CNN encryption using XOR Gate for Hardware Optimization
Ki Beom Lee, Sumin Lee, Sunghwan Joo, Hong Keun Ahn, Young Seok Jung, Seong Ook Jung · 2021
Artificial intelligence networks have been researched in many fields such as computer vision, health care, and military service. Convolutional neural network (CNN) is one of the basic neural networks that uses convolutional operations as the basis to train data and perform desired application. However, important parameters used in CNN applications suffer from security issues. Thus, the need for data protection is increasing. The traditional approach of storing training data and weight parameters encrypted in external memory has the disadvantage of large hardware area. In this paper, the CNN-based encrypting method using XOR gates is proposed to reduce hardware resources by allowing encryption and decryption to operate in a single module. The proposed encrypting method reduces hardware usage by more than 100 times compared to the conventional encrypting method.