The rise of AI has created an almost insatiable appetite for computing power. Training and running AI systems requires vast numbers of transistors, and engineers are now racing to pack more of them onto every chip. With their existing designs, however, silicon transistors are rapidly running up against physical limits on how small they can get.
Through new research published in Nature, a team led by Ya-Ping Chiu at National Taiwan University has uncovered new details about next-generation transistors that could help push past these limits.
Downsizing to 2D
Today, the most advanced transistor designs available are built from materials just one atom thick, known as 2D materials. Inside any transistor, charge carriers (either electrons or the positively charged “holes” left behind when electrons move) flow through a channel to create an electric current. Because 2D materials are so thin, engineers can exert unusually precise control over this flow, making them promising candidates for building smaller, more efficient transistors.
But as well as shrinking the channel, researchers also need to consider how current enters the material through metal contacts. So far, it has remained unclear exactly how small these contacts can become before they stop working properly. Key to answering this question is a quantity called the transfer length: the distance over which most of the current actually enters the material. If contacts become smaller than this transfer length, a bottleneck is created—damaging the transistor’s overall performance.
Probing metal contacts
To find out just how small these metal contacts can get, Chiu’s team turned to a specialized imaging technique called scanning tunneling microscopy. By slicing open working transistors and scanning a needle-like probe across the exposed surface, they could directly track how current spreads out from a metal contact into a single atomic layer of the transistor’s semiconductor component.
Strikingly, the team measured a transfer length of just 2 to 3 nanometers—far smaller than previous theoretical estimates, which ranged from tens of nanometers to nearly 170 nanometers. Crucially, these tiny transfer lengths already meet the contact-size requirements projected for the next generation of ultra-miniaturized transistors.
Keeping up with demand
The team’s findings suggest that 2D transistor contacts can be scaled down just as aggressively as the transistors themselves, clearing a major hurdle on the road to smaller, denser, more efficient chips. If contacts really can shrink this far without sacrificing performance, it would strengthen the case for 2D materials as a genuine successor to silicon—helping to keep pace with AI’s relentless demand for more powerful, energy-efficient hardware.
© 2026 Science X Network
Source: Phys.org