Researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) have developed Fractal, an operating system kernel designed to turn processors into subjects of rigorous microarchitectural study. By booting directly on bare metal, Fractal allows scientists to isolate hardware behavior from software “noise,” leading to the discovery of previously undetected speculative execution vulnerabilities in Apple’s M1 chip, including “Phantom” speculation. The tool supports x86_64, ARM64, and RISC-V architectures, providing a standardized environment for future security research.
How Fractal eliminates experimental “noise”
Traditional security research often relies on modifying existing operating systems like macOS or Linux, a process that researchers describe as unstable and prone to interference. According to MIT PhD student Joseph Ravichandran, general-purpose operating systems inject scheduler activity and address-space management into measurements, which obscures the CPU’s internal actions. Fractal solves this by implementing “multi-privilege concurrency,” a technique that allows a single experiment to switch privilege levels at runtime while maintaining a consistent address space. This approach produces clean signals and flat baselines, enabling researchers to observe chip behavior without the background interference of a standard kernel.
Fractal includes ports for standard tools like GCC, vim, and the dash shell, allowing researchers to migrate existing experiments with minimal friction.
What the M1 findings reveal about chip security
Using Fractal, the MIT team confirmed that Apple’s M1 processor implements CSV2 protections correctly for the execute stage of indirect branch prediction. However, the team identified that the CPU still fetches target instructions into the cache before these protections activate. This creates a side channel that allows user code to influence kernel cache activity. Furthermore, the team provided the first evidence that Apple Silicon is susceptible to “Phantom” speculation—a class of misprediction previously only documented in Intel and AMD processors. Mengyuan Li, an assistant professor at the University of Southern California, notes that Fractal effectively transforms ad hoc reverse-engineering into reusable research infrastructure.
Why standardized infrastructure matters for hardware research
The transition from ad hoc experiments to shared infrastructure could fundamentally change how the industry addresses hardware vulnerabilities. While previous studies on the M1’s conditional branch predictor suggested privilege isolation existed on performance cores but not efficiency cores, Fractal’s data indicated that neither core type maintains true isolation. Ravichandran attributed the earlier, conflicting results to macOS migrating threads between cores during system calls. By providing a “true independent variable” where only the privilege level changes, Fractal aims to make microarchitecture research as reliable as software development tools like QEMU or FFmpeg.
Pro tips for security researchers
- Isolate variables: Always verify if thread migration between cores is influencing your performance data before concluding that hardware isolation is failing.
- Utilize bare metal: When testing speculative execution, remove the OS kernel from the equation to ensure your measurements are not artifacts of scheduler activity.
- Prioritize reproducibility: Move toward standardized kernels like Fractal to ensure that your findings can be validated by other research groups across different architectures.
Frequently Asked Questions
What is “Phantom” speculation?
Phantom speculation is a class of misprediction where a CPU incorrectly identifies ordinary instructions as branches, triggering speculative execution that the software never intended to run.
Is Fractal limited to Apple processors?
No. According to the research team, Fractal supports x86_64, ARM64, and RISC-V architectures, making it a versatile tool for broader microarchitecture investigation.
How does Fractal differ from a virtual machine?
Unlike virtual machines or standard operating systems that manage resources for the user, Fractal boots directly on bare metal and provides primitives that allow researchers to control privilege levels and memory access with high precision.
Have you encountered unexpected side-channel behavior in modern CPU architectures? Share your experiences in the comments below or subscribe to our newsletter for the latest in hardware security research.
