Eyeq4 | Datasheet

The Mobileye EyeQ4 is a high-performance vision-based System-on-Chip (SoC) designed specifically for Advanced Driver Assistance Systems (ADAS) and autonomous driving. Leveraging a heterogeneous multi-core architecture, it delivers significant leaps in computational efficiency compared to its predecessors.

1. Core voltage (0.9V) must rise before I/O voltage (3.3V).
2. Inrush current limit per pin: < 50mA.
3. Power-down sequence: Reverse order of power-up.

The EyeQ4 architecture is built on a 28nm FD-SOI (Fully Depleted Silicon On Insulator) process technology from STMicroelectronics, which is critical for its high efficiency. Specification Details Performance Over 2.5 Teraflops (TFLOPS) or 2.5 TOPS Power Consumption Approximately 3 Watts for typical automotive use Video Processing eyeq4 datasheet

. This "asymmetric" design allows the chip to perform massive parallel processing—essentially "seeing" and "interpreting" multiple data streams from cameras and sensors simultaneously—while maintaining a remarkably low power profile of approximately 3 to 5 watts. Safety and Redundancy Core voltage (0

Thermal Management

Despite a low TDP (3-5W), the datasheet warns about localized hotspots near the VMP accelerators. A thermal pad and a small copper spreader are mandatory for ambient temperatures above 85°C. The EyeQ4 architecture is built on a 28nm

2. Dedicated Hardware Accelerators

Unlike a general-purpose GPU, the EyeQ4 features fixed-function accelerators that offer massive power savings:

Key Features Overview (From the EyeQ4 Datasheet)

The EyeQ4 is purpose-built for Level 2+ and Level 3 autonomy. Unlike general-purpose SoCs, it is a heterogeneous multi-processor unit with hardware accelerators dedicated to computer vision.

The EyeQ4 architecture utilizes a heterogeneous mix of specialized accelerators to achieve high efficiency. Specification Performance 2.5 TOPS (High variant) / ~1.1 TOPS (Mid variant) Power Consumption ~3 Watts (Automotive grade) CPU Cores 4 multi-threaded MIPS InterAptiv cores (4 threads each) Vision Accelerators