Cutting-edge AI technology represented by large models is rapidly moving forwardNew energy vehiclesPenetration, with strong power to promote the acceleration of intelligence in the automobile industry. Automobiles are being redefined, from “transportation” to “mobile agents” with perception, decision-making and evolutionary capabilities, and AI-driven has become the underlying competitiveness of future automobile companies. At the heart of this change lies the explosive growth of data and the comprehensive breakthrough of AI models in real-time, energy efficiency and security. Traditional computing power-centric architectures encounter “storage wall” bottlenecks – data handling has become a performance shackle, and computing power is difficult to release efficiently. AI cars and AI storage are forming a closely coupled symbiotic relationship, and efficient storage architecture and data processing capabilities (AI storage) have become the key to unlocking the potential of AI cars.
Deyi Microelectronics Co., Ltd. (hereinafter referred to as “Deyi Micro”) has a keen insight into this trend and clearly proposed the concept of “AI storage chip” for the first time in the industry. This is not only an upgrade of storage capacity, but also a reconstruction of the computing paradigm from the architectural level, pointing directly to the core challenges of smart cars in real-time data processing and energy efficiency optimization. Adhering to the vision of “creating more intelligence per bit of data”, Deyiwei has built three core technology pillars: storage control (high-speed, stable, and low-latency data scheduling), storage-computing interconnection (data channels that efficiently connect computing power units and storage units), and storage-computing integration (reducing data handling and improving processing energy efficiency), providing full-stack and full-scenario AI storage solutions covering edge perception, central decision-making, and cloud evolution for smart cars.
Image source: Deyiwei
AI storage chip: Break the “storage wall” and release real intelligence
In the context of the accelerated development of intelligence in the automotive industry, the demand for computing power and storage power for various components of smart cars continues to grow. At present, the computing power (TOPS) indicators of chips equipped with various smart car companies continue to rise, such as the NVIDIA Orin-X chip equipped with Zhiji L7 has reached 254 TOPS.Gasso Automotive Research InstituteAccording to the data, in 2024, the penetration rate of assisted driving of L2 and above in China’s passenger cars will be close to 48%, and the standard configuration of NOA functional models will exceed 7%. It is expected that by 2030, the penetration rate of the domestic L2 and above smart car market will exceed 90%, of which the number of NOA functions will exceed 24 million. The huge market prospects call for the synchronous upgrading of the underlying technology.
However, the increase in hashrate numbers has not fully translated into a linear improvement in user experience. Smart cars have entered the “data-driven” era, but traditional computing architectures are limited by the “von Neumann bottleneck”, and the powerful computing power of CPU/GPU often needs to wait for data to be slowly transported from the storage unit (such as DRAM and SSD), and the problem of “storage wall” is becoming increasingly prominent, mainly reflected in four dimensions of performance bottlenecks:
■High latency: Data handling delays can lead to delays in decision-making, especially in scenarios such as emergency obstacle avoidance and urban NOA that require millisecond responses.
■High power consumption: Frequent data handling consumes a large amount of system energy consumption (up to more than 60% of the total system power consumption), which seriously restricts the cruising range and equipment cooling design.
■Bandwidth bottleneck: The massive raw data torrent generated by sensors, especially lidar and high-resolution cameras, far exceeds the throughput capacity of traditional storage interfaces, forming a data “dammed lake”.
■Low computing power utilization: Powerful computing power units are often in a “starvation” state (Idle), waiting for data input and cannot be used efficiently. This is the core of the dilemma of “computing power growth ≠ experience improvement” – the storage, handling and access efficiency of data has become a key shortcoming restricting the performance release of smart cars.
The innovation path of “AI storage chip” proposed by Deyiwei directly hits this pain point, and reconstructs the data flow through the three core technology matrices of storage control, storage and computing interconnection, and storage and computing integration.
■Optimize storage scheduling: Optimize data access policies in storage media through advanced storage control technology, maximize storage bandwidth utilization, and reduce access latency.
■Open up the high-speed channel: Uses storage and computing interconnection technology to build more efficient (higher bandwidth/lower latency) and smarter (QoS/deterministic latency) connection channels, so that data can reach the computing unit faster and smoother.
■ Disruptive architecture: Explore storage and computing integration technology, where part of the calculation is directly placed inside the storage unit or performed near-end, fundamentally reducing or even eliminating the need for data handling, and achieving exponential energy efficiency and latency reduction. This is also the key to making huge computing power “full” and “work efficiently”.
Image source: Deyiwei
AI storage chip: Deyiwei uses technological innovation to reshape the storage pattern of smart cars
Deyiwei, which has been established for 18 years, has achieved rapid breakthroughs in the smart car track with continuous R&D and innovation. Through underlying technological innovation, the company has independently built a two-wheel drive technical system of chip architecture design (controller SoC) + core algorithms (FTL flash memory conversion layer, ECC error correction, wear equalization, AI acceleration, etc.), and broke through the limitations of “storage wall” with a hard core.
At the level of technological innovation, Deyiwei’s innovative solution focuses on the in-depth mining of data value, and greatly reduces the amount and frequency of data uploaded to the cloud through efficient preprocessing, intelligent caching and near-memory computing on the vehicle side, which not only improves the real-time performance of local processing, but also effectively reduces the network load and data transmission costs, providing a solid foundation for the implementation of “AI-defined cars”.
In terms of product reliability, Deyiwei’s automotive-grade storage chip products, from chip design to manufacturing, strictly follow AEC-Q100 and other international and domestic automotive standards, all products have undergone a series of harsh environmental and reliability tests such as high-temperature aging, temperature cycling, pre-burning, durability and data retention to ensure stable operation in complex and changeable vehicle environments. Based on the fully self-developed chip design and independent firmware, Deyiwei deeply optimizes the performance, reliability (data retention/lifetime), energy efficiency and environmental adaptability (wide temperature) of the product, and has successfully created a number of eMMC and BGA SSD storage products that meet the stringent AEC-Q100 Grade 2/Grade 3 automotive grade certification.
Image source: Deyiwei
Facing the higher future needs of smart cars (such as L3+ autonomous driving and cabin-driving integration central computing), Deyiwei will soon launch automotive-grade UFS storage products, providing read and write speeds of more than 2GB/s, far exceeding the bandwidth and lower latency of eMMC, and meeting the needs of ADAS domain controllers and high-computing power intelligent cockpits for high-speed data caching. In addition, the company is actively deploying cutting-edge CXL automotive-grade storage solutions, exploring the use of CXL high-speed interconnection protocols, building memory pooling and expansion architecture technologies, and providing ultra-large-capacity, ultra-low-latency, and on-demand scalable storage resource pools for future central computing platforms, which will provide key support for the deployment of 100 billion-parameter large models on the vehicle side.
With the development of smart cars to higher-end autonomous driving, Deyiwei will continue to deepen technological innovation and provide the industry with more advanced and reliable vehicle specification storage solutions.
AI storage chip: drive the evolution of intelligent vehicles and build a domestic storage ecology
Relying on the three technical pillars of storage control (stability), storage and computing integration (intelligence), and storage-computing interconnection (communication), Deyiwei’s AI storage chip is actively building an open and cooperative industrial ecology while improving the level of automotive intelligence, and is becoming an important domestic force to promote the evolution of intelligent vehicles.
At present, Deyiwei’s automotive specification products have been widely used in key fields such as automotive digital instruments, T-Box, ADAS, intelligent cockpits, in-vehicle infotainment systems, event recorders, etc., and its cooperative customers cover dozens of mainstream car companies and Tier 1 manufacturers such as Dongfeng, Changan, Geely, SAIC, Shaanxi Automobile, FAW, and Great Wall.
At the level of industrial cooperation, Deyiwei actively promotes the common development of the industry. The company is deeply involved in the discussion and formulation of next-generation in-vehicle storage standards, and has established an innovative cooperation model of “co-definition” with automakers and Tier 1 manufacturers. By proactively grasping the requirements of future models for core indicators such as storage performance (bandwidth/IOPS/latency), capacity, reliability, power consumption, and security, joint development and early verification are carried out to ensure that solutions accurately match market demand and accelerate the application of innovative technologies in mass-produced vehicles.
At the level of supply chain security, Deyiwei has built a fully integrated, independent and controllable chip supply chain, from chip design to firmware development, from packaging and testing to module manufacturing, all key links are completed in China. At the same time, the company has also established stable cooperative relations with leading NAND Flash manufacturers at home and abroad to ensure the supply of high-quality media, providing double guarantee for product quality and supply safety.
AI cars are like a “data black hole” that never gets tired, and storage is the “memory center” and “nerve conduction pathway” of its AI intelligent system. Deyiwei’s innovative AI storage chip reconstructs the data storage and processing paradigm, providing real-time performance of intelligent car perception decision-making (millisecond-level response to ensure safety) through high bandwidth and low latency, and supporting the autonomous learning and continuous evolution of smart cars with large capacity and high performance, so that every bit of data can be used faster, more efficiently and more intelligently, and truly practice its vision of “making each bit of data create more intelligence”.
In the future, with the continuous evolution and breakthrough of technologies such as storage and computing integration and CXL interconnection, the efficiency of AI storage chips will continue to improve, and its role is not only as a container for carrying data, but also as the core engine driving the intelligent leap. The performance and innovation level of AI storage chips will become one of the key yardsticks and core competitiveness to measure the intelligence of the next generation of AI vehicles. Deyiwei will continue to deepen this field and provide solid underlying support for the evolution of smart cars in China and even the world.
