OpTC - Automatic Compression and Performance Estimation for Deployment of Neural Networks on AURIX TC3xx Microcontrollers
Heidorn C, Hannig F, Riedelbauch D, Strohmeyer C, Teich J (2026)
Publication Language: English
Publication Type: Journal article
Publication year: 2026
Journal
Abstract
Compression and quantization techniques are key for deploying neural network (NN) models on highly resource-constrained microcontroller architectures. Particularly for AURIX TriCore microcontrollers, widely used in automotive applications, there is a lack of support for automatic compression, conversion, and deployment of pre-trained NN models. Therefore, in this paper, we present OpTC, a toolchain designed to bring neural networks to TriCore microcontrollers. OpTC achieves this by employing automatic sensitivity-based NN pruning and quantization. Testing whether a given neural network fits on a microcontroller, by compiling, deploying, and then evaluating memory and performance constraints can be quite time-consuming, particularly when applying our automatic compression scheme that iterates multiple pruned variants of an NN. To speed up the deployment process in our design flow, we contribute a cost modeling approach for TriCore microcontrollers that estimates the required ROM, RAM, and execution time for given neural networks without requiring compilation or explicit measurements on the target. We evaluate our approach for selected applications, including an autoencoder and a convolutional neural network for keyword spotting from the MLPerf Tiny benchmark. Our experiments show that OpTC can find compressed NN variants, that significantly reduce the memory requirements and execution time without increasing the error over the baseline model.
Authors with CRIS profile
Christian Heidorn
Lehrstuhl für Informatik 12 (Hardware-Software-Co-Design)
Frank Hannig
Lehrstuhl für Informatik 12 (Hardware-Software-Co-Design)
Jürgen Teich
Lehrstuhl für Informatik 12 (Hardware-Software-Co-Design)
Related research project(s)
Involved external institutions
Germany (DE)How to cite
APA:
Heidorn, C., Hannig, F., Riedelbauch, D., Strohmeyer, C., & Teich, J. (2026). OpTC - Automatic Compression and Performance Estimation for Deployment of Neural Networks on AURIX TC3xx Microcontrollers. Communications in Computer and Information Science.
MLA:
Heidorn, Christian, et al. "OpTC - Automatic Compression and Performance Estimation for Deployment of Neural Networks on AURIX TC3xx Microcontrollers." Communications in Computer and Information Science (2026).
BibTeX: Download