AURIX TC2xx and TC3xx microcontrollers are used in numerous automotive applications, and recently, also in some which include machine learning tasks. Yet, these applications are mainly engineered manually, and only little tool support exists for bringing neural networks to TriCore microcontrollers. Thus, we propose OpTC, an end-to-end toolchain for automatic conversion, code generation, and deployment of neural networks on TC3xx microcontrollers. OpTC supports various types of neural networks and provides compression (e.g., pruning) and optimization (e.g., approximation of special activation functions) techniques often required to reduce execution time and memory footprints while maintaining accuracy. The toolchain incorporates cost models for estimating memory utilization and execution time of a neural network model. The flexibility in supporting different types of neural networks, such as multi-layer perceptrons (MLP), convolutional neural networks (CNN) and recurrent neural networks (RNN), is shown in case studies for a TC387 microcontroller. Automotive applications for predicting the temperature in electric motors and detecting anomalies are thereby used to demonstrate the effectiveness and the wide range of applications supported by OpTC.
Session:
Poster
|