Our Solution Concept
PowerCare is based on the institutes activities and results in the three main target areas of the project:
-
Novel, vertical GaN trench MOSFETs and their behavioral models (Fraunhofer ISIT)
-
Embedded AI models integrated in a PWM controller for failure prediction of electric motors and GaN power semiconductors (Fraunhofer IMS)
-
Demonstration of GaN MOSFETs and intelligent motor control (Fraunhofer IISB)
System concept illustration.
Our Core Technologies
Vertical GaN Transistor
Vertical GaN power transistors combine the performance advantages of vertical wide band gap (WBG) transistors with the cost advantages of established silicon technology. In due time, they can replace IGBTs to reduce power conversion losses in many price-sensitive applications ranging from data center power supplies to electric vehicle traction inverters, establishing GaN as semiconductor of choice beyond consumer electronics. In PowerCare vertical GaN trench MOSFETs based on engineering substrates are being developed for applications at 48V with current rating between 35-160A per device. Higher voltage levels will be evaluated as well.
PWM Controller with Embedded AI
Failure models for inverters as well as for the connected electric motors are developed and ported to a RISC-V based power module for in situ execution. A PWM controller with RISC-V architecture will be extended with hardware accelerators for the functions
important for the models (e.g., FFT, filtering). The model size and execution speed will be optimized for real-time requirements (e.g., quantization or pruning).
Hybrid Models
The failure model for electric motors will consider
changes in load current and optionally other sensor data (vibration, acoustics, instantaneous rotation rate) that can be observed due to impending failures. Starting with the detection of bearing damage, the development will cover further faults, such as demagnetization or winding faults.
​
The failure model of the transistors and inverters are developed based on data from life tests and parameter measurements and enable the training of the failure prediction.
Intelligent Power Module
A GaN-based and AI-capable power module will be built, into which the trained failure models are integrated and executed locally, with the current and sensor data from the inverter and motor serving as input parameters.
