US 11,853,857 B2
Secure convolutional neural networks (CNN) accelerator
Mihir Narendra Mody, Bangalore (IN); Veeramanikandan Raju, Bangalore (IN); Chaitanya Ghone, Pune (IN); and Deepak Poddar, Bangalore (IN)
Assigned to Texas Instruments Incorporated, Dallas, TX (US)
Filed by Texas Instruments Incorporated, Dallas, TX (US)
Filed on Jun. 2, 2020, as Appl. No. 16/889,853.
Application 16/889,853 is a continuation of application No. 15/730,316, filed on Oct. 11, 2017, granted, now 10,706,349.
Claims priority of application No. 201741018440 (IN), filed on May 25, 2017.
Prior Publication US 2020/0293859 A1, Sep. 17, 2020
This patent is subject to a terminal disclaimer.
Int. Cl. G06N 3/04 (2023.01); G06N 3/088 (2023.01); H04L 9/40 (2022.01); G06N 3/045 (2023.01)
CPC G06N 3/04 (2013.01) [G06N 3/045 (2023.01); G06N 3/088 (2013.01); H04L 63/0435 (2013.01); H04L 63/0464 (2013.01)] 21 Claims
OG exemplary drawing
 
1. A method of signal processing comprising:
decrypting, using an input feature decryption block of an integrated circuit device, an input feature to form a decrypted input to a layer of a multi-layer convolutional neural network (CNN) using a first cryptographic key;
receiving, using a weight kernel decryption block of the integrated circuit device, an encrypted weight that corresponds to the decrypted input;
decrypting, using the weight kernel decryption block of the integrated circuit device, the received encrypted weight using a second cryptographic key;
performing, using a CNN hardware core of the integrated circuit device, a convolution on the decrypted input and the decrypted weight to generate a layer output; and
encrypting, using an output feature encryption block of the integrated circuit device, the layer output to form an encrypted layer output.