Convolution neural network model computing device and computing method

Abstract

The invention provides a convolution neural network model computing device and computing method. The computing device comprises a physically non-clonable module for generating a response r' accordingto a predetermined excitation c'; s multiplication-accumulation calculation module configured to perform a multiplication and accumulation calculation of the corresponding fuzzy weight values w'0 to w' i and the corresponding input data of the trained convolution neural network model based on the response r' of the physically non-clonable module, and obtaining the results of multiplication and accumulation, wherein the fuzzy weight value is not equal to at least one of the original weight values w0 to wi corresponding to the trained convolution neural network model, and the obtained multiplication and accumulation calculation result is the same as the original weight value of the trained convolution neural network model and the multiplication and accumulation calculation result of the corresponding input data. The computing device and the computing method of the invention can carry out intellectual property protection for the CNN model itself and have low overhead.

Description

technical field [0001] The invention relates to the technical field of information security, in particular to a convolutional neural network model computing device and computing method. Background technique [0002] In recent years, advances in technology have contributed to the rapid growth of system design complexity. In the context of a globalized economy, external economic drivers and market forces have led to more design starting points, shorter design cycles and greater time-to-market pressure. These trends have simultaneously led to the widespread use of third-party intellectual property (IP). However, privacy attacks on intellectual property, such as unauthorized use, cloning, and tampering, not only reduce profits and market share, but also cause damage to brand reputation. Therefore, the protection of intellectual property rights is extremely necessary. [0003] Convolutional neural network (CNN) is a feed-forward artificial neural network. CNN uses data convolu...

AI Technical Summary

Problems solved by technology

[0006] However, the existing CNN-related IP protection is limited to the protection of circuits and FPGA designs. These technologies can be used in the IP protection of CNN hardware accelerators, but they cannot be used in the IP protection of the CNN model itself, because the CNN model is not hardware

In addition, traditional data protection methods are implemented through encryption, when the encrypted data is decrypted for use, the decrypted data stored in the memory may be stolen by attackers, and the decryption process will affect the performance of the accelerator

Therefore, the traditional IP protection method for hardware cannot be directly used for the protection of CNN model parameters, and the traditional data encryption method will bring the performance loss of the hardware accelerator. The research on the IP protection of the CNN model is still blank.

Images