Features four CPU cores capable of executing four hardware threads each (totaling 16 threads). These cores handle upper-level system tasks, application scheduling, interface buses, and general driving policy tracking. Vision and Deep Learning Accelerators
For high-level logic, path planning, sensor fusion, and operating system management, the EyeQ4 features . Each core supports hardware multi-threading, enabling the SoC to manage dozens of concurrent software threads smoothly without context-switching bottlenecks. 3. Technical Specifications & Datasheet Metrics
The foundational magic of the EyeQ4 datasheet lies in its highly heterogeneous, many-core computing architecture. Rather than relying on a power-hungry general-purpose GPU, Mobileye engineered a proprietary mix of general-purpose RISC processor cores and highly specialized hardware accelerators. 1. General-Purpose Compute Layer Mobileye EyeQ4 Vision Processor Family - Yole Group
: Features 4 CPU cores with 4 hardware threads each, integrated with Mobileye's proprietary Vector Microcode Processors (VMP). Performance
The capabilities of the EyeQ4 have made it a popular choice for automakers looking to enhance both safety and driving comfort. It serves as the central vision processor for a wide range of ADAS functions, as demonstrated by its implementation in vehicles like the NIO ES8, the Ford Bronco and F-150, and the third-generation Haval H6. Key applications include: eyeq4 datasheet
: Extremely efficient, operating within a budget of approximately 3 Watts. Core Configuration : 4 multi-threaded MIPS InterAptiv CPU cores.
Employs two MPC cores . More versatile than a traditional GPU, MPCs are explicitly tuned to support custom perception algorithms and graphical neural networks with superior execution speed than typical open-compute hardware layers. 4. Key Supported Functional Features
While the raw TOPS number (2.5) seems low compared to modern desktop GPUs, the datasheet emphasizes . The EyeQ4 executes 8-bit integer CNN inferences at a rate of 0.25ms per layer. This allows it to detect objects (cars, pedestrians, traffic signs) at 30-36 frames per second across three distinct camera streams simultaneously.
The is a cornerstone Vision System-on-Chip (SoC) designed by Mobileye and manufactured by STMicroelectronics . It serves as a fundamental building block for Level 2 and Level 2+ Advanced Driver Assistance Systems (ADAS). Operating with a performance of 2.5 Tera Operations Per Second (TOPS) while drawing a mere 3 Watts of power , the EyeQ4 represented a massive 10x compute leap over its predecessor, the EyeQ3, with only a 20% increase in power footprint. Features four CPU cores capable of executing four
4x multi-threaded 64-bit RISC MIPS CPUs (4 hardware threads each)
: Peripheral Component Interconnect Express for high-speed chip-to-chip communication in multi-SoC architectures. 5. Embedded Software & Computer Vision Capabilities
For academic or evaluation purposes, Mobileye offers a (45 pages) covering mechanical, power, and thermal specs without register-level details.
In summary, the Mobileye EyeQ4 system-on-chip is a powerful, purpose-built vision processor that strikes an exceptional balance between high-performance, low power consumption, and automotive-grade reliability. Its advanced heterogeneous architecture enabled a significant leap in ADAS capabilities, bridging the gap from basic Level 2 assistance to the dawn of Level 3 conditional automation. As a cornerstone of automotive safety technology, the EyeQ4 continues to represent a key milestone in the journey towards fully self-driving vehicles. Rather than relying on a power-hungry general-purpose GPU,
The combination of general-purpose CPUs and purpose-built accelerators allows the EyeQ4 to achieve a , which represents a ten-fold increase over its predecessor, the EyeQ3 . This massive leap in capability is what enabled new, more advanced ADAS features.
According to the datasheet, the EyeQ4 features a unified memory architecture with:
Quad-core processors with multi-threading (up to 4 threads per core). VMP Vector Microcode Processor