Hardware Trojan attack method for on-chip interconnection structure of reconfigurable accelerator

Abstract

The invention discloses a hardware Trojan attack method for an on-chip interconnection structure of a reconfigurable accelerator, and the method comprises the following steps: 1) enabling a hardware Trojan triggering module to generate a triggering signal through employing configuration information or 20-bit image data as triggering input; comparing the configuration information or the image datawith a trigger condition predefined by an attacker to determine whether the trigger signal is valid or not; 2) when the trigger signal is valid, enabling the hardware Trojan load module to modify a control selection signal of a selector in the on-chip interconnection structure; and 3) controlling a selection signal to be modified, and changing an original data path and an operation circuit structure in the on-chip interconnection structure to cause wrong reasoning calculation. According to the hardware Trojan attack method for the reconfigurable accelerator on-chip interconnection structure provided by the invention, the reconfigurable on-chip interconnection structure is attacked, and an original data path and an operation circuit structure in the on-chip interconnection structure are changed, so that the classification precision of the neural network accelerator is reduced.

Description

technical field [0001] The invention belongs to the field of hardware security, and in particular relates to a hardware Trojan horse attack method aimed at the on-chip interconnection structure of a reconfigurable accelerator. Background technique [0002] Convolutional neural networks have been widely used to deal with various scene tasks, such as medical diagnosis, stock market prediction, semantic segmentation, face recognition, etc. For network reasoning, neural networks need to process hundreds of megabytes of weights and billions of computing operations. In order to speed up the reasoning process, neural network accelerators have been proposed to improve processing speed, which has become a hot spot in the research field and has received extensive attention. Accelerator designs based on different platforms have been proposed, including central processing units (CPUs), graphics processing units (GPUs), application-specific integrated circuits (ASICs), and field-program...

AI Technical Summary

Problems solved by technology

Therefore, implementing a hardware Trojan horse attack on the on-chip interconnect structure of a neural network accelerator will cause serious consequences to the accelerator, thus affecting the entire artificial intelligence processing system

Images