🔬 applied physics

A Wide-Regulation-Range Hybrid Switched-Capacitor Converter for 48V Automotive Power Systems

This paper presents a hybrid switched-capacitor converter with a novel multi-mode modulation scheme and reinforcement learning-assisted control that achieves wide-range voltage regulation with soft-switching capabilities and high efficiency for 48V automotive power systems.

Original authors: Georgios Spanodimos, Guanyu Qian, Xiaodan Cui

Published 2026-04-17

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Georgios Spanodimos, Guanyu Qian, Xiaodan Cui

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine your car is evolving. It's no longer just a machine for getting from point A to point B; it's becoming a supercomputer on wheels. With self-driving features and advanced AI, your car now needs to power massive amounts of electronics.

To handle this, modern cars are switching from a standard 12-volt battery system to a more powerful 48-volt system. Think of this like upgrading from a small garden hose to a fire hose: you can move the same amount of water (power) with much less pressure (current), which means less energy is wasted as heat and the wiring can be thinner and lighter.

However, there's a problem. While the main "fire hose" is 48 volts, many of the car's new gadgets (like the AI brain or the infotainment screen) need very low voltages, like 5 or 12 volts. You need a way to step that 48 volts down efficiently without wasting energy or making the system heavy and bulky.

This paper introduces a new, clever solution called a Hybrid Switched-Capacitor Converter (HSCC). Here is how it works, explained simply:

1. The Old Way vs. The New Way

The Old Way (The Heavy Lifter): Traditional converters often use big, heavy transformers (like the ones in old power bricks). They are good at changing voltage, but they are heavy and take up a lot of space. In a car, every gram counts because weight kills fuel efficiency.
The New Way (The Light Dancer): This new converter uses capacitors (which store electricity like a bucket stores water) instead of heavy transformers. It's much lighter and smaller. However, traditional capacitor-based converters are like a rigid robot: they can only step down voltage in one specific ratio (like 48V to 24V). If you need 12V or 5V, they struggle or become inefficient.

2. The "Three-Step Dance" (The Innovation)

The authors invented a new "dance routine" for the electricity to follow.

The Old Dance: The electricity would jump from the input to the output in a fixed pattern. It was efficient, but only if the output voltage was exactly half the input.
The New Dance (3-Mode Modulation): The new converter adds a third step to the dance.
- Step 1: Charge the bucket.
- Step 2: The Secret Move: Instead of just waiting, the system briefly dumps some energy to the ground to "reset" the bucket's level. This allows the system to change how much voltage it steps down, just like a dimmer switch changes light brightness.
- Step 3: Deliver the power to the load.

Because of this extra step, the converter can smoothly adjust from stepping 48V down to 12V, all the way down to 5V, without getting stuck.

3. The "Soft Landing" (Zero Switching Loss)

When you flip a light switch, you sometimes hear a tiny click or see a spark. In electronics, this "spark" is wasted energy and heat.

Hard Switching: Turning a switch on or off when there is still voltage or current flowing is like slamming a car door. It's loud, damaging, and wastes energy.
Soft Switching (ZCS/ZVS): This new converter is like a magic door that only opens when the hallway is empty.
- It waits for the electrical current to hit zero (Zero-Current Switching) or the voltage to hit zero (Zero-Voltage Switching) before flipping the switch.
- This means no sparks, no heat, and maximum efficiency. The paper shows this works beautifully even when the car is running at different speeds (load conditions).

4. The "AI Coach" (Reinforcement Learning)

Controlling this converter is tricky because the rules change depending on how much power the car needs at that exact moment. It's like trying to drive a car where the steering wheel feels different every time you turn it.

Instead of a human engineer trying to guess the perfect settings, the authors used Reinforcement Learning (RL).
Think of this as an AI coach. The AI played thousands of simulations, trying different settings. When it did a good job (fast response, stable voltage), it got a "reward." When it messed up, it got a "penalty."
Eventually, the AI learned the perfect way to tune the controller to handle any situation instantly, ensuring the car's electronics get steady power even when the AI brain suddenly needs a massive power boost.

The Results: Why Should You Care?

The team built a physical prototype (about the size of a small smartphone) and tested it.

Efficiency: It achieved over 92% efficiency. This means almost all the energy from the battery goes to the electronics, with very little lost as heat.
Versatility: It can handle a wide range of voltage drops (from 48V down to 5V) without losing that high efficiency.
Weight: It is incredibly light (only 22 grams), which is a huge win for electric vehicle range.

In a nutshell: This paper presents a new, lightweight, and super-efficient power adapter for the cars of the future. It uses a clever three-step dance to adjust voltage smoothly and an AI coach to keep everything running perfectly, ensuring your self-driving car has the power it needs without draining the battery.

1. Problem Statement

Modern automotive architectures are transitioning from 12V to 48V low-voltage buses to support high-power AI workloads, autonomous driving systems, and electrified loads. This shift requires Power Conversion Systems (PCS) that can:

Handle a wide range of voltage conversion ratios (from 48V down to 12V for legacy systems, and sub-5V for computing XPUs).
Maintain high efficiency and power density.
Provide tight voltage regulation with fast transient response.

Limitations of Existing Solutions:

Transformer-based converters (e.g., DAB, LLC): While capable of high step-down ratios, they suffer from high weight, volume, and magnetic losses (leakage inductance, core losses), which hinder power density improvements.
Traditional Switched-Capacitor Converters (SCCs): Offer high power density but are typically constrained to a fixed 2:1 conversion ratio. Operating away from this natural ratio leads to hard-switching losses and sharp efficiency degradation.
Control Challenges: Existing resonant SCCs often rely on open-loop operation or simple frequency modulation. Closed-loop control for variable-ratio resonant converters is difficult due to nonlinear dynamics and variable-frequency operation, making traditional linear controller tuning (like PI) suboptimal.

2. Methodology

The authors propose a Hybrid Switched-Capacitor Converter (HSCC) featuring a novel Three-Mode (3M) Modulation scheme and an advanced control strategy.

A. Converter Topology

The HSCC utilizes a Flying-Capacitor Multi-Level (FCML) topology with a small output resonant inductor ( $L_r$ ). This replaces bulky magnetic transformers with capacitors for energy transfer, significantly reducing size and weight.

B. Three-State Modulation Scheme (3M)

Unlike conventional 2-state resonant operation, the proposed scheme introduces a three-state operating sequence to achieve variable step-down ratios while preserving soft-switching:

State 1 ( $T_1$ ): The flying capacitor ( $C_r$ ) is charged from the input ( $V_{in}$ ).
State 2 ( $T_2$ ): $C_r$ is disconnected (floating), and the output capacitor ( $C_{out}$ ) discharges to ground. This state allows $V_{Cr}$ to deviate from the fixed $V_{in}/2$ steady-state value.
State 3 ( $T_3$ ): $C_r$ delivers charge to the output.

By controlling the duration of State 1 ( $T_1$ ) and allowing States 2 and 3 to be determined by zero-crossing events, the converter achieves a variable voltage conversion ratio (ranging from 2:1 to 5:1).

C. Soft-Switching Strategies

Zero-Current Switching (ZCS): Achieved for transitions between State 2 $\to$ 3 and State 3 $\to$ 1. An adaptive zero-crossing detection algorithm automatically identifies these instants to ensure zero-current switching, minimizing switching losses.
Zero-Voltage Switching (ZVS): An additional discharge interval ( $T_{dis}$ ) is introduced to fully discharge the parasitic output capacitance ( $C_{OSS}$ ) of the switches before turn-on. This is particularly beneficial for light-load efficiency.

D. Control Architecture

Closed-Loop Regulation: A Proportional-Integral (PI) controller regulates the output voltage by adjusting the duty cycle of State 1 ( $T_1$ ).
Reinforcement Learning (RL) Tuning: Due to the converter's nonlinear, variable-frequency nature, traditional analytical tuning is insufficient. The authors employ an offline Reinforcement Learning (RL) agent (using Proximal Policy Optimization) to tune the PI controller gains. The RL agent optimizes for voltage tracking error, settling time, and stability without requiring an accurate analytical system model.

3. Key Contributions

Novel 3M Modulation: A three-state operating sequence that enables continuously variable step-down conversion ratios (beyond the fixed 2:1 limit) while maintaining soft-switching conditions.
Hybrid Soft-Switching: Simultaneous support for both ZCS and ZVS modes, ensuring high efficiency across a broad range of load and conversion conditions.
RL-Based Control Tuning: A control framework that integrates RL to tune PI parameters for complex, nonlinear resonant converters, achieving fast transient response (settling in ~6 cycles) without complex modeling.
High Power Density Design: A prototype that eliminates heavy transformers, achieving a total weight of only 22g for a 100W converter.

4. Experimental Results

A hardware prototype was built and tested with the following specifications:

Input: 24V (scaled from target 48V for safety).
Output: 5V – 10V (Conversion ratios of 0.2 to 0.4).
Power: Up to 100W.
Components: GaN FETs, 20µF resonant capacitor, 150nH inductor.

Key Performance Metrics:

Efficiency:
- Peak Efficiency: Exceeded 92% at 100W for conversion ratios of 0.3 and 0.4.
- Wide-Range Efficiency: Maintained >88% efficiency across all loads for a 3:1 conversion ratio.
- Note: Efficiency dropped slightly at the lowest ratio (0.2) due to non-ZCS switching of the high-side switch, suggesting a need for multiphase operation in future iterations.
Regulation: Successfully demonstrated closed-loop regulation with a 6-cycle settling time under 10% load transients.
Soft-Switching: Waveforms confirmed ZCS operation over the 0.2–0.4 ratio range and successful ZVS turn-on for the low-side switch.
Density: The 22g prototype is significantly lighter than comparable transformer-based solutions.

5. Significance

This work addresses a critical bottleneck in next-generation automotive power systems: the need for high-efficiency, high-density, wide-range DC-DC conversion.

Scalability: The proposed HSCC can serve as a versatile front-end converter (48V $\to$ 12V) or a direct Point-of-Load (PoL) converter (48V $\to$ sub-5V), simplifying the power distribution network.
Efficiency vs. Flexibility Trade-off: It breaks the traditional trade-off where variable-ratio converters suffer from efficiency losses, proving that soft-switching can be maintained even outside the natural resonant ratio.
Control Innovation: The integration of RL for controller tuning offers a new paradigm for managing the complex dynamics of resonant converters, potentially applicable to other nonlinear power electronic systems.

In conclusion, the paper demonstrates that Hybrid Switched-Capacitor Converters with multi-mode modulation are a viable, high-performance solution for the evolving 48V automotive power architecture, offering a compelling alternative to traditional magnetic-based converters.