The watershed of high-level intelligent driving, the era of the great chip is coming: Black Sesame Intelligence releases a new weapon.

As the autonomous driving technology is continuously approaching a qualitative change node, this year is undoubtedly a watershed for the landing and explosion of autonomous driving, and the upstream and downstream of the industrial chain have pressed the accelerator button.

On December 30, BST AI announced the launch of its high-computing power chip platform specially designed for the next-generation AI model - the Huashan A2000 family.

The Huashan product line of BST AI has always carried an important mission, and the launch of the Huashan A2000 family is a powerful inheritance of this mission.

The Huashan series has always been committed to providing strong technical support for the intelligentization of automobiles, promoting the transformation of automobiles from traditional transportation tools to intelligent mobile terminals. The advent of the Huashan A2000 family is undoubtedly a more solid step on this path.

The Huashan A2000 family has higher computing power and can easily handle complex AI computing tasks. In the intelligent driving scenario, whether it is the real-time analysis of road conditions, the accurate identification of surrounding vehicles and pedestrians, or the rapid decision-making for various unexpected situations, the support of strong computing power is indispensable. With its high computing power advantage, the Huashan A2000 family is like installing a super intelligent brain in the car, enabling it to perceive and understand the surrounding environment more acutely and accurately.

BST AI launched the Huashan A2000 family with a clear and ambitious goal. It aims to empower the automotive industry with this high-computing power chip platform and accelerate the high-level intelligent driving to become a standard configuration.

Today, although intelligent driving has made certain progress, high-level intelligent driving has not yet been widely popularized. The emergence of the Huashan A2000 family is expected to break this situation, enabling more cars to have the ability of high-level intelligent driving and improving the intelligence level of the entire industry.

A New Milestone of High-Level Intelligent Driving

In recent years, high-level intelligent driving has shown a gradually popularizing trend in the automotive industry, which is profoundly changing the pattern of the automotive industry. 

And entering 2024, the explosion of Baidu's intelligent travel platform Luo Bu Kuai Pao and Tesla's entry into Robotaxi, etc., have implanted the concept that "autonomous driving has become a reality" into the minds of more ordinary consumers. 

From the technical side, autonomous driving has entered a new mature stage at all levels. According to the report, since the second half of 2023, high-level intelligent driving technology has entered a new stage of development, and urban NOA has become the main target of the industry. Leading intelligent driving manufacturers are actively launching end-to-end large models to compete for the optimal solution of urban NOA. 

Many subdivided technologies have made many significant progress. For example, sensor technology is constantly innovating, and devices such as cameras and radars (including lidar, millimeter-wave radar, etc.) are increasingly sophisticated, which can capture the environmental information around the vehicle more accurately and comprehensively. 

At the same time, advanced algorithms are also continuously optimized. Since the popularity of large models, technologies such as deep learning and machine learning have been widely applied in the field of intelligent driving, which has greatly improved the vehicle's ability to analyze, process and make decisions on the obtained data.

For example, through a large amount of data training, the intelligent driving system has been able to realize functions such as automatic following, automatic lane changing, and automatic parking, and even in some specific scenarios, it can also deal with unexpected situations and perform emergency braking or avoidance operations, which greatly improves the convenience and safety of driving.

However, more technical difficulties remain to be overcome.

At the sensor level, no matter which sensor is currently used, there are certain technical shortcomings. For example, the current lidar technology can provide accurate three-dimensional point clouds, but its performance will decline in rainy and snowy weather conditions; the camera can accurately capture road condition information, but its perception effect will be affected by changes in light, etc.

In the research and development of AI chips, improving the computing power, power consumption, and performance of the chips is the current technical problem for enterprises. In the autonomous driving scenario, the complex environment and real-time nature place extremely high requirements on the computing power and performance of the chips. For example, the decision-making of autonomous driving must be completed in an instant, and any delay may cause serious consequences.

When a pedestrian or vehicle suddenly enters the road, the chip must calculate the best avoidance or braking plan based on the current road conditions and vehicle status in a very short time (usually calculated in milliseconds or even microseconds). However, as the level of autonomous driving increases, the fineness of environmental perception and the complexity of decision-making also increase significantly, and maintaining this real-time performance is a huge challenge for the chip performance.

In addition, the vehicle has a long service life and may drive in various extreme conditions, which requires the autonomous driving chip to be stable and reliable under long-term continuous working conditions.

No matter in the high-temperature desert environment, the cold polar region, or the high-humidity coastal area, the chip cannot fail or have a performance decline. For example, during a long-distance self-driving tour, if the chip malfunctions due to long-term operation, the autonomous driving function will fail, putting passengers in danger.

High-Level Intelligent Driving towards Full-Scenario General Intelligent Driving

Based on the high requirements of autonomous driving for chip performance, the birth of the BST AI Huashan A2000 family is even more significant.

BST AI observes that the future development of algorithms will focus more on improving efficiency and performance. As the industry enters the era of large models, the Transformer algorithm structure and the hybrid model architecture will lead the new technological trend. BST AI predicts that from 2025, the high-level intelligent driving capability will gradually become a standard configuration. 

In the intelligent driving scenario, the vehicle needs to process massive data from multiple sensors in real time, including image information captured by the camera and distance and speed data detected by the radar, etc. 

For example, in a complex urban traffic environment, facing numerous vehicles, pedestrians, and various traffic signs, this chip can quickly analyze the relevant information and provide strong support for driving decisions. At the same time, it also has a lower latency characteristic. In the intelligent driving process, any delay may lead to serious consequences, such as the inability to respond to unexpected situations in a timely manner. 

For this, BST AI has proposed the concept of full-scenario general intelligent driving. By introducing the information of the driving scene into the knowledge-enhanced representation space based on the knowledge paradigm, this information can be deduced as the general knowledge in the scene semantic space, and then the scene is inferred through the reflection of knowledge, thereby guiding the realization of a better intelligent driving experience. 

The general intelligent driving has the general ability to achieve advanced perception, decision-making, and execution, and can fully cover different scenarios of urban roads, highways, day and night changes, and various climate conditions. 

A highlight of the Huashan A2000 family chip is that it adopts a self-developed new NPU architecture - "Jiu Shao".

Jiu Shao is the computing core of the high-performance AI chip launched by BST AI to meet the technical requirements of autonomous driving. Two innovations, the new-generation universal AI toolchain BaRT and the new-generation dual-chiplet interconnection technology Blink, jointly empower the full play and flexible expansion of the computing performance of "Jiu Shao", forming a powerful intelligent driving technology base. 

This innovative architecture has significant advantages in AI model adaptation and power consumption optimization. In terms of AI model adaptation, the "Jiu Shao" architecture has a high degree of flexibility and compatibility. 

With the continuous development of intelligent driving technology, AI models are also continuously updated and improved. The "Jiu Shao" architecture of the Huashan A2000 family chip can easily adapt to various new AI models, ensuring that the chip is closely combined with the latest intelligent driving algorithms to achieve more accurate environmental perception and decision-making. 

In addition, the Jiu Shao architecture also has a three-layer memory architecture with low latency and high throughput, including a large-capacity and high-bandwidth NPU dedicated cache, an on-chip shared cache for the core module, and a symmetrical dual data path and a dedicated DMA engine. It improves performance and effective bandwidth, reduces the dependence on external memory bandwidth, and achieves the ultimate balance between performance, bandwidth, and cost. 

In terms of power consumption optimization, the "Jiu Shao" architecture performs even more outstandingly. The intelligent driving chip needs to consume a certain amount of electricity during operation, and excessive power consumption will not only affect the vehicle's cruising range but also may increase the heat dissipation cost. Through advanced design concepts and technical means, the "Jiu Shao" architecture effectively reduces the power consumption of the chip, enabling the Huashan A2000 family chip to ensure high-performance computing while minimizing the energy consumption of the vehicle, making a contribution to energy conservation and environmental protection. 

It can be seen that the standard configuration of high-level intelligent driving puts strict requirements on the chip, especially in complex scene processing and real-time decision-making. The Huashan A2000 family chip precisely meets these needs.

With the release of the Huashan A2000 family chip, the local autonomous driving chip pattern will usher in a major change, and the process of autonomous driving will also move forward further.