In cryptography, an attack can be performed by injecting one or several faults into a device, thus, disrupting its functional behavior. Commonly used techniques to inject faults consists of introducing perturbations in the source voltage, clock frequency, temperature, or irradiating with a laser beam, etc.
Several security mechanisms can be implemented to detect any stress applied on the device. For example, Secure-IC Digital Sensor, as a universal fault injection sensor, is able to detect multiple kind of attacks (temperature, voltage, clock frequency, laser exposure, EM exposure) or the Secure-IC Active Shield operating as a smart dome over the sensitive part of the chip. Secure-IC s Cyber Escort Unit monitors for abnormal operating conditions step by step escorting the program execution, allowing real-time detection of zero-day attacks. Other IPs, not destined to be sensors, like Secure-IC PUF or Secure-IC TRNG, are able to detect alteration on the chip environment, thanks to their embedded Health Tests.
All these opportunistic sensors have output status or alarm alerts about any abnormal environmental conditions. But, for classical systems it is difficult to interpret all these data, which can reach several hundreds or thousands of bits when a matrix of Secure-IC Digital Sensors is embedded, and more so to react properly and quickly in case of a proven attack.
This is why it is necessary to instate a security module, as a security headquarter, handling all security events and statuses. Smart Monitor has the ability to centralize all information concerning the health of the environment. In addition, Smart Monitor can be interfaced with customer s sensors (analog or digital) to increase the sources of information and specific data.
Furthermore, respecting the Security Policy in a given environment where the chip is operating, Smart Monitor uses Artificial Intelligence to classify faults. This method excludes false-positives. For instance, voltage variations can be usual in some electrical installations, but it may be identified as an attack by some sensors. However, after a learning phase, with training data distinguishing characteristics of such faults as a non-attack scenario, and according to its Security Model, Smart Monitor will not raise an alarm in this case. So, a phase of machine learning aims to determine the security model specific to the environment based on an acquisition campaign, depending on device categories, geographic areas, technology nodes, etc. It is, then, able to make a real-time diagnostic of the environment and apply a well-adapted Security Policy in case of a detected attack. It also profiles the anatomy of an attack (nature, temporality, locality, intensity, attack phase, etc.).