At the moment, our processors are built on silicon. But fundamental limits on what can be done with that material has researchers eyeing ways to use materials that have inherently small features, like nanotubes or atomically thin materials. At least in theory, these will let us do what we’re now doing, just more efficiently and/or with physically smaller features.
But can these materials allow us to do things that silicon can’t? The answer appears to be yes, based on research published earlier this week. In it, the researchers describe transistors that can be reconfigured on the fly so that they perform completely different operations. They suggest this can be useful for security, as it would keep bad actors from figuring out how security features are implemented.
Doping vs. security
The researchers, based at Purdue and Notre Dame, lay out an argument for why this sort of reconfigurable circuitry could have security implications. It comes down to the materials science of silicon transistors. They require areas of silicon that either hold negative or positive charge (creatively named p- or n-type semiconductors). These are created by doping, or adding small amounts of certain elements to the silicon. This is done during the manufacturing, and the doping is locked into place at that point. This means that the operation of individual transistors is locked into place when the chip is made.
That becomes an issue for security-focused hardware. If any of the features are implemented in a silicon chip (as opposed to being purely software-based), then they’d have to be physically committed to the chip hardware itself. And, since that hardware is static, knowing the chip layout would mean understanding something about how the security hardware operated, potentially exposing its vulnerabilities. That’s not an abstract fear; we have advanced microscopy techniques that can examine hardware at the level required, and there’s some indication that they’ve already been used to do so.
The solution to this, the authors argue, is to create transistors that aren’t committed to a particular function. And it’s not possible to do that with silicon. But it turns out that atomically thin materials, which have been studied for other reasons, aren’t inherently p- or n-type semiconductors. Their behavior is set by their environment, as they’ll carry a positive or negative charge depending on what’s injected into the material from the metal conductors that wire up the transistor. So, the researchers decided to test whether they could actually build a reconfigurable transistor.
While there are a variety of atomically thin materials—graphene, MoS2, and more—the researchers decided to work with something called black phosphorus. The material is formed of multiple layered sheets, with each sheet composed entirely of phosphorus atoms chemically linked to each other. Unlike graphene, which is planar, the chemical bonds of phosphorus cause these sheets to have regular ridges and troughs, like corrugated metal. (When we last visited this material, it was being used to make fast-charging batteries.)
Black phosphorus was chosen because it has a small bandgap, which means that it doesn’t require a large voltage difference to operate. Unfortunately, that also meant that the difference between its on and off states was small. This problem was exacerbated by the fact that the hardware was designed so that current could flow in both directions. When in the off state, it became possible for current to flow forward or backward at a low level, making it harder to register “off” as a lack of current.
To deal with this problem, the researchers substantially redesigned the transistor. In silicon, a transistor has source and drain electrodes to allow current to flow across the transistor and a gate electrode that switches this current on or off. For the reversible version with black phosphorus, the researchers used two gates, which reinforced the on/off signal. They also added what they called a “polarity gate,” which blocked the flow of current when the gate was supposed to be off.
With these in place, the researchers saw excellent performance: operation at small voltages and a clear difference between the on and off states, with the difference growing as the voltage was ramped up.
NAND, NOR, and XOR
With that in place, the researchers built an actual bit of logic. This had a single-bit key that set the state of the gate. With the bit in one state, the hardware would perform a NAND (not-and) function. Flip the bit, and instead it would perform a not-or operation (NOR). And, based on the graphs in the paper, it worked exactly as it should. The researchers also showed that it was possible to create a similar device that could switch between exclusive-or (XOR) and NOR just by tweaking some details of the configuration.
The key thing is that the state of the bit can be set dynamically at runtime. Without knowing the state of the bit, there’s no way to know what operation these gates are performing by looking at the hardware. Even if you have the complete hardware layout, there’s no way to tell what these gates might be doing.
Is this important? Maybe not—we’re a long way from implementing any of these new materials in anything close to production hardware. But it’s exciting to see it being considered, because we’ve not seen a lot of reports like this. As the authors argue, “research in the field typically focuses on demonstrating operations that are also achievable with traditional transistors, and efforts that to try to leverage the unique properties of 2D materials, such as ambipolarity, to deliver new functionalities are rare.”