US 12,001,945 B2
Event driven mathematical engine and method
David Schie, San Jose, CA (US); Sergey Gaitukevich, San Jose, CA (US); Peter Drabos, San Jose, CA (US); and Andreas Sibrai, San Jose, CA (US)
Assigned to AIStorm Inc., Houston, TX (US)
Filed by David Schie, San Jose, CA (US); Sergey Gaitukevich, San Jose, CA (US); Peter Drabos, San Jose, CA (US); and Andreas Sibrai, San Jose, CA (US)
Filed on Apr. 26, 2019, as Appl. No. 16/396,570.
Claims priority of provisional application 62/663,121, filed on Apr. 26, 2018.
Prior Publication US 2019/0332929 A1, Oct. 31, 2019
Int. Cl. G06N 3/06 (2006.01); G06N 3/048 (2023.01); G06N 3/049 (2023.01); G06N 3/065 (2023.01); H01L 27/146 (2006.01)
CPC G06N 3/065 (2023.01) [G06N 3/048 (2023.01); G06N 3/049 (2013.01); H01L 27/14612 (2013.01)] 7 Claims
OG exemplary drawing
 
1. An event driven device comprising a neural network whose input layer comprises pixel charge data coupled through transfer gates to a plurality of analog multiply and sum devices, the multiply and add devices comprising level shifts which are used to adjust a reset level of the analog multiply and sum devices;
wherein after the processing of each exposure by the analog multiply and sum devices, the reset level of the analog multiply and sum devices are set to the potential of the exposure such that in a subsequent exposure a spill and fill circuit will transfer no charge to the multiply and sum devices unless a pixel value has changed.