Mobileye Announces EyeQ6 And EyeQ Ultra Chips For Assisted And Automated Driving

Mobileye and its parent company Intel

INTC
may not be attending CES 2022 in person, but they are making some news.The Israel-based market leader in vision-based advanced driver assist systems (ADAS) has announced two new generations of chips and new or expanded partnerships with three automakers. The new EyeQ6 family is expected to launch in mid-2023 for ADAS and the EyeQ Ultra is expected to power level 4 automated vehicles from 2025. 

Mobileye has dominated the market for vision-based ADAS since the original EyeQ chips shipped in 2004, but the company has faced significantly increased competition recently. Nvidia

NVDA
has been winning many production programs in recent years, especially in the Chinese market with both its Xavier system on-a-chip (SoC) for partially automated ADAS and its Orin SoC for more fully automated systems. 

Qualcomm

QCOM
has also been gaining traction with its Snapdragon Ride family of SoCs which was selected by General Motors

GM
for new ADAS applications beginning in 2022 as well as the Ultra Cruise system coming in 2023. Like Mobileye, both Nvidia and Qualcomm are also supplying full software stacks for ADAS and automated driving with their SoCs. 

Mobileye’s current mainstream chip is the EyeQ4 which debuted in late 2018 and is in millions of vehicles on the road today. Among the capabilities of EyeQ4 is support for Road Experience Management (REM), Mobileye’s crowd-sourced high-definition mapping system. Roadbook, the map product powered by REM is already being used by Nissan in the hands-free ProPilot Assist V2 that debuted in Japan in 2019 and is coming to the US market later in 2022 on the Ariya. REM data is also available to other map providers. 

From mid-2023, EyeQ6 will become Mobileye’s main ADAS SoC in two forms. The Eye6L is the entry level chip to power forward facing camera systems that are now on most new vehicles. It’s a low-power chip designed single package, behind the windshield applications where minimal heat generation is required. This can enable automatic emergency braking, lane keeping assist and hands-on lane centering solutions like Nissan’s first-generation ProPilot Assist. 

The EyeQ6H is for more full-featured hands-free systems with multiple surround-view cameras. It’s performance matches the EyeQ5 SoC that is launching in production programs this year. 

The EyeQ Ultra takes Mobileye performance to a whole new level. It features 12 CPU cores with 24 threads based on RISC-V archictecture. There are also GPU, vision processing and image signal processing cores and 16 convolutional neural network accelerator clusters. All in, Mobileye claims the Ultra is capable of 176 trillion operations per second (TOPS). By comparison the Nvidia Orin can do 250 TOPS and most Orin applications are using either two or four SoCs. 

However, the run performance numbers don’t tell the whole story since the software that runs on these chips is very different. Mobileye has in the past claimed that its proprietary perception software runs more efficiently and doesn’t require the same raw computational power. The target with the EyeQ Ultra SoC is to enable a single chip solution that can provide full level 4 automated driving capabilities for consumer level vehicles. 

For comparison, the L4 robotaxi platform that Mobileye plans to begin pilot deployments of in Munich, Germany and Tel Aviv, Israel later this year uses a compute platform with 8 EyeQ5H SoCs. The Ultra is expected to match or exceed this performance at lower cost and power consumption. The first sample chips for Ultra are expected by the end of 2023 with full production slated for 2025.