Author | Yesterday
Key Highlights:
1, FPGA downstream application scenarios are extensive, the demand is growing steadily, in the future in the 5G, AI, data center, automatic driving and other high computing demand driven, FPGA chip market demand growth is certain, according to Frost & Sullivan expects that the global FPGA market size in 2025 will be expected to reach 12.58 billion US dollars, 16-25 years of average CAGR is about 11%;
2. Compared with CPUs, GPUs, ASICs and other products, the profit margin of FPGA chips is higher: the profit margin of medium and low density million gate-level and tens of millions of door-level FPGA chip research and development enterprises is close to 50%, the profit margin of high-density 100 million door FPGA chip research and development enterprises is nearly 70%, and the gross profit margin of Xilinx in the past ten quarters remains above 65%, which is higher than the gross profit margin of NVIDIA and AMD in the same period;
3, FPGA entry barriers are high, the need for software and hardware collaborative development: FPGA dedicated EDA software technical barriers are high, the hardware structure is complex and the yield is low, making the global FPGA market a duopoly competition pattern, of which, Xilinx market share is more than half of the perennial, and Intel (Altera) accounts for more than 80% of the market share, the industry Matthew effect is obvious;
4. Benefiting from the evolution of 5G, AI, automatic driving and other technologies and the development of data centers to promote market scale expansion, xilinx, the FPGA leader, has achieved a V-shaped reversal in revenue in the past two years, and FY22Q2 revenue, gross profit and net profit have maintained double-digit year-on-year growth;
5, the advent of the 5G era, it is expected that the future market construction scale of 5G mobile communication base stations is tens of millions, coupled with the pricing of FPGAs is positively correlated with on-chip resources, the future unit price is expected to be further increased, therefore, as the core components of 5G infrastructure and terminal equipment, FPGA is expected to achieve a simultaneous increase in volume and price;
6. Due to its flexibility and high-speed computing capabilities, FPGAs are widely used in the field of AI acceleration cards, benefiting from the broad development prospects in the field of artificial intelligence, and the demand for FPGAs applied to the FIELD of AI will continue to improve in the future;
7, data center FPGA is mainly used for hardware acceleration, low latency + high throughput laid the core advantages of FPGA, compared to the traditional CPU solution, the use of FPGA solutions to deal with custom algorithms can achieve significant acceleration effect, starting from 2016, Microsoft Azure, Amazon AWS, Alibaba Cloud servers have begun to deploy FPGA accelerators;
8, the use of FPGA heterogeneous computing platform, can handle the increase in the number of sensors brought about by the explosive growth of data, shorten the overall response time of the system brought about by the synchronization and fusion of multiple sensors, and improve flexibility and scalability, to achieve scalability from edge sensors to domain controllers, while providing dynamic reprogramming capabilities, reduce system costs and losses, in mid-June 2020, FPGA leader Xilinx has about 70 million automotive chips used in ADAS;
9. In the future, driven by the 5G wave, FPGA is expected to achieve a simultaneous increase in volume and price, and FPGA leader Xilinx will continue to benefit from the demand of FPGA application markets such as AI, data centers, and automatic driving.
As one of the branches of logic chips, FPGA (Field-Programmable Gate Array) chips are based on programmable devices (PAL, GAL) and are semi-customized, programmable integrated circuits, known as "universal chips". FPGAs offer field programmability (high flexibility), short time to market (saving tapeout cycles), lower cost than fully customized ASICs (saving tapeout costs), and greater parallelism than general-purpose products.
The downstream application scenarios of FPGAs are extensive, and the demand is growing steadily, mainly covering network communications (5G), industrial Internet of Things, consumer electronics, data centers, automotive electronics (automatic driving), artificial intelligence (AI) and other fields. Among them, network communication, consumer electronics, and automotive electronics are its main application scenarios, accounting for more than 80% of the total demand. In the future, driven by the demand for high computing power such as 5G, AI, data centers, and automatic driving, the market demand growth of FPGA chips is certain. In addition, as Intel, AMD and other companies gradually combine CPUs and FPGAs in high-computing scenarios and increase investment in heterogeneous computing, the global FPGA market size will be further opened. According to Frost & Sullivan, the global FPGA market size is expected to reach $12.58 billion in 2025, with an average CAGR of about 11% in 16-25 years.
Compared with CPUs, GPUs, ASICs and other products, FPGA chips have higher profit margins. It is reported that the profit margin of medium and low density million gate-level and tens of million-level FPGA chip research and development enterprises is close to 50%, and the profit margin of high-density 100 million-door FPGA chip research and development enterprises is nearly 70%. As shown in the chart below, Xilinx's gross margin in the past ten quarters remained above 65%, higher than the gross margins of NVIDIA and AMD in the same period.
FPGA entry barriers are high, requiring software and hardware collaborative development: FPGA-specific EDA software technical barriers are high, the hardware structure is complex and the yield is low, so the global FPGA market is in a duopoly competitive pattern all year round, the top four giants are Xilinx, Intel (Altera), Lattice and Microchip, CR4≥90%. Among them, Xilinx's global FPGA market share is more than 50% all year round, the top1 concentration is second only to the PC CPU, GPU market, and Intel (Altera) together account for more than 80% of the FPGA market share, the industry Matthew effect is obvious.
There are two important indicators of FPGA: process process and logic gate density
At present, the FPGA demand structure is still dominated by processes above 28nm and logic units below 100K.
In terms of process process, due to its high cost performance and yield rate, FPGA chips with 28-90nm process occupy a major position. The advanced process has lower power consumption and higher performance, and it is expected that FPGA chips with processes below 28 nm will enter a period of rapid development. In terms of logic gate density, the FPGA chip demand for logic cells below 100K is currently the largest, followed by the 100K-500K logic unit part.
As Xilinx's largest market, the Asia-Pacific region, especially China, has a significant impact on the company's revenue. According to Frost & Sullivan, in the 2019 Chinese FPGA market in terms of sales, the market share of FPGA chips in the 28-90nm process and the FPGA chip below 28nm was 63.3% and 20.9%, respectively; the market share of FPGA chips below 100K and FPGA chips with 100K-500K logic units was 38.2% and 31.7%, respectively.
Benefiting from the evolution of 5G, AI, autonomous driving and other technologies and the development of data centers to promote market scale expansion, Xilinx, the FPGA leader, has achieved a V-shaped reversal in revenue in the past two years. Xilinx FY22Q2 revenue increased 22.1% year-over-year to $936 million, gross profit increased 16.7% year-over-year to $632 million, and net profit increased 21% year-over-year to $235 million.
As of the close of trading on January 11, 222, Xilinx had gained 49.84% in 21 years and -5.43% in 22 years so far, outperforming the gains of the S&P 500 ETF (SPY: -1.1%), the Philadelphia Semiconductor Index (SOXX: -2.04%) and the NASDAQ 100 ETF (QQQ: -3.02%) in the same period.
Founded in 1984 and inventing the Field Programmable Logic Gate Array (FPGA) in the same year, Xilinx is the world's leading provider of complete solutions for programmable logic. As the inventor of FPGAs, programmable SoCs and CAPS, Xilinx is committed to providing the industry with the most flexible processor technology, with highly flexible programmable chips powered by a range of advanced software and tools for network communications, automotive electronics, consumer electronics and data centers. At present, the company meets more than 50% of the world's demand for FPGA products.
Xilinx's revenue is primarily driven by the four businesses of AIT (Aerospace & Defense, Industrial, Test & Measurement), Automotive, Broadcast & Consumer Electronics, Wired & Wireless, and Data Center.
The communication field is the most widely used scenario for FPGAs
Compared with other types of chips, on the one hand, the programmable characteristics (flexibility) of FPGAs are highly suitable for the continuous iterative upgrading of communication protocols; on the other hand, the strong real-time processing and parallel processing capabilities of FPGAs can effectively meet the high-speed communication protocol processing requirements in the communication field. Therefore, FPGA chips are widely used in wireless communication and wired communication equipment.
With the advent of the 5G era, the volume and price of FPGAs have risen together. In terms of quantity, due to the higher frequency of 5G radio, in order to achieve the same coverage goal as 4G, it is necessary to have about 3-4 times the number of 4G base stations (taking China as an example, by the end of 20 years, the total number of China Mobile communication base stations reached 9.31 million, a net increase of 900,000 for the whole year, of which the total number of 4G base stations reached 5.75 million), and it is expected that the future market construction scale will be tens of millions. At the same time, due to the high concurrent processing requirements of large-scale antenna arrays, compared with 4G single base stations, the FPGA usage of 5G single base stations will increase from the original 2-3 blocks to 4-5 blocks. Therefore, as a core component of 5G infrastructure and terminal equipment, FPGA usage will also increase. In terms of unit price, FPGA is mainly used in the baseband of the transceiver, due to the increase in the number of channels and the increase in computational complexity in the 5G era, the scale of the FPGA used will increase, and because the pricing of the FPGA is positively correlated with on-chip resources, the unit price is expected to be further increased in the future. FY22Q2, Xilinx Wired & Wireless revenue increased 45.6% year-over-year to $290 million, accounting for 31% of total revenue.
2019-2028 China and Global 5G Base Station (Wireless Master) Investment Scale Forecast (RMB 100 million)
Source: Zhiyan Information, Founder Securities
FPGAs can be used as data center accelerators, AI accelerators, Smart NICs (smart network cards), and accelerators in network infrastructure. In recent years, due to the prosperity of artificial intelligence, cloud computing, high-performance computing (HPC), autonomous driving, etc., it has brought new market impetus to FPGAs and catalyzed incremental space.
AI accelerates FPGA requirements
Due to their flexibility and high-speed computing power, FPGAs are widely used in the field of AI acceleration cards. Compared with GPUs, FPGAs have obvious energy efficiency advantages, and compared with ASICs, FPGAs have greater flexibility, which can match the faster evolution speed of AI neural networks and keep up with iterative updates of algorithms. Benefiting from the broad development prospects in the field of artificial intelligence, the demand for FPGAs applied to the field of AI will continue to improve in the future. According to SemicoResearch, the market size of FPGAs in AI application scenarios will triple in 19-23 years to reach $5.2 billion. Compared with the $8.3 billion FPGA market in 21 years, the application potential of AI cannot be underestimated.
FPGA has great potential in AI applications
Source: Founder Securities
A more promising market for FPGAs is the data center
Data centers are one of the emerging application markets for FPGA chips, and low latency + high throughput have laid the core advantages of FPGAs. Data center FPGAs are mainly used for hardware acceleration, compared with the traditional CPU scheme, the processing of custom algorithms can achieve significant acceleration effect: for example, the Microsoft Catapult project in the data center uses FPGAs instead of CPU schemes, when processing Bing's custom algorithms 40 times faster, the acceleration effect is remarkable. Therefore, starting in 2016, Microsoft Azure, Amazon AWS, and Alibaba Cloud's servers have all begun to deploy FPGA accelerators for computing acceleration. In the context of the acceleration of global digital transformation by the epidemic, the requirements for chip performance in future data centers will be further improved, and more data centers will adopt FPGA chip solutions, which will also increase the value ratio of FPGA chips in data center chips.
Source: Gartner, Cisco, China IDC Circle, Founder Securities
Large-scale commercialization of autonomous driving increases the demand for FPGA mass production
In the continuous evolution of the automotive industry from ADAS to fully autonomous driving, through the heterogeneous computing platform using FPGAs, it can handle the explosive growth of data caused by the increase in the number of sensors, shorten the overall response time of the system caused by the synchronization and fusion of multiple sensors, and improve flexibility and scalability, realize scalability from edge sensors to domain controllers, and provide dynamic reprogramming capabilities to reduce system costs and losses. In addition, FPGAs provide flexible, low-cost, high-performance solutions for a wide range of rapidly growing automotive electronics application needs. In mid-June 2020, FPGA leader Xilinx has about 70 million automotive chips used in ADAS.
AMD's xilinx acquisition was postponed to 22Q1
Following Intel's acquisition of FPGA Dragon 2 Altera in 2015, AMD announced in October 2020 that it plans to spend $35 billion (in stock form) to acquire FPGA giant Xilinx, trying to expand its TAM by entering the FPGA market, while enriching the product line and forming a complete high-performance computing system with existing CPU processors, GPU graphics cards, and accelerated computing cards. According to the latest news on December 31, 2021, the acquisition is expected to be completed at 22Q1 as all approvals have not yet been obtained, which is a delay from the originally expected schedule.
In the future, driven by the 5G wave, FPGA is expected to achieve a simultaneous increase in volume and price, and FPGA leader Xilinx will continue to benefit from the demand for FPGA application markets such as AI, data centers, and autonomous driving.
Note: This article is original by the US stock research agency team, please indicate the source when reprinting, thank you!