By – Dr. Saswat Kumar Ram, Asst. Professor, Dept. of Electronics and Communications Engineering, SRM University -AP
Detecting Hardware Trojans:
A Hardware Trojan is malicious circuitry inserted into a chip during design or manufacturing.AI/ML can learn the expected behaviour of a legitimate chip and identify deviations.Techniques such as neural networks, SVMs, random forests, and autoencoders can be trained using circuit characteristics such as:Power consumption, timingbehaviour, Switching activity, Gate-level features, Test responses.
Detecting Side-Channel Attacks:
Chips can unintentionally reveal secret information throughPower consumption, Electromagnetic emissions, Execution timing, Cache behaviour.An attacker may analyse power traces to recover an AES encryption key.AI can learn the relationship between power/EM traces and cryptographic operations and identify suspicious patterns.
Protecting Cryptographic Keys:
AI can monitor access to sensitive areas containingEncryption keys, Authentication credentials, Secure memory, Trusted execution environments. If the access pattern becomes abnormal, the system can automatically block access, isolate the security region, Zeroize sensitive keys, generate an alert, Switch to a secure operating mode. This is particularly interesting for AI-enabled secure SoCs.
Detecting Fault-Injection Attacks:
Attackers may deliberately introduceVoltage glitches, Clock glitches, Laser faults, Electromagnetic disturbances to make a chip behave incorrectly.AI can monitor voltage, clock, timing and execution characteristics and distinguish between normal variations and malicious fault injection.
Detecting Abnormal Memory/Bus Access:
AI can learn normal access patterns CPU, Cache, Memory, Bus, Peripherals. If a processor normally accesses a protected memory region only occasionally but suddenly generates thousands of accesses, the AI model can flag this as suspicious.This can help in detecting data exfiltration, unauthorized memory access, privilege escalation, malicious firmware, Bus attacks.
AI-Based Runtime Security:
This is one of the strongest research directions.Instead of checking the chip only during manufacturing, AI can provide runtime security. This is very essential in current scenario.
Federated Learning for Chip Security:
Multiple devices can learn about attacks without sending their sensitive raw hardware data to a central server.The devices share model updates rather than sensitive data.This could be particularly useful for IoT, automotive chips, medical devices, and edge computing.
