Efficient utilization of vector DSP architectures for upcoming automotive applications trends

Conference: AmE 2021 – Automotive meets Electronics - 12. GMM-Symposium
03/10/2021 - 03/11/2021 at online

Proceedings: GMM-Fb. 99: AmE 2021

Pages: 6Language: englishTyp: PDF

Walluszik, Konrad; Hassan, Muhammad; Schaefer, Juergen (Infineon Technologies AG, Neubiberg, Germany)

Among the major megatrends in the automotive industry is the development of safer, smarter and more eco-friendly cars. This development requires innovative products for future automotive applications which drives the development of the E/E architectures towards zone or domain based architectures. Such a hierarchical concept enables the usage of sensor information across multiple system functions. Such new functions which either extracts more precise data out of a sensor signal or take multiple system states into consideration require additional performance in order to address the computational demands of signal processing. Vector DSP architectures based on a SIMD concept offer a new performance class besides scalar CPUs in next generation of automotive microcontroller products. Especially for larger matrix/vector based operations a huge performance gain is visible. But even for algorithms with smaller matrix/vector operations we are demonstrating a high HW utilization considering the application context. This allows to accelerate algorithms like neural networks, Kalman Filters or other sensor signal processing algorithms significantly compared to scalar CPU architectures. This in turn can be used to enhance the complexity of signal processing while still staying within the applications real-time performance budget.