Core follows smart movement and leads the way | Chaoxing Future releases a new generation of intelligent driving computing chip "Jingzhe R1"

On December 27, 2022, Chaoxing Future, a provider of high-energy-efficiency computing solutions for intelligent driving, held an annual product launch and strategy launch with the theme of "Core Follows Intelligent Movement, Leading the Way". Dr. Liang Shuang, founder and CEO of Superstar Future, shared his observations and thoughts on the computing needs of intelligent driving on behalf of the team, announced the new strategy of "Superstar Future 2.0", and launched a series of new products: high-energy-efficiency AI processing architecture "Pinghu", new generation Intelligent driving computing chip "Jingzhe R1" and full-process development tool chain "Luban". This is also the first time that Chaoxing Future has held a formal launch event for the public.

● Inflection point/Timing/Positioning/Easy to use/Changing situation

Grasp the pace of the industry, mass production is king, and make products that meet market demand.

Liang Shuang said that in the past few years, the entire industry has seen a significant change in its understanding of the development path and future expectations of intelligent driving. During this process, Chaoxing Future has also been repeatedly thinking about several issues: What are the long-term and short-term trends of intelligent driving in the future? What? Where is the most technologically advanced and commercially valuable link in the entire industry chain? What is Superstar’s best and most advantageous ability in the future? In the end, the team summarized their observations and thinking into five key words.

The key word is "inflection point". There is a rule in the commercial market: when the penetration rate of an emerging species exceeds 15%, it will quickly change consumer habits and logic, and then rise rapidly. Research data shows that from January to September this year, the L2 and L2+ ADAS assembly rate of China's passenger cars has reached 33.5%. At the same time, it can be seen that Shanghai has introduced a policy that by 2025, the proportion of new cars with L2 and L3 functions will exceed 70%. And there are also predictions that China’s smart car sales will reach 20 million by 2025 . Chaoxing Future firmly believes that pre-installed and mass-produced smart driving will definitely enter the fast lane in the next few years. Data is the most important support, and mass production is king.

Keyword 2 "Timing". After years of exploration, the focus of the intelligent driving market has begun to shift from open scene L4 to front-loaded mass production L2+ systems, converging to a more pragmatic and progressive development route. Chaoxing Future believes that intelligent driving must be combined with the pace of industry development to define products that consumers can “afford and be willing to use”. Supporting high-level autonomous driving through a central large-power computing chip solution is a majestic Mount Everest, and may not be realized until close to 2030; the biggest market opportunity in the next few years will still come from mass production solutions represented by integrated parking and cabin parking. Intelligent driving must eventually return to its business essence, advocating cost reduction and efficiency improvement, iterative evolution, and laying eggs along the way.

Keyword three: "positioning". Looking back at the smart car market in the past few years, we can observe that there have been obvious stratifications in the market positioning, technical needs, and cooperation models of vehicle products. The highest sentiment in the industry is usually in the flagship market, where complete vehicles often cost more than 400,000 yuan, supporting the brand's sense of technology and influence. However, further analysis shows that the largest shipments are still concentrated in the economical market priced between 100,000 and 300,000, accounting for 70% of the overall sales of smart cars. This type of vehicle products tend to pay more attention to sales and need to be launched to the market as soon as possible, so they pay more attention to cost control. Therefore, they require extreme software and hardware optimization at the intelligent level, and the requirements for suppliers' turnkey delivery capabilities will be higher. The two types of markets have very different requirements for technology selection, which also tests the understanding and positioning of upstream suppliers.

Keyword 4: “Easy to use”. As the industry's understanding of intelligent driving mass production systems deepens, everyone gradually realizes that: just because the hardware equipment is made does not mean that the OEM can use it; the low cost of individual hardware does not mean that the overall cost of system migration is low; hardware Even if the computing power is large, it does not mean that the actual performance of the system is strong enough. All this shows the importance of software tools. Chaoxing Future believes that building a complete and easy-to-use software stack requires a comprehensive understanding of the AI ​​software ecosystem and a deep understanding of the requirements of intelligent driving scenarios. It also requires rapid implementation to form a data closed loop and system iteration.

Keyword 5: "Change of situation". The structure of the automobile industry chain is also undergoing major changes. Under the traditional model, Tier1 occupies a dominant position, and there is a linear relationship between "software supplier/chip supplier Tier2 - component integrator Tier1 - OEM". With the advent of the intelligent era, the supply chain has transformed into a network structure centered on the main engine factory, and the cooperation model of all parties has become more open and diverse. Chaoxing Future believes that open cooperation is more needed in this era: OEMs will be more involved in intelligent driving and will have a wider choice of suppliers; Tier1 will begin to integrate full-stack software and hardware capabilities, transforming from integrators to solution providers ; And software/chip suppliers have more opportunities to face the OEMs and directly understand the needs of end customers.

● Super Star Future 2.0 Era

Intelligent Efficient leads the way | Intelligent Efficient

Chaoxing Future was founded in 2019. When the team was first established, it was characterized by the cross-domain genes of "intelligent driving" and "energy-efficient computing", with industry-leading computing architecture design capabilities and algorithm optimization capabilities as its core advantages. Based on the above-mentioned profound insights into the development path of intelligent driving, Chaoxing Future officially proposed a new development strategy "Intelligent Efficient leads the way | Intelligent Efficient" in the 2.0 stage of the company's development.

In the future, Chaoxing will adhere to the route of collaborative optimization of software and hardware, positioning itself in Tier 2 of the smart car industry chain, targeting OEMs, Tier 1, autonomous driving solution providers and other customers, providing smart driving computing chips as the core, supporting AI processing architecture, multi-modal Energy-efficient computing solution for intelligent driving using three-dimensional perception algorithms. The company will focus on front-end mass production scenarios, create "double-wheel iterations" of software and hardware, promote product upgrades through data drive, and realize the product vision of "computing, integration, and connection". Chaoxing will be committed to becoming a leader in energy-efficient computing for intelligent driving in the future, and will work with partners to promote the large-scale implementation of intelligent driving.

Based on the new strategy, Chaoxing Future released three major new products at this event.

High-energy-efficiency AI processing architecture "Pinghu"

The first is "Pinghu", a high-energy-efficiency AI processing architecture customized for intelligent driving scenarios. The Pinghu architecture integrates a systolic array tensor computing engine and a low-power vector processing unit, which can widely support current mainstream neural networks, process various types of dense and non-dense operators in a targeted manner, and achieve efficient operation of the overall model. In response to the bandwidth problem commonly faced by high-performance computing, Pinghu architecture is designed with a two-dimensional tensor cache and a high-performance DMA module, which can support the calculations of the tensor engine and vector engine with full throughput in various situations, and cooperate with the compiler to implement Ultimate on-chip data reuse. Combining the ultimate optimization of "computing" and "memory access", the Pinghu architecture can achieve extremely high performance utilization and computing energy efficiency. Actual measurements show that the utilization rate of typical neural networks for intelligent driving running on Pinghu can be higher than the industry leader on average. Level 20%.

"Pinghu" comes from Chairman Mao's famous "Water Melody Songtou·Swimming". Chaoxing hopes that the results of years of persistent and in-depth research can bring everyone the most efficient and easy-to-use AI processing architecture, and a "natural moat" for intelligent driving to come to fruition. Become a "thorough".

New generation of intelligent driving computing chip "Jingzhe R1"

Based on the Pinghu architecture, Superstar Future Design has launched a new generation of intelligent driving computing chip "Jingzhe R1". Jingzhe R1 can provide 16TOPS@INT8 AI hard computing power and 30KDMIPS general computing power. It can operate in high throughput mode and low latency mode according to different scenarios. It is equipped with a high-performance computing library and is positioned for L2+ level intelligent driving applications and is accurately vector-oriented. product market.

LPDDR4x dual-channel memory provides responsive data movement capabilities and supports efficient real-time computing. According to actual measurement, the core energy efficiency ratio of Jingzhe R1 is 4TOPS/W, reaching the first echelon level in the industry. It can realize natural passive heat dissipation design at the system level and reduce the overall domain control cost. The average utilization rate of a typical network running smart driving on Jingzhe R1 can reach more than 70%, fully unleashing the chip performance and not being a paper tiger.