Eyeq4 Datasheet

: Optimized for computer vision tasks, with support for a wide range of algorithms used in ADAS and autonomous driving, such as detection, tracking, and scene understanding.

2 cores offering higher efficiency than standard CPUs and more versatility than a GPU. Programmable Macro Array

Designed from the ground up to meet ASIL (Automotive Safety Integrity Level) standards. Hardware failure detection mechanisms, such as error-correcting code (ECC) on all internal memories and registers, catch potential data corruption.

Two Programmable Macro Array cores, providing high compute density for fixed-function hardware acceleration. eyeq4 datasheet

The Mobileye EyeQ4 is a 28nm FD-SOI, high-performance System-on-Chip (SoC) designed for camera-based Advanced Driver Assistance Systems (ADAS), delivering over 2.5 teraflops of processing power at 3W. Featuring six VMP cores, two MPC cores, and two PMA cores, it supports up to 8 simultaneous cameras for advanced computer vision and autonomous emergency braking. For more details, visit Mobileye .

The is a high-performance vision processor designed specifically for Advanced Driver Assistance Systems (ADAS) and semi-autonomous driving. Launched in 2018, it represented a significant leap in computational efficiency, providing approximately 10 times the processing power of its predecessor, the EyeQ3, while maintaining a very low power envelope. Core Technical Specifications

The EyeQ4 achieves its processing efficiency through a . Rather than relying solely on general-purpose CPU cores or power-hungry GPUs, Mobileye utilizes proprietary, specialized accelerators tailored for computer vision. : Optimized for computer vision tasks, with support

| Feature | Specification | | :--- | :--- | | | 28nm FD-SOI (STMicroelectronics) | | Total Computing Cores | 14 cores | | CPU Cores | 5 MIPS-based cores (4x interAptiv + 1x M-class ) | | Specialized Accelerators | - 6x Vector Microcode Processors (VMP) - 2x Multithreaded Processing Clusters (MPC) - 2x Programmable Macro Arrays (PMA) | | Compute Performance | Up to 2.5 Teraflops (TFLOPS) / TOPS | | Concurrent Camera Input | Processes data from up to 8 cameras simultaneously | | Memory Interfaces | 2x LPDDR4 SDRAM interfaces (32-bit each, up to 1.6GHz) | | Storage Interfaces | 2x 128MB Flash (for redundant code memory) | | I/O and Connectivity | - 4x MIPI CSI-2 camera input ports - 3x CAN ports - 3x UART ports - 3x I2C interfaces - 4x SPI interfaces - 1Gb Ethernet port |

The EyeQ4 was engineered to support a "safety cocoon" around the vehicle, enabling features essential for semi-autonomous and autonomous driving: The Evolution of EyeQ - Mobileye

References: Mobileye EyeQ4 Product Brief (Public, Rev 1.3), Intel Automotive SoC Overview, ISO 26262 ASIL Decomposition Guide for EyeQ4, AEC-Q100 Grade 2 Thermal Validation Report (public summary). Featuring six VMP cores, two MPC cores, and

The CPU complex executes Mobileye's RSS model—a mathematical framework that defines what it means for an automated vehicle to drive safely. It evaluates surrounding traffic behaviors to prevent the host vehicle from initiating dangerous maneuvers. Deep Learning Object Classification

According to the datasheet, the EyeQ4 features a unified memory architecture with:

Deep Learning Accelerator. Dedicated high-performance AI engine. The main source of horse power for convolutional neural networks. ZF and Mobileye Safety Technology Chosen by Toyota