ADAS Mapping by ARM

Towards 2020

Advanced Driver Assistant System (ADAS) is a key application leading towards autonomous drive in the not distant future. Almost all car OEMs and tier 1s are developing and competing with each other with the aim to introduce semi-autonomous driving vehicles into market by 2020. ADAS requires much higher computing performance than existing other automotive applications. It needs sophisticated technologies such as mobile and consumer electronics tailored for automotive qualifications which are quite huge challenges for today’s automotive industries.

Under such circumstances, a great deal of attention is paid to ARM processors for their unparalleled performance with high power efficiency and broad ecosystem, which will help a lot for the automotive space.

Actually ADAS contains very wide applications but here we want to focus on camera, sensor and radar related applications and discuss how ARM processors will be used in the near future.

Simplified ADAS application workflow is described below: sensing, perception, decision making (sometimes displayed to the monitor/dashboard) and then actuated by chassis application like braking or steering system.

For sensing, a multitude of sensors such as radar, lidar, cameras, MEMS or ultrasonic sensors are used here.

• Radar/Lidar: Already widely used and established in the automotive and related technologies, they are good at measuring the distance and finding objects even in bad weather or night time. The price points have been decreasing for mass production. High performance CPU such as ARM® Cortex®-R4, Cortex-R5 and Cortex-M7, associated with real time responses taking into account safety critical events and information are required for this application.

• Front/360° view camera: This emerging application requires the highest computation power ever for the SoC. There are several approaches to meet requirements and also a lot of algorithms are developed. It is getting more common to use not only CPU but also additional hardware accelerator like DSP, FPGA or GPU to achieve higher performance under the constrained power and space. Cortex-A15 or ARMv8-A processors supporting 64-bit such as Cortex-A57 and Cortex-A53 will be main streams in this space and ARM Mali GPU can be also used for GPU compute acceleration.

• MEMS/Ultrasonic sensor: A huge amount of sensors are used today in automotive from affordable cars to premium cars. Ultrasonic sensors are mainly used for parking assistant systems to measure distance and MEMS like G-sensors to detect acceleration and collision. The number of these sensors per vehicle will increase to obtain more precise data, but on the other hand the cost will decrease, as is the common trend in automotive semiconductor. To achieve higher performance with high power efficiency and affordable price than 8, 16-bit MCU or current 32-bit MCU, the ARM Cortex-M0+, Cortex-M3 and Cortex-M4 are very suitable CPUs here.

Possibility of Data Fusion in Vehicle

ADAS will bring a lot of extra MCUs and SoCs into vehicles. They will start communicating or cooperating with chassis applications like braking system or steering system via domain controllers. Communication via automotive Ethernet, data fusion seems to be inevitable to happen to handle enormous amount of data but it is not clear yet what kind of new SoC will be needed or if improved Graphic SoC (for GPU compute) to handle the extra job. Perhaps real-time CPUs supporting functional safety related to ISO26262 will be needed to make decisions. Nevertheless, the performance requirements for the SoC for data fusion should be higher than ever. I’m thinking that if heterogeneous computing is to happen in automotive space in the future, this will be the place. We offer ARMv8-A processors supporting 64-bit, ARMv8-R architecture supporting virtualization by hardware with real-time response, and Mali GPU for GPU compute and ARM’s announced Open CL support for NEON for heterogeneous multiprocessing and parallel computation for performance and efficiency.

Summary

ADAS, which is definitely required to support automotive quality and safety, at the same is very similar to mobile and consumer application technologies with rapid development cycle. ARM’s solutions in terms of technologies and ecosystem address such complicated requirements. ADAS is a unique domain where cooperation or co-development with third-party of software companies is needed to remain competitive or to develop new algorithms for fast time to market. ARM’s ecosystem with its broadness and diversity, enables the automotive industry to overcome the difficulties it faces today by providing new partnerships and to leap into the next stage.