023.01)
CPC G06N 3/063 (2013.01) [G06F 9/3814 (2013.01); G06F 9/3877 (2013.01); G06F 9/4498 (2018.02); G06F 9/5027 (2013.01); G06N 5/04 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method, comprising:
extracting, by a drain module of a neural network accelerator from one or more processing elements of the neural network accelerator, data elements in an output tensor of a first layer of a neural network in, the data elements computed by the one or more processing elements in an execution of the first layer, the data elements organized in a first sequence;
reorganizing, by the drain module, the data elements based on a configuration descriptor register for a second layer of the neural network, the configuration descriptor register providing one or more signals to one or more multiplexers in the drain module for reorganizing the data elements, the reorganized data elements having a second sequence that is different from the first sequence, the second layer subsequent to the first layer in the neural network;
writing, by the drain module, the data elements into a memory of the neural network accelerator in accordance with the second sequence; and
performing, by the one or more processing elements, an execution of the second layer of the neural network by using the data elements read from the memory in accordance with the second sequence.