AI Accelerators Drive Cost Reduction and Efficiency in ADAS
Intense Competition in Advanced Driving Assistance Systems, Hailo Proposes a New Solution.Author | JansonEditor | ZhihàoAs the new car manufacturing movement enters the second half characterized by intelligence, Advanced Driver Assistance Systems (ADAS) have become the primary focus for OEMs and Tier 1 suppliers.At the same time, with this new wave of car manufacturing, the technical application of ADAS has become a core competitive area for automakers and Tier 1 suppliers.As this industry transformation deepens, many semiconductor companies are emerging in the automotive sector with their specialized technologies, driving cars toward a comprehensive mobile computing platform.With the combined push of electrification, intelligence, and the Internet of Vehicles, the pace of technological innovation in the automotive industry is accelerating. However, when discussing this digital transformation, it is essential to mention the critical area of vehicle system safety.As complex functions such as assisted driving, intelligent interaction, and multimedia entertainment are integrated, the computational performance required for automotive electronics has increased significantly, along with higher safety performance standards for systems.Currently, most mainstream end-to-end autonomous driving solutions on the market are built on Transformer + BEV models, necessitating the industry to provide more specialized and high-performance processor technology solutions to meet the demands of these advanced applications.In this context, Hailo, a hardware company focused on artificial intelligence acceleration, has conducted in-depth optimization of end-to-end computing processing and launched chip solutions specifically designed for intelligent driving, aiming to meet the industry’s stringent requirements for high-performance computing and further promote the development of intelligent driving technology.01.Accelerating Intelligence in the Automotive Industry: The AI Market Faces ChangesIn the global automotive industry’s process of intelligence, the AI market is facing new challenges and opportunities. At this juncture, Hailo provides a unique perspective on the strategic positioning and future development trends of the Chinese automotive market.Hailo believes that the Chinese automotive market is of extremely high strategic importance.In fact, Chinese automakers have already become global leaders in advanced driver assistance systems (ADAS) and autonomous driving technology. The rapid development of the Chinese market aligns with Hailo’s dynamic and rapidly evolving product portfolio and capabilities.Notably, the market believes that this leading position will also continue in the field of autonomous driving. Chinese manufacturers, with their innovative and agile characteristics, quickly bring innovative solutions to market, achieving rapid growth and occupying significant shares in both local and global markets.In addition, in the AI layout, Hailo believes that safety and regulations are the main growth drivers for autonomous driving. Currently, automakers are required to include various safety applications. These applications require AI processing based on sensor perception, paving the way for easily adding more advanced features on existing hardware and sensors.On the other hand, as drivers spend more time on the road and in traffic jams, the demand for driver assistance and automated driving/navigation (NOA) functions is becoming increasingly common.To this end, Hailo proposes a dedicated accelerator solution for popularizing ADAS, distinguishing it from so-called integrated solutions. The Hailo-8 AI accelerator solution can provide greater flexibility and cost advantages for ADAS solutions.
02.End-to-End Algorithm Acceleration: What Kind of Processor Does the Industry Need?In the rapidly evolving technology of smart vehicles, global automakers are facing a new trend of intelligent upgrades for ADAS systems. Since last year, ADAS technologies represented by Bird’s Eye View (BEV) have gained widespread attention and adoption in the industry due to their high-precision environmental perception capabilities. However, the implementation of BEV technology demands extremely high computational power, which also represents significant cost and technical challenges.Traditional ADAS systems are mostly based on Convolutional Neural Networks (CNN), while transformer BEV technology requires stronger AI processing capabilities.In simple terms, to achieve BEV, the vehicle’s Electronic Control Unit (ECU) must be equipped with sufficient computational power. Additionally, a typical BEV solution usually requires six cameras, and high-end systems may exceed ten cameras. The vast amount of image data provided by these cameras needs to be processed through AI neural network models, which requires extremely high computational demands.Vendors capable of implementing BEV solutions on the market typically use platforms that are quite expensive. The high cost is partly due to expensive training costs and the high computational power demands of the chips themselves, which naturally leads to higher costs; on the other hand, significant computations come with high power consumption, which brings about the demand for high-performance cooling systems, sometimes even requiring active cooling systems.These factors combined mean that BEV’s ADAS solutions can typically only be applied to high-end models. However, the market clearly expects this technology to be popularized in mass-market models. To achieve this, a balance must be struck in terms of cost.To this end, Hailo’s strategy is to combine automotive SoCs with one or more Hailo AI processors to provide high flexibility and scalability of AI computing for ADAS/AD ECUs.This flexibility allows for customized adjustments based on the number of sensors and the complexity of AI models, supporting designs from simple to highly complex systems.The design of Hailo processors considers power consumption and cost efficiency, enabling them to process advanced AI models, such as 3D object detection for ADAS/AD perception and intelligent in-car experiences (e.g., large language models LLM). This feature makes Hailo’s technology suitable not only for high-end models but also for providing advanced intelligent driving functions for mass-market vehicles.In this regard, the Hailo-8 AI accelerator demonstrates its advantages. Compared to other high-end and expensive SoCs on the market, the Hailo-8 offers a cost-effective solution. It is specifically designed to support AI computing, taking on important computational tasks throughout the BEV system. By combining the Hailo-8 AI accelerator with conventional automotive SoCs like Renesas, a balance of cost and performance can be achieved. The front-end SoC handles image and sound inputs, while the responsibility for AI computations is handed over to the Hailo-8 AI accelerator.Moreover, the power consumption performance of the Hailo-8 AI accelerator is particularly outstanding, achieving a computational capability of 26 TOPS with only 2.5 watts of power consumption. This combination not only lowers overall costs but also maintains the efficient operation of the system.Although integrated solutions are currently popular, the superiority of specialized chips cannot be denied.Physical laws tell us that general-purpose chips have strong performance but weaker specialization; specialized chips can optimize algorithms specifically, sacrificing some degree of generality, but achieving specialized functions with good energy consumption performance, helping enterprises reduce costs and increase efficiency, thereby promoting the development of ADAS.03.Cost Reduction and Efficiency Become Mainstream: The Advantages of AI Accelerator Solutions are ObviousFrom the four core advantages of the Hailo-8 AI accelerator, it is not difficult to see that in the current situation where car manufacturers are fiercely competing in a “price war” to reduce costs and increase efficiency, the Hailo AI accelerator solution has significant advantages.From the functional execution perspective, compared to integrated ADAS solutions, the Hailo-8 AI accelerator has a relatively obvious advantage in the field of autonomous driving, as its design fundamentally differs from the operation methods of traditional CPUs and GPUs when executing neural network models.Traditional processors, when processing neural networks, are not specifically designed for such tasks, resulting in additional overhead during execution, particularly in data transfer, especially the data exchange between the processor and memory.The Hailo-8 AI accelerator adopts a distributed architecture, divided into multiple blocks, each containing computing, memory, and control resources. This design allows each layer in the neural network model to directly utilize these resources, reducing data transfer overhead between the processor and memory, thereby significantly lowering costs.During the execution of the Hailo-8 AI accelerator, the entire neural network model is analyzed and resources are allocated during initialization, allowing image data obtained from the camera to be continuously processed through all layers within the chip until results are obtained. This processing method is not only fast but also energy-efficient.According to reports, the architecture of the Hailo-8 AI accelerator is known as data flow architecture, which is its proprietary patented technology. This enables the Hailo-8 AI accelerator to quickly process cutting-edge neural networks in an energy-efficient manner, bringing significant specialization and efficiency to autonomous driving.Compared to traditional CPUs and GPUs, the Hailo-8 AI accelerator is faster and more efficient in executing neural network models, allowing it to refine each computation unit of the neural network model, thus performing more precise prediction operations.This precise predictive capability means that whether processing traditional CNN models or emerging large models, the Hailo-8 can provide excellent performance and power consumption control. This is crucial for the real-time performance and energy efficiency of autonomous driving systems, as these systems need to quickly and accurately process vast amounts of sensor data and respond immediately.Therefore, the Hailo-8 AI accelerator not only enhances the performance of autonomous driving systems but also brings cost-effectiveness and sustainability improvements to the entire industry.At the same time, Hailo has made good progress in new product collaborations.Recently, Hailo reached a collaboration with Zhixing Technology, where Zhixing Technology selected the Hailo-8 AI accelerator, paired with Renesas R-Car V4H SoC, to jointly drive its iDC High domain controller.
Integrating the Hailo-8 AI processor, the iDC High domain controller provides vehicles with smarter driving assistance functions, such as highway cruising/automated assisted navigation (NoA), automated memory parking, and potentially future city cruising/NoA applications. The cost-effectiveness of these applications will be significantly enhanced, making it possible to popularize them in mass-market models. A Chinese automaker is set to begin mass production of vehicles equipped with the iDC High domain controller in the second half of this year.The high performance of the iDC High enables it to provide BEV 3D perception for autonomous driving applications, enhancing safety and comfort through a 10V5R sensor configuration. The low power characteristics of the Hailo-8 accelerator allow the domain controller to adopt passive cooling technology, which not only reduces the material costs of the entire vehicle but also simplifies the vehicle integration process.Hailo CEO Orr Danon expressed his views on this collaboration, believing it is an important step in popularizing advanced autonomous driving applications in a safer and more economical way across all vehicles. The partnership with Zhixing Technology marks another milestone for Hailo in applying its AI technology to the foundational aspects of the global automotive industry. The global automotive industry is seeking AI solutions capable of achieving autonomous driving and parking, which need to possess high performance, economic efficiency, stability, and scalability.Zhixing Technology’s CTO Lu Yukun also expressed positive views on the collaboration, emphasizing the advanced AI capabilities and efficiency brought by the Hailo-8 AI accelerator to the market, which will benefit all drivers. The unique advantage of the Hailo-8 AI accelerator lies in its ability to provide efficient AI acceleration for cutting-edge neural networks while maintaining low energy consumption, which is significant for promoting automotive innovation.Takashi Fushi of Renesas Electronics also emphasized the unprecedented functionality and price competitiveness brought by the integration of Renesas R-Car SoC and Hailo-8 AI accelerator.04.Conclusion: AI Accelerators Become a New Choice for ADASWith the growing demand in the automotive market and the acceleration of intelligence processes, the application of ADAS is not only required for high-end models.Currently, the mainstream sinking market, including mid-range and low-end markets, also has a strong demand for ADAS.Hailo’s innovative process and market dynamics demonstrate its ongoing efforts in promoting automotive electrification and intelligence, further pushing ADAS technology into lower-tier markets through specialized optimization and low-energy products.We can look forward to its continued influence in the future automotive technology field.
