World’s first AI Chip Made of Carbon

 

Chinese scientists build world’s first AI chip made of carbon and it’s super fast



Chinese researchers have developed a chip that it could be a game-changer in modern computing

In a breakthrough that could redefine the future of computing, Chinese scientists have developed the world’s first carbon-based microchip capable of running artificial intelligence tasks using a novel ternary logic system, leapfrogging today’s binary-dependent technology.
The achievement, led by researchers from Peking University and Beijing University of Posts and Telecommunications, marks a pivotal step towards next-generation semiconductors that could outperform silicon in speed, efficiency and scalability.
The team unveiled a carbon nanotube (CNT) transistor chip that processes data not just in ones and zeros, but also a third state, enabling computations to occur faster and with less energy. In experiments, the chip demonstrated flawless accuracy in image recognition tasks, a milestone detailed in the January issue of the journal Science Advances.
CNTs are tiny tubes made from rolled-up sheets of graphene. They offer exceptional mechanical and electrical properties as well as excellent stability, and are mainly used as conductive additives in lithium-ion batteries.

However, due to their superior electrical characteristics and ultra-thin structure, CNTs are also a promising material for next-generation semiconductor technology. Compared to traditional silicon-based chips, CNTs offer higher carrier mobility, lower power consumption and the potential for even smaller transistor sizes – key advantages that could help overcome the limits of Moore’s Law, which predicts a doubling of semiconductor capacity every two years.

As early as 2009, CNTs were listed as a future technology option in the International Technology Roadmap for Semiconductors.

Most modern electronic devices rely on binary logic systems, where all data is represented using zeros and ones. Therefore, introducing a third state could allow for more efficient data processing, enabling faster computations without increasing circuit complexity. This approach, known as ternary logic, improves information transmission efficiency within the same physical space.

The research team designed a novel CNT transistor, introducing a concept known as source-gated transistors (SGTs). By adjusting the gate voltage, the CNT transistor can move between three distinct current states, forming the foundation for ternary logic circuits.

Building on this, the researchers constructed a neural network capable of learning and reasoning by mimicking the connections between neurons in the human brain. To validate their ternary logic circuit, they conducted extensive experiments, showing that this CNT-based neural network achieved 100 per cent accuracy in classifying handwritten digits – highlighting its immense potential in artificial intelligence applications.
Beyond its computational efficiency, the CNT transistor boasts high stability and strong resistance to interference. It also offers advantages in power consumption and processing speed. According to the researchers, their design is highly scalable and could be integrated into large computing systems.

This technology has broad applications in high-performance computing, artificial intelligence, machine learning and low-power storage devices. It could also be used in IoT devices, where energy efficiency is a key concern.

One of the lead researchers, Peng Lianmao, a member of the Chinese Academy of Sciences, has been exploring carbon-based chip technology for more than two decades. His team has made significant progress in developing high-performance CNTs and realising the precise arrangement of nanotube arrays.

In 2020, they successfully fabricated an eight-inch CNT wafer, with its integrated circuits outperforming silicon-based CMOS devices of a similar size – a milestone that placed China at the forefront of international research in this field.

By 2024, Peng, together with Professor Zhang Zhiyong from Peking University, developed the world’s first CNT-based tensor processor chip, integrating 3,000 CNT transistors capable of performing image contour recognition and extraction tasks.

CNT chips still lag behind silicon chips in terms of integration density. For instance, Nvidia’s RTX 5090 GPU, announced in January 2025, contains a staggering 92 billion transistors – far beyond the existing capabilities of CNT technology.

However, the race for carbon-based chips is a new frontier for the global semiconductor industry, and China currently holds a leading position in this field.

“Our ultimate goal is to make carbon-based chips mainstream within the next 10 to 15 years, enabling their widespread use in supercomputers, data centres, smartphones and other mainstream electronic devices,” Peng said in an interview published on Peking University’s website.

If successful, this transition could redefine the future of computing, pushing beyond the limits of traditional silicon-based technology and ushering in a new era of high-performance, low-power semiconductor solutions.

We are thrilled to extend a warm welcome to the 

China Scientist Awards!

Join us for the China Scientist Awards, a premier event in the realm of research. Whether you're joining virtually from anywhere in the world, this is your invitation to explore and innovate in the field of research. Become part of a global community of researchers, scientists, and professionals passionate about advancing research.

Comments

Post a Comment

Popular posts from this blog

New Frost-Resistant Sand-Control Agent

Image
  Across China: Chinese researchers develop new frost-resistant sand-control agent LANZHOU -- A Chinese research team has developed a chemical sand-fixation material suitable for use in cold desert regions, which is expected to serve as a new tool for sand control and desertification prevention in such areas. The application of chemical materials to stabilize shifting sands is one of the primary methods of desertification control. This approach involves the use of adhesive chemical substances to bind loose sand particles together, thereby mitigating encroachment by wind-blown sand. However, conventional chemical sand-fixation materials have been mainly designed for hot and arid regions. In colder, high-altitude or high-latitude desert regions, such as the Qinghai-Xizang Plateau and the Mongolian Plateau, which are located in Northwest, Southwest and North China, cooler temperatures often render traditional methods of sand control ineffective. Researchers from the Northwest Institut...
Read more

Bamboo In The World’s Longest Sea Bridge

Image
  How Chinese engineers used bamboo in the world’s longest sea bridge, and more Chinese scientists have been working on new technologies to make stronger and more durable bamboo-based materials for large-scale infrastructure, reinforcing the country’s position as a leader in the emerging eco-friendly industry. Bamboo, which is known for its rapid growth and high strength, was engineered into composite panels used on the scenic platforms that stretch for kilometres along the  Hong Kong-Zhuhai-Macau Bridge , the official Science and Technology Daily reported on Wednesday. Six years after they were installed on the bridge – the world’s largest sea crossing – the bamboo-based panels have withstood sunlight, typhoons and seawater corrosion, remaining “as solid as ever”, according to the report. Lou Zhichao, from Nanjing Forestry University’s Bamboo Research Institute, said most of the world’s bamboo was grown in developing countries. “China is not only the world’s largest bamboo pr...
Read more

Blueprint Of Human Brain Cortex

Image
  Chinese scientists uncover complex blueprint of human brain cortex BEIJING -- Researchers from the Institute of Automation at the Chinese Academy of Sciences have revealed the intrinsic relationship between the topological structure of human brain cortex connections and genetic characteristics, according to a study published in the Journal of Neuroscience. Fan Lingzhong, a researcher at the institute, explained that neural networks operate as human think, learn, or perceive the world, with trillions of connections enabling rapid information transfer. The study addresses a fundamental question: How do these intricate connections form, and why do distinct brain regions exhibit such orderly distribution across the cortex? "The brain begins following a genetic 'blueprint' during embryonic development," Fan noted. Researchers proposed a hypothesis: genetic encoding and cortical connectivity are not directly correlated due to the vast numerical disparity between genes and...
Read more