Chinese scientists have developed a new type of TPU — or tenor processing unit — that utilizes carbon nanotubes in place of the traditional silicon semiconductors that are typically used.
They have said that the new chip — which is a special type of computer chip — could create artificial intelligence technology that is significantly more energy-efficient than what’s being done today.
One of the big problems with AI models is that they are very data-intensive. In other words, they require massive computational power just to run, which means they need a lot of electricity.
Because of this, there are big obstacles in scaling up and training, especially as the demand for applications that run on AI continues to increase.
It’s a major reason why there are scientists across the world who are working to try to make new components that can handle all the computations that are necessary for these AI applications while using significantly less energy than they do now.
In 2015, scientists at Google created the TPU to address just this challenge. The specialized chips are dedicated accelerators for hardware parts that can complete complex math calculations that are used to run and train various AI models.
The TPU enables these models to train much more efficiently and faster, since they allow the tasks to be offloaded from a computer’s graphics processing unit (GPU) and central processing unit (CPU).
The new chip the scientists from China have developed, though, uses carbon nanotubes, making them the first to do so. These tiny structures consist of carbon atoms, and can replace the traditional materials that are used in semiconductors such as silicon.
The structure, according to the Chinese scientists, allows electrons to then flow through without much resistance, which as a result makes them excellent at conducting electricity.
All of this information was published in late July in the Nature Electronics journal by the scientists.
The Chinese scientists reported that the TPU they designed consumes only 295 microwatts of power and has the ability to deliver up to 1 trillion different operations per watt.
The Edge TPU that Google has created, by comparison, is able to perform as many as 4 TOPS (trillion operations per second) and uses two watts of power.
Based on this data, the carbon-based TPU that the Chinese scientists have created is almost 1,700 times more energy-efficient than Google’s.
The co-author of the paper, Zhiyong Zhang, who works at the Peking University, commented:
“From ChatGPT to Sora, artificial intelligence is ushering in a new revolution, but traditional silicon-based semiconductor technology is increasingly unable to meet the processing needs of massive amounts of data. We have found a solution in the face of this global challenge.”
There are 3,000 carbon nanotube transistors included in the Chinese TPU. It’s also designed in a very unique way, with a full network of the processors being arranged in a pattern that looks like a grid.
