ANTMAN: An Efficient and Interpretable RTL-Level Run-Time Detection Framework for Stealthy Branch Predictor Attacks on BOOM

Imagine your computer processor as a super-fast, hyper-organized chef in a busy kitchen. To save time, this chef doesn't wait for every order to be fully confirmed before starting to cook; instead, they guess what the customer might want next and start chopping vegetables or heating pans based on that guess. This is called "speculative execution," and it makes computers incredibly fast. However, just like a chef who accidentally leaves a trail of chopped onions on the counter, these guesses leave tiny, invisible footprints in the machine's memory.

Hackers have learned to read these footprints. By watching how long it takes the chef to realize a guess was wrong and start over, they can figure out secret recipes (like passwords or encryption keys) without ever actually seeing the ingredients. This is known as a "side-channel attack." For a long time, security experts tried to catch these hackers by counting how many times the chef picked up a knife or opened a drawer (using hardware counters). But the hackers got too sneaky; they learned to hide their footprints so well that the counters couldn't see them, especially in the newer, open-source RISC-V processors that are becoming popular in everything from phones to satellites.

Enter ANTMAN, a new security framework designed by researchers to catch these invisible intruders. Think of ANTMAN not as a counter, but as a super-observant kitchen inspector who doesn't just count actions but understands the story of what's happening. Instead of looking at isolated events, ANTMAN watches the entire kitchen for a specific, chaotic dance that only happens when a hacker is trying to steal secrets.

The researchers built this system on the BOOM processor, a famous open-source design that acts like a test kitchen for RISC-V chips. They first let the computer run thousands of normal cooking tasks (benign workloads) and then let a hacker try to steal secrets using two different tricks: one simple and one complex. By recording every single signal inside the processor's brain, they used a data-mining technique called "association rule mining." Imagine this as teaching a computer to recognize that whenever the chef drops a spoon, the oven beeps, and the lights flicker all at once, it's not an accident—it's a signal that a thief is in the kitchen.

The paper finds that these "secret dances" are incredibly consistent. When a hacker tries to exploit the processor's guessing mechanism, it triggers a specific chain reaction: the processor realizes it guessed wrong, tries to correct its path, and then has to wipe its entire mental slate clean to start over. This happens across many different parts of the processor simultaneously. The researchers discovered 16 specific rules that describe this chaotic pattern. For example, if the "branch misprediction" signal in the main unit matches the "flush" signal in the memory unit and the "redirect" signal in the fetch unit all at the same time, it's almost certainly an attack.

When they built a tiny, non-intrusive monitor into the processor's design to watch for these 16 rules, the results were striking. In their simulations, ANTMAN caught the hackers 100% of the time. It didn't matter if the hacker was using a simple trick or a complex one; the system spotted the attack after just 25% of the attack was complete for the simple version, and even caught the complex, unseen variations before the secret was fully stolen. Crucially, it never cried "wolf" during normal cooking tasks, achieving zero false alarms.

The researchers are careful to note that this was tested in a simulated environment (a perfect digital model of the processor), not on a physical chip in a real-world device yet. However, the simulation shows that by designing the security directly into the processor's blueprint (RTL level) rather than trying to patch it later, they can stop these stealthy attacks much faster than previous methods. They argue that while older methods relying on hardware counters are like trying to catch a ghost by counting how many times a door creaks, ANTMAN is like recognizing the ghost's specific shadow pattern. This approach offers a promising, "secure-by-design" path forward for protecting the next generation of open-source computers against invisible threats.