Secure Convolutional Neural Networks (CNN) accelerator
AssigneeTexas Instruments Incorporated
InventorsMihir Narendra Mody, Veeramanikandan Raju, Chaitanya Ghone, Deepak Poddar
Life of this patent
Estimating the term…
Abstract
A CNN based-signal processing includes receiving of an encrypted output from a first layer of a multi-layer CNN data. The received encrypted output is subsequently decrypted to form a decrypted input to a second layer of the multi-layer CNN data. A convolution of the decrypted input with a corresponding decrypted weight may generate a second layer output, which may be encrypted and used as an encrypted input to a third layer of the multi-layer CNN data.
Claims20
A system, comprising:
- an integrated circuit device comprising a convolutional neural network (CNN) hardware engine and a first memory coupled to the CNN hardware engine;
- and
- a second memory coupled to the CNN hardware engine, the second memory configured to store an encrypted weight of the CNN,
- wherein the CNN hardware engine is configured to:
- of an image frame;
- receive the encrypted weight of the CNN that is stored in the second memory;
- decrypt the encrypted weight to generate a decrypted weight of the CNN;
- perform a convolution based on the input feature and the decrypted weight to generate an output;
- and
- store the output in the first memory.
An integrated circuit device, comprising:
- a first memory;
- and
- a convolution neural network (CNN) hardware engine coupled to the first memory, the CNN hardware engine configured to:
- receive an input feature of an input frame;
- receive an encrypted weight of the CNN from a second memory external to the integrated circuit device;
- decrypt the encrypted weight to generate a decrypted weight of the CNN;
- perform a convolution based on the input feature and the decrypted weight to generate an output;
- and
- store the output in the first memory.
A non-transitory computer readable medium storing program instructions that, when executed by an integrated circuit device, cause the integrated circuit device to:
- receive an input feature of an input feature for a convolutional neural network (CNN);
- receive an encrypted weight of the CNN from a first memory external to the integrated circuit device;
- decrypt the encrypted weight to generate a decrypted weight of the CNN;
- perform a convolution based on the input feature and the decrypted weight to generate an output;
- and
- store the output in a second memory internal to the integrated circuit.
Dependencies are inferred from the claim text.