Searching for esoteric hardware vulnerabilities may be missing the point.
Speculative execution seemed like a good idea at the time. As the power/performance benefits of each node shrink began to dwindle, companies like Intel figured out ways to maintain processor speeds at the same or lower power.
There were other approaches, as well. Speculative execution and branch prediction are roughly equivalent to pre-fetch in search, which has gotten so good that often the user doesn’t even have to finish typing a word or phrase before the search has begun. The idea was ingenious for its day. It saved money, which was passed along to consumers, and at the time the added complexity and risk of side-channel attacks seemed minimal.
In a paper published this week, researchers from Worcester Polytechnic Institute and the University of Lübeck found a way to dramatically accelerate reverse engineering of physical page mapping. While the researchers found the problem is limited to Intel chips, the reality is that all processors need to be revisited and rethought from the perspective of how data is stored and accessed.
The paper is worth a read for anyone who wants to dig into how what appeared to be a good idea at the time can go terribly wrong years later. In effect, this is delayed post-silicon debug of the worst kind. The Jeep hack of the CAN bus in a car is roughly the equivalent. As one top automotive executive remarked years later, “We never thought anyone would hack a car.”
There are several takeaways that everyone needs to consider. First, what is considered secure today may not be secure in 5 or 10 years. This sounds obvious, but it requires thinking about security as part of the initial architecture. Where will keys be stored? Can they be modified, and if so, by whom? What is the authentication strategy for users, and can it be updated? And can the overall chip architecture be modified in the field to resist attacks, possibly with the help of AI?
Unless multi-layered security is an inherent part of the architecture, nothing will work. So far, it appears the best approach to security is resilience, because not every possible attack surface or approach can be identified when a chip is designed. Nothing can withstand attacks forever, particularly as hackers gain access to more and more powerful computers and now-affordable tools such as scanning electron microscopes.
Second, designing in sufficient security will become even more difficult as architectures become more heterogeneous and systems become more distributed. While digital logic will continue to shrink, increasingly it will be mixed with other processing elements developed at a variety of process nodes. Some of that will be on-chip, some of it will be off-chip in a package, and some of it may involve multiple devices networked together. In effect, security will require a system of systems, and it will have to determine whether anything has been compromised based upon data screening and movement.
Third, making this all work will require much more cooperation across the supply chain. As the number of sensors continues to grow, so does the amount of data being generated by those sensors. That data needs to be processed, moved and stored, and each one of those steps creates new vulnerabilities that never existed before. For hackers, the real objective isn’t hacking into a processor. It’s the growing value of that data.
So while speculative execution is an interesting door into this world, it’s one of the harder ones to crack. And that fact that it has been discovered by a team of researchers spanning two continents should be an indication of just how difficult it was to find. Put in perspective, this is like searching for water vapor in lab while the water level outside the building is approaching flood stage.
Related Stories/Blogs
Meltdown And Spectre, One Year Later
Paul Kocher: New vulnerabilities have emerged around software techniques that damage hardware.
Finding Security Holes In Hardware
Emphasis on performance, backward compatibility and system complexity are creating vulnerabilities that are difficult to fix.
Next Wave Of Security For IIoT
New technology, approaches will provide some protection, but gaps still remain.
Using AI Data For Security
Pushing data processing to the edge has opened up new security risks, and lots of new opportunities.
Open-Source RISC-V Hardware And Security
Experts at the Table, Part 1: The advantages and limitations of a new instruction set architecture.
Leave a Reply