News

Amberella Unveils AI Domain Controller SoC for Multi-sensor Perception in AVs

January 10, 2022 by Abdulwaliy Oyekunle

To accelerate efficient performance in autonomous vehicles, the new AI domain controller from Ambarella promises robust single-chip processing for multi-sensor perception in an advanced driver-assistance system (ADAS).

As the world moves towards smarter, connected, and safer vehicles, more technology is needed to keep these systems at the cutting edge. 

One important device or electronic control unit (ECU) tool is domain controllers, which designers can use in a complex autonomous vehicle system. 

The domain controller is employed to control a wide range of autonomous vehicle functions, as well as in an ADAS, which targets safety systems by gathering and mapping sensor data from cameras, lidar, radar, sonar, and so on for path planning and decision making.

 

In ADAS, domain controllers are employed in different use cases including active safety, user experience, and body control.

In ADAS, domain controllers are employed in different use cases including active safety, user experience, and body control. Image used courtesy of Aptiv

 

As the automotive industry is shifting attention to autonomous vehicle systems, one company, Ambarella, hopes to take the performance in ADAS to the next level by introducing its new CV3 domain controller system-on-chip (SoC).

In this article, we'll consider how this new solution might accelerate better performance in ADAS and level 4 (L4) autonomy? And, how could this single-chip processor eliminate the need for developing separate software for the ECUs in autonomous vehicles?

 

Design Consideration for Domain Controllers in Autonomous Vehicles

Path planning and perception are two arduous tasks carried out by autonomous vehicles. 

Thus, in advanced autonomous systems, a reliable high computing microcontroller or processor with large bandwidth and memory is required to run and process the vast amount of streams of data gathered from the sensors.

 

The block diagram of the CV3 AI domain controller.

The block diagram of the CV3 AI domain controller. Image used courtesy of Ambarella

 

An ADAS domain controller utilizes logic libraries as part of its main processor components, enabling a high level of computing algorithms and manipulating data. 

An efficient logic library that includes adders and registers should be considered to balance the tradeoffs between numerical precision and the bandwidth of the memory available for computation.

Furthermore, the centralized SoC should be conscientiously designed to make timely decisions such as emergency braking, pedestrian detection, and so on based on the data gathered and mapped from the sensors.

Above all else, safety is a major concern in ADAS. The components in ADAS electronic control units should be designed to meet safety practices and operations. 

Generally, the SoC should meet international standard safety requirements such as the ISO 26262, and may also include TS 16949 quality management standard and the automotive SPICE (ASPICE), which is based on automotive quality management processes.

In an attempt to improve the performance of the domain controller system, designers could also consider advanced FinFET process technologies. These process technologies could enable the integration of a wide array of critical functions such as image processing and neural network processing on a single integrated circuit (IC) die.

 

Introducing the Novel CV3 AI Domain Controller

Hoping to keep up with advancing ADAS, Ambarella's latest domain controller SoC, the CV3, is designed to handle neural network algorithms. 

Additionally, the product offers up to 500 eTOPS of CVflow artificial intelligence processing for neural network computation.

As for performance, it claims to be accelerated in ADAS and L2+ to L4 autonomous vehicles, as its CVflow computer vision AI processor provides high throughput, low latency, and low-power neural network processing.

The new system-on-chip also features a general vector processor (GVP) for computer vision processing. Its processing cores are up to 16 Arm Cortex -A78AE and an automotive graphics processing unit (GPU).

To enable the gathering of a large chunk of image and video streams of data, the CV3 has an image signal processor (ISP) that provides a high-resolution of image and video recording. This ISP claims to help reveal clear images in low-lighting conditions and streams low bit rates of videos with efficient encoding in H.265 and H.264 formats.

 

The CV3 domain controller claims to enable autonomous vehicles to make a good decision for path planning.

The CV3 domain controller claims to enable autonomous vehicles to make a good decision for path planning. Image used courtesy of Ambarella

 

Commenting on the CV3 AI domain controller, Fermi Wang, President and CEO of Ambarella, believes that its on-chip ISP and simultaneous radar processing that utilizes its Oculii adaptive AI algorithms, Ambarella can help the automotive industry open up greater levels of perception and accuracy across all environmental conditions.

The product can be chiefly used in applications that involve a single- / multi-camera advanced driver-assistance system and parking assistance systems.

All in all, this domain controller SoC hopes to fill a need in ADAS and autonomous vehicle systems. 

 


 

Interested in other ADAS and automotive industry news? Read on in the articles down below.

Banned ADAS Headlight Technology Could Prevent Millions of Accidents

Startup Arbe Hopes to Take Center-stage in ADAS with 4D Radar

Ditching Radar: Tesla Bets on Computer Vision for the Future of ADAS