Interview: Optimizing AI Models for Automotive SoCs

In developing AI-based driver assistance systems, Bosch relies on its own optimization toolchain to tailor neural networks specifically to the target hardware. As the company explains in an interview with Markt&Technik, the focus is not on developing new neural networks, but rather on their efficient implementation on automotive SoCs. Bosch explains how its proprietary AI optimization toolchain tailors models to the target platform through hardware-aware transformations, quantization, and benchmarking. Key factors here include not only latency and computing power, but also perception quality, determinism, and real-time capability. In the interview, Texas Instruments complements this perspective with details on the new TDA5 SoC family and the C7 NPU. In addition to high energy efficiency and scalable AI computing power, the focus is particularly on the heterogeneous architecture and integrated security features. Also of interest is the close collaboration between the two companies in optimizing transformer-based models such as BEVFormer, as well as the use of a Virtualizer Development Kit that enables software development and validation even before the hardware is available. (oe)

Go to the post