Back to results
US 12,238,126 B2Granted
Systems and methods for learning-based high-performance, energy-efficient, and secure on-chip communication design framework
AssigneeThe George Washington University
InventorsKe Wang, Hao Zheng, Ahmed Louri
Life of this patent
Estimating the term…
Abstract
Systems and methods are disclosed for improving on-chip security, while minimizes the latency and cost of security techniques to improve system-level performance and power simultaneously. The framework uses machine learning algorithms, such as an artificial neural network (ANN), for runtime attack detection with higher accuracy. Further, a learning-based attack mitigation method using deep reinforcement learning is disclosed, where the method may be used to isolate the malicious components and to optimize network latency and energy-efficiency.
Claims17
An on-chip router comprising:
- an on-chip input port receiving input data packets;
- an on-chip output port transmitting output data packets;
- an on-chip dynamic error detection and correction circuit having a decoder at said on-chip input port to decode and detect error at the input data packets, and an encoder at said on-chip output port to apply an error correction to the encoded data packets and output the output data packets;
- an on-chip artificial neural network module;
- an on-chip controller, wherein the controller instructs the artificial neural network module to detect malicious circuit modifications in the router, wherein said malicious circuit modifications comprise transient faults in the input data packets;
- and
- a bypass channel that directly connects said on-chip input port to said on-chip output port and bypass said on-chip dynamic error detection and correction circuit,
- wherein the artificial neural network module detects malicious circuit modifications by recognizing abnormal network behaviors caused by the transient faults, and
- wherein the artificial neural network module applies one or more machine learning algorithms to detect the malicious circuit modifications.
6 more elements
A system comprising:
- one or more processing cores;
- and
- one or more on-chip routers, wherein each of the on-chip routers is comprised of an on-chip input port connected to said one or more processing cores and one or more adjacent on-chip routers, and an on-chip output port connected to said one or more processing cores and one or more adjacent on-chip routers, each of said one or more on-chip routers further comprising a on-chip dynamic error detection and correction circuit, a bypass channel that directly connects said on-chip input port to said on-chip output port and bypass said on-chip dynamic error detection and correction circuit, an artificial neural network module, and a controller;
- wherein the controller instructs the artificial neural network module to detect malicious circuit modifications in the one or more on-chip routers, wherein said malicious circuit modifications comprise transient faults,
- wherein the artificial neural network module detects malicious circuit modifications by recognizing abnormal network behaviors caused by the transient faults, and
- wherein the artificial neural network module applies one or more machine learning algorithms to detect the malicious circuit modifications.
3 more elements
A method for on-chip network security comprising:
- receiving one or more data packets at an input port at the artificial neural network module of one or more on-chip routers, wherein each on-chip router is comprised of a controller and wherein the artificial neural network is trained to detect malicious circuit modifications including transient faults in the received data packets at the input port;
- analyzing runtime attributes of the data packets at the artificial neural network module;
- identifying malicious circuit modifications from the runtime attributes of the data packets;
- bypassing at a bypass channel, one of the one or more on-chip routers by connecting the input port to an output port of the one or more on-chip routers;
- and
- transmitting an instruction from the controller to use the bypass channel to isolate routers in which the malicious circuit modifications are detected to directly connect the input port with an output port and bypass the router,
- wherein the artificial neural network module detects malicious circuit modifications by recognizing abnormal network behaviors caused by the transient faults, and
- wherein the artificial neural network module applies one or more machine learning algorithms to detect the malicious circuit modifications.
5 more elements
Dependencies are inferred from the claim text.