Mobileye rejects FDSOI for EyeQ chip

May 23, 2016 // By Peter Clarke
EyeQ5
Mobileye NV has turned away from the fully depleted silicon-on-insulator manufacturing (FDSOI) process technology promoted by ST and is aiming at 10nm or finer FinFET process for its 5th generation of EyeQ SoC for driving vision systems.

Mobileye NV has collaborated with STMicroelectronics for more than a decade on the EyeQ series of vision processors and the two companies have announced that they have started co-developing the EyeQ5. However, with a FinFET process as the target, it will likely be supplied by either TSMC or Samsung.

The EyeQ5 is expected to act the central computer for the fusion of images for fully autonomous driving vehicles for deployment in 2020. The chip will have eight dual-threaded CPU cores and eighteen vision processor cores and will increase the maximum performance by a factor of 8 over that of the previous generation EyeQ4.

To meet power consumption and performance targets, the EyeQ5 will be designed in advanced 10nm FinFET technology or an even smaller node, the companies said. As recently as January 2016 Jean-Marc Chery, ST's COO, was stating that the EyeQ4 processor from Mobileye NV would be manufactured using FDSOI. It now seems that however good FDSOI may be for computation efficiency more credible roadmaps down to 10nm and 7nm is persuading developers to select FinFET process technology.

The EyeQ5 is expected to be capable of more than 12 Tera operations per second, while keeping power consumption below 5W, to maintain passive cooling. Engineering samples of the EyeQ5 are expected to be available in the first half of 2018.

Although when ST first became involved with Mobileye it was as a manufacturing partner the co-operation continues with ST now providing support in terms of physical design and providing IP such as memory and high-speed interfaces, and system-in-package design to ensure the EyeQ5 meets automotive qualification. ST will also contribute to the safety- and security-related architecture of the product.

The EyeQ5 accelerator cores are optimized for a variety of functions in four categories: deep neural networks; machine learning; signal processing; and computer vision. The diversity of the accelerators means applications can use the core most suitable for each task thereby