Osaka researchers accelerate protons to record 132 MeV using graphene and AI detection

At the University of Osaka, physicists have found a way to push protons to nearly half the speed of light using a technique that sounds almost poetic: they let the particles ride a wave of electric force generated by a laser pulse moving through ultrathin graphene. The achievement—accelerating protons to 132 megaelectronvolts, a new record for this method—suggests a path forward for building particle accelerators that are simpler and more resilient than the massive conventional machines that have dominated the field for decades.

The challenge that laser-driven acceleration has always faced is fragility. When researchers use ultrashort, high-intensity laser pulses to accelerate ions, the delicate targets holding the material can be damaged by the weak prepulse that arrives just before the main pulse hits. This vulnerability has limited how much energy researchers can reliably extract. The Osaka team, led by Takumi Minami, found a workaround by switching to longer laser pulses and graphene targets—sheets of carbon just nanometers thick. Graphene's unusual combination of extreme thinness and surprising durability meant the targets could survive the prepulse intact, remaining ready when the main laser pulse arrived.

What happened next, according to simulations, was that the laser created a propagating electrostatic wave moving through the plasma—essentially an electric field that traveled forward like a surfer's wave. Protons caught in this moving field experienced acceleration over an extended period, several picoseconds long, which allowed them to gain far more energy than shorter-pulse methods typically deliver. The extended acceleration window proved to be the key difference. Where conventional short-pulse techniques max out at lower energies, the long-pulse approach with graphene targets pushed protons to 132 MeV, beyond what the field had previously achieved this way.

But finding those protons afterward turned out to be nearly as difficult as creating them. High-energy protons are rare events, leaving faint signals scattered across millions of detector images. Manually searching through that volume of data would be impractical. The research team, working with senior author Yasuhiro Kuramitsu, deployed a convolutional neural network—a form of artificial intelligence trained to recognize the subtle patterns that individual proton signals leave behind. The neural network achieved 99.2 percent precision in identifying high-energy proton signals, allowing the researchers to confirm that protons had indeed reached the 132 MeV threshold.

The convergence of these two advances—long-pulse laser acceleration and AI-powered detection—points toward something larger. If experiments can automatically analyze their own results and use that feedback to adjust their parameters in real time, laser systems could eventually optimize themselves without human intervention between shots. That vision of autonomous, self-improving experiments remains on the horizon, but the Osaka results demonstrate that the technical pieces are falling into place. The work was published in Progress of Theoretical and Experimental Physics and represents a step toward next-generation accelerators that might be smaller, more efficient, and capable of exploring higher-energy physics with fewer of the constraints that have bound the field to massive, centralized facilities.