A team of Chinese scientists has unveiled the world’s smallest and most energy-efficient transistor in a breakthrough poised to anchor the next generation of high-performance AI hardware.
The researchers achieved the feat in ferroelectric transistors (FeFETs), which function similarly to neurons in the human brain as they integrate memory and processing in a single unit, thereby reducing the time lost in data transfer.
In conventional semiconductor chips, data storage and computation occur in separate areas, creating efficiency bottlenecks for communication “across the wall”.
According to Qiu Chenguang of Peking University, the “in-memory computing capability of FeFETs aligns closely with the future evolution of AI chips”.
“The industry views them as one of the most promising devices for enabling brain-inspired neuromorphic computing,” Qiu told Science and Technology Daily, the official ministry newspaper.
Qiu, with Peng Lianmao, a member of the Chinese Academy of Sciences, led the team that developed the world’s smallest and most energy-efficient ferroelectric transistor. Their work was published this month in the peer-reviewed journal Science Advances.
The design “overcomes long-standing limitations of traditional FeFETs, including excessive energy consumption and mismatched logic voltage, paving the way for large-scale application”, Science and Technology Daily said.
The rapid advancement of artificial intelligence (AI) has intensified global demand for computing power and accelerated research into semiconductor materials.
To enhance computational capabilities, one approach is to build larger and faster chips. Another is changing how chips “think”.
The Chinese scientists focused on the latter, as FeFETs possess both storage and computing capabilities, unlike traditional semiconductor transistors.
A major challenge facing current FeFETs is that data writing and erasing require a relatively high operating voltage of about 1.5 volts.
In contrast, modern logic circuits typically operate at voltages below 0.7 volts. In other words, while FeFETs offer superior performance, they are like a heavy door that is difficult to push open.
Using advanced processing techniques, the team refined the transistor’s structure.
“We scaled the gate electrode down to just 1 nanometre, achieving atomic-level precision,” Qu explained. “This enabled the formation of a strong electric field within the ferroelectric layer, allowing FeFETs to work with minimal external energy at a voltage as low as 0.6 volts.”
“This nanogate FeFET operates at ultra-low voltage and consumes significantly less energy, about one-tenth of the lowest reported energy consumption values internationally,” he added. “It also supports high-speed memory operations with a response time as fast as 1.6 nanoseconds.”
According to the Peking University website, the processing and design techniques developed for ferroelectric memory chips based on this new mechanism have already been patented, representing fully independent intellectual property.
Qiu said the study marked “the first international discovery that FeFETs exhibit unexpected advantages when scaled down in size, demonstrating their strong potential for building future sub-1-nanometre node chips”.
“Such a transistor can be used both to build highly energy-efficient data centres and to assemble next-generation high-performance AI chips.” – South China Morning Post
