Ambarella Introduces New AI SoCs for In-Vehicle Fleet Telematics

Ambarella recently introduced the CV72AX and CV75AX SoCs, designed to meet the sophisticated needs of modern fleet telematics and vehicle gateways. Ambarella leveraged Samsung Foundry's 5 nm FinFET process technology to design these new SoCs for advanced AI performance, energy efficiency, and processing capabilities. 

 

The CV72AX and CV75AX are designed for fleet telematics. 

 

The New AI SoCs: CV72AX and CV75AX

The CV72AX offers up to six times the AI performance of its predecessor, while the CV75AX offers two times the performance. With these advanced capabilities, the SoCs enable advanced vision transformers and multimodal vision language models (VLMs). This enhanced AI capability supports instant event alerts, multi-channel video pre-categorization, and natural-language searches, improving video analysis efficiency.

Thanks to Samsung’s 5 nm FinFET process technology, the chip has high power efficiency, performance capabilities, and effective operation across a broad temperature range of -40°C to 105°C. The SoCs’ CPU performance has also doubled compared to the previous generation, and they now offer support for LPDDR5 memory with twice the DRAM bandwidth.  

 

Block diagram of the CV72AX. 

 

The CV72AX integrates a high-performance image signal processor, supports up to 10 cameras, and includes a H.264/H.265 video encoder. Ambarella’s advanced Vivid HDR technology utilizes AI to enhance image quality in scenes with high contrast between light and dark areas.

The CV75AX includes an ultra-low-power sleep mode for its on-chip microcontroller (MCU), reducing power consumption to less than 5 mW. This feature eliminates the need for an external MCU to monitor vehicle sensors while parked, thereby minimizing design complexity and reducing the bill of materials (BOM) cost.

 

Featuring Ambarella's Latest CVflow 3.0 AI Engine

Ambarella baked the latest version of its CVflow AI engine into the two new chips. The company's CVflow chip architecture taps into heart of computer vision algorithms, including a dedicated vision processing engine programmed with a high-level algorithm description. This critically sets it apart from general-purpose GPUs or CPUs and enables CVflow chips to scale to trillions of operations per second while consuming very little power. 

 

Overview of Ambarella's CV flow and SW framework 

 

CVflow allows customers to map their custom CNN networked trained with popular tools, including PyTorch, TensorFlow, ONNX, and Caffe, to run on Ambarella's processors. 

 

Toward Connected Vehicles

The automotive industry's increasing reliance on artificial intelligence (AI) and telematics solutions has created a substantial demand for more powerful and efficient in-vehicle computing systems. AI-driven applications, such as advanced driver assistance systems (ADAS), driver monitoring systems (DMS), and fleet management solutions, require robust computing platforms to deliver real-time analysis, decision-making, and enhanced safety features.

Ambarella hopes that its CV72AX and CV75AX SoCs will mark a leap forward for in-vehicle AI systems. These SoCs are built to transform fleet telematics and vehicle gateways with their enhanced AI engines, improved image processing, and low-power operation. 

 

All images used courtesy of Ambarella