Japan's AI Designed Wireless Power System Cuts Voltage Swings to Under 5%

Chiba University targets a weak point in wireless charging

Scientists at Chiba University in Japan have developed a method using machine learning that reduces voltage fluctuations in a wireless power transfer system to less than 5%, compared with swings of up to 18% in conventional designs. Their experimental system achieved 86.7% efficiency at a frequency of 6.78 MHz while delivering more than 23 watts of output power.

The research, published in IEEE Transactions on Circuits and Systems in June 2025, addresses a practical problem: keeping electricity flowing steadily when the device receiving it changes its power needs. Conventional systems can become less efficient under changing loads, with efficiency falling as low as 65%.

The team, led by Professor Hiroo Sekiya, combined detailed numerical models with a genetic algorithm to choose circuit components. Instead of relying entirely on equations that simplify how components behave, the method simulates a circuit with the imperfections that affect real hardware.

Wireless power transfer, or WPT, already appears in smartphones, electric toothbrushes and induction stovetops. The Chiba University work could help extend its use to electric vehicles, industrial equipment and other applications that need dependable power as operating conditions change.

As of August 22, 2025, the results are based on laboratory experiments. Turning the approach into widely available products will require further engineering, safety testing and standards that allow equipment from different manufacturers to work together.

Why a changing load makes charging difficult

A wireless power system sends energy through electromagnetic fields rather than a physical cable. A transmitter coil creates a field at a particular frequency, and a receiver coil tuned to that frequency captures the energy and converts it back into electricity. This matching of frequencies is known as resonance.

The receiving device is called the load. Its electrical characteristics affect how the power circuit behaves. In a smartphone, for example, the battery's resistance changes as it fills up. A system that delivers steady voltage at one stage of charging may behave differently at another.

Voltage fluctuations can slow charging, reduce efficiency or damage sensitive electronics. Larger batteries, including those in electric vehicles, make the challenge more demanding because their loads can change substantially during charging.

Engineers traditionally calculate the required values of inductors and capacitors with analytical equations. Inductors store energy in a magnetic field, while capacitors store it in an electric field. Choosing their values helps determine how a wireless power circuit responds.

Those calculations often depend on idealized assumptions. Actual components vary because of manufacturing tolerances, unintended electrical effects such as parasitic capacitance, and environmental changes. When these factors are left out of a design, the finished circuit can behave differently from the prediction, especially as the load changes.

How the team built a more realistic model

Sekiya's team modeled the complete WPT circuit using differential equations. These equations describe how electrical quantities change over time, allowing the researchers to follow the circuit's behavior through successive steps rather than relying on a simplified calculation.

The model includes effects such as diode parasitic capacitance. A diode directs current through a circuit, but it also has an unintended ability to store a small electrical charge. At the frequencies used in this system, such behavior can influence performance and needs to be included in the design.

The researchers solved their equations numerically, step by step, until the circuit reached a steady operating state. That gave the optimization process a way to assess how each proposed design would perform once its initial changes had settled.

This detailed modeling is central to the approach. The team sought stable voltage and efficient energy transfer across a range of loads, including lighter loads where conventional designs are particularly prone to instability.

A genetic algorithm searches for better circuit settings

The researchers used a genetic algorithm to optimize the circuit. Inspired by natural selection, this type of artificial intelligence evaluates different configurations, selects promising candidates and adjusts their parameters over repeated rounds.

Three objectives guided the search: stable output voltage, efficient power delivery and low harmonic distortion. Harmonic distortion means that a signal contains additional frequency components beyond its intended frequency. Reducing it helps preserve signal quality and limits interference with other electronics.

The algorithm repeatedly tested and adjusted circuit parameters to find a design that performed well across different load conditions. This reduces the need for painstaking manual tuning of component values for each new device or application.

Sekiya, the professor leading the research and identified as its lead author, described how the virtual design process brings the modeling and optimization together:

Our machine-learning approach models the system virtually, runs simulations, and uses artificial intelligence to optimize performance by minimizing energy lost as heat and preserving signal quality.

The result is what engineers call load independent operation: maintaining stable output voltage and high efficiency as the receiving device's electrical demands vary. The numerical design process is intended to adapt to different loads, environments and uses by accounting for them during optimization.

What the laboratory system achieved

The team demonstrated its method on a class-EF WPT system combining a class-EF inverter with a class-D rectifier. An inverter produces the alternating electrical signal used to drive the transmitter, while a rectifier converts received alternating current into direct current. Both circuit types are known for efficiency and suitability for wireless power applications.

The experimental system held output voltage fluctuations below 5%, against fluctuations of up to 18% in conventional designs. Using those figures, the reduction is more than 13 percentage points. For a receiving device, a smaller voltage swing means a steadier electrical supply as its load changes.

Power transfer efficiency reached 86.7% at 6.78 MHz, with output exceeding 23 watts. Compared with the 65% efficiency cited for conventional systems under changing loads, that is a difference of 21.7 percentage points. Efficiency describes how much of the input power reaches the output rather than being lost along the way.

Performance also improved at light loads. The system kept energy dissipation in the transmission coil nearly constant across different loads, indicating effective current regulation and limited wasted energy. That behavior matters because a wireless charger needs to remain stable when a device draws less power as well as when demand is high.

Why the design could matter beyond phones

Electric vehicles, industrial robots and medical implants are among the applications identified for this approach. Their energy requirements can change rapidly or unpredictably, making stable delivery a practical concern rather than simply a laboratory target.

For vehicles, the challenge involves larger and more complex batteries whose charging loads can vary dramatically. Industrial equipment needs dependable power while operating demands change. Medical implants are another potential use for a design intended to keep power delivery consistent across varying loads.

The published account linked to ColombiaOne's wireless power article described the possible benefits for larger batteries and circuit construction:

The improvement could be especially important for larger, more complex batteries such as those in electric vehicles, where charging loads can vary dramatically. The method also simplifies circuit construction and could reduce cost and size, with the team aiming to make wireless power commonplace within five to ten years.

The potential savings come from simplifying the design process and circuit construction. Smaller, less costly systems could be easier to manufacture in large numbers and integrate into everyday products. These are prospective benefits as the laboratory method moves toward practical deployment.

The ambition is wireless power as familiar as Wi-Fi

The Chiba University team envisions wireless power becoming as common as Wi-Fi, with devices charging in homes, offices, factories and public spaces. A design that remains stable as loads change could make that ambition easier to pursue across equipment with different power requirements.

The expectation that wireless power could become commonplace within five to ten years is an outlook for adoption. Further engineering, rigorous safety testing and the development of industry standards remain necessary before broad commercial use.

Standards would help establish reliability and interoperability. Interoperability means that devices and charging equipment from different manufacturers can work together, an essential requirement for a technology intended to serve many types of products.

The chronology of the work and its prospects is short:

• June 2025: The Chiba University research was published in IEEE Transactions on Circuits and Systems.
• August 22, 2025: The reported results remained based on laboratory experiments, with wider applications still prospective.
• The following five to ten years: The team aims for wireless power to become commonplace as engineering, safety work and standards develop.

AI is changing how power circuits are designed

The wireless power project also illustrates a broader move toward artificial intelligence in electrical engineering. Designing inverters, rectifiers and other power electronics has traditionally required extensive expertise, repeated testing and substantial manual adjustment.

Evolutionary algorithms and large language models are among the AI tools being explored to streamline that work. Automated optimization can reduce development time, limit human error and help designers account for the complex behavior of real components.

A review available through ScienceDirect's power electronics review describes the potential of AI for circuit design:

AI-driven design optimizes high frequency wireless power transfer system parameters. GPT-based AI models significantly reduce design time and complexity in power electronics.

The review points to applications across power electronics, including transportation and the Internet of Things, or IoT. The latter refers to connected devices that exchange information. The Chiba University experiment uses a genetic algorithm to optimize a physical circuit, while the review describes a wider range of AI methods being applied to design tasks.

From experimental circuits to everyday equipment

The team's laboratory results offer a route toward more reliable wireless power systems across a range of applications. As AI assisted design methods mature and become more widely used, the same principles could support practical equipment that handles changing loads with less manual adjustment.

Industry observers envision environments in the next decade where devices charge automatically upon entering a room, vehicles recharge while parked or moving, and industrial equipment operates without power cables. These remain expectations for future applications.

Such uses could bring benefits beyond convenience, including reduced electronic waste, improved safety and new forms of connectivity and automation. For factories and other workplaces, removing power cables could also reduce the physical constraints on equipment.

The immediate achievement is more specific: a circuit designed through detailed numerical modeling and a genetic algorithm delivered steadier voltage and higher efficiency in laboratory testing. The practical task is to carry that performance into products that are reliable, safe and compatible with the equipment people already use.

Key Points

• Chiba University researchers led by Professor Hiroo Sekiya developed a machine learning method for designing stable, efficient wireless power systems.
• The research was published in IEEE Transactions on Circuits and Systems in June 2025.
• Output voltage fluctuations fell below 5%, compared with swings of up to 18% in conventional designs.
• The experimental system achieved 86.7% efficiency at 6.78 MHz and delivered more than 23 watts of output power.
• Detailed circuit models and a genetic algorithm accounted for real component behavior and optimized performance across changing loads.
• Potential applications include smartphones, electric vehicles, industrial equipment and medical implants.
• Wider adoption requires further engineering, safety testing and standards, with commonplace wireless power envisioned within five to ten years.