Power Semiconductors in Robotics How Infineon Powers the Joints of Humanoid Robots with GaN

Related Vendors

FAULHABER Drive Systems

Taiwan Machine Tool & Accessory Builders' Association (TMBA)

SIMTOS (Seoul International Machine Tool Show)

GaN is revolutionizing the drive technology of humanoid robots, but the challenges go far beyond compact joints. A powerful nervous system consisting of Ethernet and quantum-resistant hardware security determines the success of the machines.

The robotics market is undergoing rapid expansion. Driven by an aging population and the widespread labor shortage, artificial intelligence is increasingly merging with the physical world. The industry refers to this as "Physical AI," or physical artificial intelligence. Machines are evolving from mere automatons to intelligent systems that perceive, understand, and interact with their surroundings.

At the power electronics trade fair PCIM, semiconductor manufacturer Infineon addressed this technological challenge. Adam White, Head of Power & Sensor Systems at Infineon, sees his company playing a key role in this evolution: "We are shaping the industry from the power grid to the data center to Physical AI," White explains. Modern humanoid robots, perhaps the most complex mechanical systems of our time, often feature up to 70 motorized joints. Controlling these components presents developers with massive challenges.

Gallery

This is precisely where conventional semiconductors reach their limits. The electronics for motor control account for almost half of the entire semiconductor bill of materials in a robot. If these components are too heavy and bulky, a physical chain reaction occurs: heavier components require stronger motors, which in turn consume more power, drastically reducing the already limited battery life.

The GaN Advantage for Compact Joints

To break this cycle, semiconductor manufacturers like Infineon are increasingly relying on gallium nitride, or GaN. This innovative material enables significantly higher switching frequencies, which dramatically reduces the size of the required passive components and, consequently, the PCB footprint. Accordingly, Infineon showcased reference designs at the trade fair that achieve an extreme power density of 3.3 kilowatts per cubic inch.

"We know very well that, for example, in an elbow joint, the size is crucial because no one wants bulky and heavy motors," White explains this necessity. He adds: "Moreover, a bulky and heavy motor naturally requires a lot of energy."

The efficiency gains are also evident in thermal management. Nenad Belancic, Head of Application Management for Robotics at Infineon, emphasizes: "It's not just about size. It's also about the thermal performance you can achieve with GaN." The higher frequencies reduce the thermal power loss in the system. According to Belancic, this "not only relieves the boards but also the motor, as it cools there." Previously common, heavy cooling systems can thus be drastically reduced in size.

Dexterity: The Electronics of the Robotic Hand

Even more complex than an elbow is replicating the human hand. The so-called "dexterous hand" is considered one of the greatest challenges in robotics. To delicately grasp objects and operate tools, electronics and mechanics must work perfectly together in the smallest of spaces.

Infineon relies on highly integrated motor controllers from the Motix family, which enable exceptionally high component density. The drives are complemented by a phalanx of sensors. Radar systems, microphones, and specialized image processing chips help the robot with orientation, but for pure motion, internal sensors are crucial. Current and position sensors capture even the smallest movements of the finger joints in real time. "Dexterity for the hand will be the critical path," predicts Belancic, referencing specialized sensors that ensure safe and precise feedback.

System Architecture and Cost Drivers in Motor Control

Depending on the safety requirements, conventional PSOC microcontrollers or high-security Aurix microcontrollers from the automotive sector control the joints. Economically, actuation remains the most expensive segment of the robot. Excluding the costly central processing units for machine learning (such as large GPUs), Infineon calculates its own semiconductor costs to average 500 US dollars per robot.

Belancic comments: "If you ask me, this number will certainly rise in the coming years because these robots' capabilities will increase." He identifies "actuation, data flow, and human-like behavior to support us humans" as key drivers for the growing demand for semiconductors.

Coexistence of Technologies: GaN Does Not Replace Silicon

Will gallium nitride completely replace conventional semiconductors like silicon in robotics? The experts clearly deny this. Instead, a parallel use will be established. Adam White sees the strengths of gallium nitride in areas where high switching frequencies necessitate miniaturization. However, he adds: "Silicon still offers significant advantages in terms of cost-performance ratio for a number of other joints that our customers are interested in."

Belancic also anticipates a long-term coexistence: "We will see both in the coming years." While integrated GaN modules offer clear advantages for joints from the shoulder downwards, he sees different requirements for heavily stressed components. "For the absolute high-performance joints, I believe the transition will be slower," says Belancic. In areas with extremely high torque, robust silicon solutions will retain their justification for the time being.

The Nervous System: Why Ethernet Resolves Data Bottlenecks

In addition to sheer driving force, a humanoid robot requires a gigantic nervous system to process the incoming masses of data. Cameras, microphones, and hundreds of sensors in the joints continuously generate information. "There is more and more data being generated," explains White. "You need the ability to transfer this information as quickly and effectively as possible, without latency."

While fieldbuses have traditionally dominated industrial plant networking, a technology known from IT is now moving to the center of the robotics chassis: Ethernet. The network protocol provides the necessary bandwidth to enable motors and sensors to communicate synchronously and without delay. "We believe Ethernet will become highly standardized in certain humanoids and in robotics within industrial architecture," White states confidently. Infineon is already supporting this development with concrete reference designs, including partnerships with the Nvidia platform Thor.

Hardware Security in the Age of Quantum Computers

However, the increasing networking and autonomy of Physical AI also entail immense risks. A robot that moves freely in factories or even in public spaces in the future represents a potentially devastating target for cyberattacks. Functional safety protects humans from the machine's malfunctions—but cybersecurity protects the machine from hackers.

"We want to ensure they are not taken over by external or malicious actors," warns White. The company is therefore integrating authentication chips and platform security solutions directly at the board level. A particular focus is already placed on post-quantum cryptography (PQC). Since future quantum computers could easily break today's encryptions, engineers are now implementing quantum-resistant hardware root of trust into the architecture of robots. "We don't just want to think about today, but also about tomorrow," White explains this forward-thinking step.

Accelerated Market Maturity through Global Partnerships

To accelerate the development of humanoid robots from initial design to mass production, semiconductor architects are increasingly relying on strategic collaborations. A recent example is a new partnership with the Vietnamese developer Vinrobotics. Both companies announced plans to establish a joint center of excellence in Hanoi. While the subsidiary of the Vingroup conglomerate contributes its expertise in mechanics and artificial intelligence, Infineon provides the complete electronic foundation, from sensors and power switches to cryptographic security chips. The choice of the Asian location is strategically motivated. The Southeast Asian region is rapidly developing into a massive hub for production, significantly driving the demand for automation solutions.

Virtual Training and the Evolution of Mechanics

Before these highly secure, connected, and intricate robots go into mass production, they must learn. This learning process is increasingly shifting to virtual spaces. Using digital twins, the systems complete millions of training cycles in simulation before taking their first physical step. Semiconductor manufacturers provide models of their hardware so developers can test the exact thermal and electrical behavior of the chips in advance on a computer.

The electronics industry thus provides all the tools for the age of Physical AI. The bottleneck is now increasingly shifting toward classic actuation. To fully exploit the efficiency advantages of the high switching frequencies of the new power electronics, new electromechanical concepts are required.

Belancic holds suppliers accountable: "This is not just about semiconductors. The motor industry must also address new motor technologies." Only when this bridge between electronics, software, and cutting-edge mechanics is achieved will nothing stand in the way of widespread use of humanoid robots. 

Actuator Design and Edge AI

Improved Motion Quality for Humanoid Robots