Renesas Electronics Corporation and Fixstars Corporation, a developer of multi-core CPU/GPU/FPGA acceleration technology, announced the joint development of a suite of tools that enables rapid software optimization and simulation for autonomous driving systems (AD ) and advanced driver assistance systems (ADAS) specifically designed for Renesas' R-Car system-on-chip (SoC) devices. These tools enable rapid development of network models with highly accurate object recognition from an early stage, leveraging the performance of the R-Car SoC.
"Renesas continues to create integrated development environments that enable customers to take the software-first approach, " said Hirofumi Kawaguchi, vice president of Automotive Software Development at Renesas. "By supporting the development of deep learning models tailored for R-Car, we help our customers create AD and ADAS solutions while reducing time-to-market and development costs."
"GENESIS for R-Car, which is a cloud-based evaluation environment created together with Renesas, allows engineers to evaluate and select devices early in development cycles and has already been used by many customers," said Satoshi Miki, Ceo of Fixstars. "We will continue to develop new technologies for the purpose of accelerating machine learning operations (MLOps), which can be used to maintain the latest versions of software in automotive applications."
Today's AD and ADAS applications use deep learning to achieve highly accurate object recognition. Deep learning inference processing requires huge amounts of computation and memory capacity. Models and programs executable on automotive applications must be optimized for a dedicated SoC, this is because real-time processing with limited arithmetic units and memory resources can be a challenging task. In addition, the software evaluation process during verification must be accelerated, and updates must be applied repeatedly to improve accuracy and performance.
The tools developed by Renesas and Fixstars.
- R-Car Neural Architecture Search (NAS) tool for generating network models optimized for R-Car
This tool generates deep learning network models that efficiently use the CNN (convolutional neural network) accelerator, DSP, and memory on the R-Car device. This allows engineers to quickly develop lightweight network models that achieve extremely accurate object recognition and fast processing times even without deep knowledge or experience with the R-Car architecture.
- R-Car DNN Compiler for compiling network models for R-Car
This compiler converts optimized network models into programs that can take full advantage of R-Car's performance potential. It converts the network models into programs that can be executed quickly on the CNN IP and also performs memory optimization so as to maximize performance in the context of high-speed but capacity-constrained SRAM memory.
- R-Car DNN Simulator for rapid simulation of compiled programs
This simulator can be used to quickly test the operation of programs on a PC, rather than on the actual R-Car chip. Using this tool, developers can generate the same operational results that would be produced by R-Car. If the accuracy of inference processing recognition is affected during the process of creating lighter models and optimizing programs, engineers can provide immediate feedback to model development, thus shortening development cycles.
Renesas and Fixstars will continue to develop deep learning software with the joint "Automotive SW Platform Lab" program and create operating environments that maintain and improve recognition accuracy and performance by continuously updating network models.
You might also be interested in:
