Scientists in Japan have developed a new, highly efficient method for designing wireless power transfer (WPT) systems. Based on machine learning, the method enables a system to maintain stable voltage even as loads change — a key step towards broader adoption of wireless power.
WPT systems are already a key part of numerous devices, from smartphones and biomedical sensors to induction stovetops, which use the mechanics of WPT to heat cookware. But one of the key problems with current technology is that it struggles with power fluctuations. This is because they’re load-dependent, meaning a system’s performance is significantly affected by what device is being powered (the load).
Devices like smartphones rely on constant, regulated voltage to charge their batteries safely. A battery’s resistance to electricity changes as it fills up, which in a load-dependent WPT system can cause the voltage to fluctuate. This can damage the device or reduce charging speed.
In comparison, the new machine-learning approach is load-independent (LI), meaning they can deliver consistent power and maintain high efficiency no matter what device is being charged. In the example of the smartphone battery, this means that power will continue to be efficiently transferred at a steady voltage regardless of fluctuations in resistance that may occur as the battery fills up.
This is especially important for larger batteries in more complex applications, such as electric vehicles, where load can shift dramatically during charging.
The researchers revealed their findings in a new study published June 2025 in the journal IEEE Transactions on Circuits and Systems.
Related: DARPA smashes wireless power record, beaming energy more than 5 miles away — and uses it to make popcorn
WPT systems function through a process called resonance, the same way as a radio or television receives an over-the-air signal. A transmitter broadcasts a specific frequency of electromagnetic wave by adjusting the way power oscillates between a capacitor and inductor. When the wave reaches a receiver circuit set to the same frequency, the two resonate, greatly amplifying the signal.
In a radio, the signal is then sent to other components to be further amplified and demodulated to create sound, while in a WPT system this resonance enables the receiver to capture and store the energy being transmitted wirelessly.
Machine learning to boost wireless power
The new technique uses machine learning to model and optimize less load-dependent power transfer systems. The process involves building a virtual model of the system, then running simulations of the model in action while an artificial intelligence observes it.
The AI judges how well the system is operating, based on criteria like how much power is lost as heat and how clean the electrical signal remains. It then uses a trial-and-error method to optimize the system so that it operates at peak efficiency, transferring power with minimal fluctuations and energy dissipation.
Using their new method, the researchers reduced fluctuations down to 5%, compared to 18% using a load-dependent system, according to the study. They also increased power transfer efficiency up to 86.7%, while load-dependent systems can operate as low as 65% efficiency.
Load-independent WPT systems have broad implications far beyond wirelessly charging devices, said study lead author Hiroo Sekiya, a professor at Chiba University’s Graduate School of Advanced Integration Science.
“We are confident that the results of this research are a significant step toward a fully wireless society,” he said in a statement. “Moreover, due to LI operation, the WPT system can be constructed in a simple manner, thereby reducing the cost and size. Our goal is to make WPT commonplace within the next 5 to 10 years.”
This study also illustrates the ways AI can be used to improve electrical circuit design, leading to a transformation in “how power electronics are designed, moving toward a future of automated circuit design,” the researchers said in a statement.
