Researchers at MIT have taken the first small—but real—steps toward the next generation of processors, building the first RISC-V compliant processor powered by carbon nanotubes. The processor can handle 32-bit instructions and 16-bit memory addressing, and the team got it to execute a Hello World! demo. That’s a far cry from what silicon chips can do, to be sure, but it’s an important step.
While chipmakers aren’t done squeezing every last bit of power out of silicon, the life of silicon processors is coming to an end. Despite what the logic of Ant-Man might suggest, we can only miniaturize things so much. That means we need to find something else to take the place of silicon. Researchers have long looked at carbon nanotubes as a potential new material for processors, but it’s a possibility as rife with pitfalls as it is full of potential.
Carbon nanotubes are natural semi-conductors, and they’re extremely small, making them an ideal option for further miniaturizing processors. However, they have to be grown, and it’s proven difficult to grown them precisely. Getting this to work requires 100% purity, but researchers have topped out at 99.9%. That sounds closer than it is; applications like this require absolute precision.
Further, silicon semiconductors can be modified to have positive and negative biases through a process called “doping,” but the microscopic size of carbon nanotubes makes that incredibly difficult to do.
Working around the pitfalls of carbon nanotubes
MIT researchers, working with scientists at Analog Devices, Inc., have found some workarounds for these hurdles.
First, the researchers let the nanotubes grow as they will. This is a somewhat chaotic process. Along with the useful, single carbon nanotubes, metallic nanotubes also grow, along with aggregated clusters of nanotubes. After letting the forest of nanotubes grow, the team applied a layer atop the surface that was then removed by sonication. The layer took the aggregated clusters with it, leaving the other ones in place.
Next, the team etched off most of the layer of nanotubes so that only the ones they wanted were left in place, and then applied a layer of oxide that acted as a doping agent. It’s not as precise as the process used for silicon processors, but it worked for this application.
Finally, the team worked around the metallic nanotubes that aren’t useful for a processor by modifying the RISC design tool. The team found that some functions are less sensitive to the presence of metallic nanotubes, and modified the open-source design tool to account for that, avoiding the functions that are sensitive to those metallic nanotubes.
The end result is the RV16X-NANO chip. This chip features 14,000 transistors—a far cry from the billions on current generation CPUs—but with that all-important 100% yield. The team then was able to execute that Hello World! demo.
Not quite the answer
There are a ton of ways to improve on the existing design, researchers say. The design has to be tolerant of imperfection, though, and are the exact opposite of that. The chip has to allow for incorrectly-oriented carbon nanotubes and metallic nanotubes.
Ultimately, researchers want to get to a place where they can control the growth of carbon nanotubes enough to make single-nanotube transistors, and this experiment is about working around that. It’s not a solution. But what it does do is prove that it is possible to make carbon nanotube chips, and it improves the overall understanding of how carbon nanotubes could work in a processing environment. Both of these are steps in the right direction, even if they’re not steps straight forward.