Fixing Security Holes

Why chipmakers need to pay attention to side-channel attacks.

popularity

Connected devices can do everything from save lives to improve the quality of life. They also destroy that quality or cause harm if these things or systems of things are not secure.

Security is a complex multi-level problem. It spans the entire seven-layer OSI communication stack, as well as the software that is used to run, manage and operate hardware. And it needs to be dealt with from multiple angles, from the smallest IP block and memory to the bus that connects them together.

The goal in most cases isn’t to make devices impenetrable. Given enough time and resources, and enough incentive, even the most sophisticated security systems can be hacked. As hardware engineers, though, what we should be worrying about is slowing down the hackers, minimizing the damage, and making it unprofitable for them to hack into devices.

The best known of these approaches uses power analysis—whether simple or differential—to create what is known as a side-channel attack. The simple power analysis looks at current consumption over time to show what function is being performed at any particular moment. That can be mapped using a standard oscilloscope. Differential power analysis, in contrast, uses multiple traces of a pair of data sets to compute the difference.

Both of these approaches are well understood, and so are the solutions. Injecting noise and avoiding conditional branches, for example, can stop simple power analysis attacks. And in differential power analysis, temporal noise can be injected through such techniques as varying clocks or creating dummy operations.

In both cases, the solution involves a combination of obfuscation and adding protocol limits to the number of transactions that can be performed with a given key. If a key can only be used a few hundred times before it is destroyed or replaced with a new key, the damage from a side-channel attack can be minimized.

Breaking into an insulin pump or a connected washing machine isn’t nearly as attractive if you only can get into one small piece of a device and then get stopped at the next door and a whole series of other doors that are difficult to find, break through, and which may lead nowhere except down blind alleys. And it’s even less attractive if that effort doesn’t lead an attacker any closer to breaking into another device sold by the same company because all of those keys are different.

But all of this requires chipmakers and systems companies to deal with security as a system-wide problem. Increasingly, implementing good security will a prerequisite for even competing in a market. For hardware vendors, a good starting point is to close up some of the best-known attack vectors, and then to start building a system that can evaluate and objectively quantify security measures that are built into devices, just as IP is characterized for power, performance and noise.

The time has come to close the loopholes, and the best way to do that is by providing incentives for companies to invest in this technology, and disincentives for those that do not.