Researchers at the University of Michigan in Ann Arbor have demonstrated how someone could install a virtually undetectable backdoor on a microprocessor during the fabrication process that could be exploited later to gain complete access to systems running the tampered chips.
The method, detailed in a technical paper innocuously titled ‘Analog Malicious Hardware’, was presented recently at the IEEE Symposium on Security and Privacy. The researchers described it as the first fabrication-time processor attack of its kind and the first to demonstrate an analog attack that is substantially smaller and stealthier than a digital attack.
The attack involves the addition of a single, booby-trapped logic gate to a chip that is ready for fabrication and the use of an extremely stealthy process for triggering changes in the gate’s functionality so it eventually acts in a malicious manner. The attack method is virtually undetectable because it involves no significant changes to the chip’s circuitry or design, according to the researchers.
A logic gate is sort of an electronic on-off switch consisting of transistors and wires that controls the operations of a chip. Modern microprocessors can have hundreds of millions of logic gates. The attack demonstrated by the researchers involves the use of a single such gate, with a capacitor inside capable of storing a minute electrical charge. The gate is designed in such a manner that its function can be flip-flopped—or switched from off to on—when the accumulated charge in the capacitor reaches a certain pre-defined threshold.
The Michigan University research paper describes a method where the rogue gate can be placed in such a way on the chip that it can siphon charges from a nearby wire when certain commands are issued. “If the wire toggles infrequently, the capacitor voltage stays near zero volts due to natural charge leakage,” the researchers said in their report. However, when the wire is toggled frequently the capacitor in the rogue gate begins to charge and eventually reaches a voltage threshold that causes the gate to flip to a malicious state.
Attackers can craft attack triggers that ensure the modification to a malicious state happens only when a sequence of specific and unlikely events happens. As a result even the most diligent post-fabrication tests are unlikely to catch it, the researchers said.
The researchers demonstrated the feasibility of building such backdoors in microprocessors by fabricating a chip based on open-RISC 1200 technology. “Experimental results show that our attacks work, show that our attacks elude activation by a diverse set of benchmarks, and suggest that our attacks evade known defenses,” they noted in the paper.
Modern chip design companies can open themselves up to such issues when they use third parties to fabricate their designs, the researchers said. Attackers can make minute changes to the chip and set it up so the modifications become malicious only when a specific and rare sequence of events happen, thereby evading detection during post-fabrication tests.
Yonatan Zunger, head of infrastructure for Google Assistant described the proof-of-concept as one of the “most demonically clever” computer security attacks in many years. “It's an attack which can be performed by someone who has access to the microchip fabrication facility, and it lets them insert a nearly undetectable backdoor into the chips themselves,” Zunger said in a Google+ post.
Among those who might want to attempt such an attack would be state-level actors, he said. “I don't know if I want to guess how many three-letter agencies have already had the same idea, or what fraction of chips in the wild already have such a backdoor in them.”