In recent years, the rise of artificial intelligence, cloud services, big data and other technologies has put forward unprecedented requirements for the network, and the extensive business layer requirements have led to the rapid growth of data centers. As the infrastructure for large-scale computing, massive data storage and provision of Internet services, data centers are developing towards high bandwidth and new transmission systems. The network transmission rate is moving towards 100Gbps, and will rapidly develop to 200Gbps and 400Gbps in the future.
More and more virtual machines and containers, ever-changing application requirements, and increasing attention to network security have put enormous pressure on the CPUs of computing nodes in server infrastructure. In the post-Moore’s Law era, the growth rate of CPU computing power has slowed down, and the gap with the growth rate of the data center network transmission rate has gradually widened. New architectures are needed to enhance the performance of computing power and network transmission. Smart NIC (SmartNIC) requirements.
Compared with traditional network cards, it is only responsible for data link transmission, network stack algorithms and protocols. As a programmable network card, the smart network card can meet the data plane network processing requirements and be compatible with the existing network protocol ecology, thereby offloading the network processing workload and tasks of the CPU, and providing a low-latency, high-bandwidth network in the data center. services, reduce the computational burden of the CPU, and improve the efficiency of the overall solution.
In addition, by using smart network cards to improve the computing power of each computing node, under the same computing power, fewer servers are required to use smart network cards, thereby reducing the initial hardware investment cost of servers, physical space and corresponding operation and maintenance, etc. Ancillary resource consumption is also saved, greatly reducing the total cost of ownership for large-scale deployment of network services.
Smart NIC market size forecast
Driven by the development of various emerging technologies, the market scale of the global smart NIC industry has gradually increased. It is predicted that the smart NIC market will reach the level of tens of billions of dollars in the future, and the smart NIC industry will usher in rapid development.
The Dilemma and Breakthrough of Domestic Smart NICs
Under the current trend, major cloud and server manufacturers are urgently seeking solutions based on smart NICs to reduce their hardware investment and energy consumption costs, and smart NIC solutions are gradually maturing. Current industry-leading manufacturers include Broadcom, Intel, Mellanox (acquired by NVIDIA), Xilinx (acquired by AMD), Marvell, etc.
These are the world’s leading smart NIC manufacturers. In contrast, domestically, China’s smart NIC industry as a whole is far behind the international giants. The reasons for the backwardness are insufficient technology accumulation and upstream EDA and advanced manufacturing processes are controlled by foreign countries. Although the entry barrier for the smart NIC software industry is low, the innovation speed of Chinese local enterprises also lags behind the international giant smart NIC hardware barriers. At the same time, Chinese manufacturers lack commercial experience, and most of their products stay in the laboratory stage, making mass production difficult.
Under this industry pattern and predicament, domestic manufacturers are also striving to make breakthroughs and seek a new situation for the development of the domestic smart network card industry. Among them, Corigine, as a leader in the local smart network card industry, is actively deploying. Coreqiyuan smart network card SmartNIC has been mass-produced since last year. As a leading supplier of intelligent network core chips and system solutions in China, Coreqiyuan can provide domestic, self-controllable core chips and high-speed smart network card solutions. Independent and controllable localization from chips to network cards.
It is understood that the smart network card of Chipqiyuan is currently the only mature DPU complete solution based on SoC architecture in China, which fills the domestic technical gap in this field and has achieved mass production.
Chip Qiyuan smart network card based on programmable SoC (Source: Chip Qiyuan)
The solution adopts the industry-leading NP many-core technology architecture to achieve efficient and flexible network packet processing, and has the characteristics of high performance, low cost, low power consumption, high expansion, and strong programmability.
A list of the advantages of the chip Qiyuan smart network card (source: chip Qiyuan)
In terms of architecture, smart NICs can be designed based on three types of core processors: ASIC, FPGA and MP (Multi-core, including SoC-GP and SoC-NP). Smart NICs based on different core processors have different characteristics.
Smart NICs based on ASIC design have a high threshold and long R&D cycle (requires a development cycle of about two years). The initial R&D cost of medium-complexity ASICs ranges from several hundred to twenty million US dollars. It has the advantage of high cost performance. There are deficiencies in programmability;
FPGA-based smart network cards can provide powerful computing power and sufficient flexibility, and have the characteristics of low latency and high power consumption, but the architecture also has the disadvantages of high complexity and imperfect open source ecology;
The intelligent network card based on MP design not only guarantees certain programmability, but also guarantees good performance.
Suppliers can achieve a single-point breakthrough product model through the technical path of different processor combinations, or seek a balance between different demand points. Based on the industry’s advanced NP-many-core technology architecture, the SoC solution can support hardware offloading including CHKSUM, TSO, OVS, eBPF/XDP, etc., as well as on-chip flexible interfaces and rich hardware modules in the network processor chip. ARM is used for business control and management and other features, and can provide a competitive smart NIC solution, perfectly solve the problems of insufficient ASIC programmability and insufficient GP performance, and seek the best balance between flexibility and performance.
Core Qiyuan DPU internal architecture diagram (Source: Core Qiyuan)
Host virtualization solution based on Chipqiyuan DPU
Coreqiyuan smart network card solution can provide chips, boards, driver software and a full set of cloud network solution products, covering advanced DPU chip technology, smart network card SmartNIC board card design and manufacturing technology and the ecosystem of collaborative cloud network software binding Software technologies, including software-defined network security, virtualization technology and modern clustered storage and load balancing technologies, can provide highly competitive solutions for domestic 5G communications, cloud data centers, network security and other applications.
With the rapid growth of data volume and transmission speed in the era of big data and cloud computing, smart network cards based on DPU chips can accelerate the development and realization of new requirements through flexible programmability, which can greatly improve the performance of cloud data centers and effectively Saving service costs will inevitably bring about a trend of replacing traditional network cards, with huge market prospects. Chip Qiyuan provides the industry-leading and mature DPU programming development kit SDK, which fully supports P4/C and other mixed network application programming.
Chipqiyuan DPU programming development SDK
Coreqiyuan’s complete smart network card products and solutions have won orders from many leading customers. Among them, the first batch of smart network card orders of China Mobile Research Institute are the products purchased by Coreqiyuan. At the network level of China Mobile’s big cloud system, Coreqiyuan intelligent network card transparently unloads the key cloud network flow table of the SDN controller to the network card hardware. Medium acceleration can completely release a large number of server CPU and memory resources consumed by network services, and comprehensively improve the virtualization efficiency of China Mobile’s cloud machine. At present, the two parties have established a joint laboratory to jointly promote the development of next-generation smart network cards.
Under the tide, the DPU industry pattern has changed
On the other hand, let’s take a look at the DPU chip. DPU is the abbreviation of Data Processing Unit (Data Processing Unit). engine. Under the computing power economy, DPU is developing rapidly and is expected to become the third largest computing power pillar in data center scenarios after CPU and GPU.
It can be understood that the DPU is born to process network data. The emergence of the DPU effectively relieves the pressure on the CPU. The DPU-based smart network card will become the core network component in the cloud data center equipment, and gradually undertake all network data processing and distribution. It can fundamentally realize many advantages of software-defined networking (SDN) and network function virtualization (NFV), effectively reduce the performance loss of cloud computing, release CPU computing power, reduce power consumption and greatly reduce the operation of cloud data center cost.
Comparison of the working principle of DPU (network transmission operation) and the working principle of traditional network cards (Source: Toubao Research Institute)
Data from the Toubao Research Institute pointed out that the growth cycle of China’s DPU market scale is in line with the data center upgrade cycle. Usually, the data center bandwidth upgrade cycle is about 3 years. China will enter the next round of server equipment and DPU replacement cycles in 2023-2025. The market size will increase significantly. It is estimated that by 2025, the size of China’s DPU market is expected to be close to US$4 billion.
China’s DPU market size, forecast for 2020-2025 (Source: Head Leopard Research Institute)
However, due to the technical difficulty of DPU system development and the long R&D cycle, the product supply has been monopolized by foreign companies for a long time. In 2020, China’s DPU market products are mainly provided by companies such as NVIDIA, Intel, Broadcom and Marvell. From the perspective of domestic manufacturers, my country is still in its infancy in the field of DPU, and there is a certain gap with the advanced level of foreign countries.
However, with the strengthening of software and hardware strength of Chinese enterprises and the development of soft and hard decoupling, a number of local enterprises have emerged in China’s DPU industry. Among them, as the only domestic smart network card provider, Coreqiyuan has the world’s leading DPU core chip technology. With its excellent product performance and ecological layout, the situation of international giants controlling the market is expected to be broken, and the market will usher in new changes. .
In response to the existing pattern of the DPU market, the Toubao Research Institute believes that if the initial capital is sufficient, the start-up DPU team can independently develop the processor micro-architecture, and such enterprises may become the main force to promote the domestic replacement of DPU. At the same time, the routes for Chinese local suppliers to seize market share include reducing costs and acquiring customers, and moving from customization to standardization. From customization to standardization: intermediate software service standardization + advanced software maintenance customization, infiltrating application scenarios.
At present, Chipqiyuan is undergoing a strategic transformation from an asset-light operation model of self-developed mid-to-high-level software, outsourcing processors and network chips, to a model of self-developed network chips + mid-to-high-level software, using basic network chip production lines to cooperate with software The service achieves revenue. In terms of processor deployment, the selling point is flexible adaptation, and differentiated technology paths meet user and market needs.
Looking at the industry situation, it can be found that mergers and acquisitions to expand production lines are the usual means for DPU head suppliers to consolidate their existing market positions. It is difficult for local start-ups to subvert the market position of leading companies through the same technology and products. Only by combining business models and logical innovations can they have the opportunity to break through the blockade of leading companies.
Bringing together the world’s top talents, with clear product, technology path planning and market strategy planning, Chipqiyuan is expected to become the leader of the DPU track.
Mimicpro——Starting the process of localization of high-end EDA
In addition to the smart network card and DPU chip business, Coreqiyuan has also made great achievements in the field of high-end EDA.
Looking back at the development of the semiconductor industry, China’s IC design industry has developed rapidly. There are currently more than 2,000 fabless design companies in China, but the number of EDA companies has always hovered in the single digits. In the field of high-end EDA, domestic companies have always had to rely on foreign products. And technology, which has greatly restricted the technological progress of domestic related industries and even the development of enterprises.
In order to better improve the simulation and verification efficiency of IC design customers, Chipqiyuan, which has insight into this trend, has launched its self-developed SoC prototype and simulation system MimicPro products in the high-end EDA field. The domestic market has officially mass-produced in June this year. and supply.
SoC prototype and simulation system MimicPro
As an FPGA-based prototyping platform that provides high performance, high capacity and high scalability for ASIC/SoC designers, the MimicPro product has been fully considered and designed on the combination of software and hardware, and realized the The hardware-integrated solution will provide chip design companies with a lower cost and faster computing simulation EDA solution, accelerate the design and software development and verification process, and improve the success rate of tape-out.
In addition to providing a highly reliable hardware platform, MimicPro products also have system-level Auto Partition, Built-in Memory analyzer, Local Debug and Fault injection. (fault injection) and other functions, which can locate and eliminate faults faster and achieve optimal performance. In addition, the blessing of the internationally certified USB IP and other technologies will also provide greater help for users to reduce costs and improve verification efficiency.
The MimicPro product of Chipqiyuan is currently the leading prototype verification and simulation system in China, and has won purchase orders from many domestic and foreign chip design leading customers. Behind winning the trust is the support of the top R&D team and globalization strategy of Chipqiyuan.
Looking back on the R&D process of MimicPro, Mike Shei, vice president in charge of system R&D, said: “In the past 20 years of work, my team and I have experienced the R&D of the first and second generation EDA tools. MimicPro is the third-generation EDA tool developed by us. The MimicPro project was established in January 2018. In just over 3 years, the R&D team has undergone Partition testing, Runtime R&D, and started 32- FPGA project, more than 1000 days and nights, the core Qiyuan software and hardware R&D teams from Silicon Valley in the United States, Shanghai and Nanjing in China worked closely together to take advantage of global resource allocation and fought a successful battle in the first batch of MimicPro mass production projects. , completed all the commissioning and evaluation of MimicPro, and successfully delivered it to customers.”
With the delivery of MimicPro products, the first shot of localization of SoC prototypes and simulation systems in the high-end EDA field was officially launched. It is understood that the MimicPro prototype verification and simulation system currently launched by Chipqigen is divided into 1 and 4 FPGAs. It is reported that the next generation of MimicPro products with 32 FPGAs will also be launched in the second half of this year. Officially released.
write at the end
Chip Qiyuan was established in 2015. With a mature team that has been deeply involved in the chip field for more than 20 years, Chip Qiyuan “targets” the highly-growing network communication-related fields, focusing on 5G, cloud data centers and cloud computing industries. “.
In just over five years, it has successfully developed four products including smart network card, SoC prototype and simulation system MimicPro, USB3.x IP, and network search engine TCAM chip, leading the domestic development in these fields. New situation, and won a large number of orders from first-class customers at home and abroad, helping the local industry to break through.
Over time, we expect Xinqiyuan to bring more surprises.
The Links: LQ104S1DG34 MG50J6ES50 BSM300GB120DLC