On February 14 of this year, a $49.8 billion acquisition transformed AMD, which had been lying dormant for several years, into a global leader in FPGAs. The target of the acquisition, Ceres, also “married into the family” and became a central player in AMD’s “grand strategy” for heterogeneous computing clusters.
Perhaps spurred by the M&A of the century, the Chinese FPGA track on the other side of the ocean has also been soaring since the beginning of the year, with continuous financing.
China’s FPGA track intensifies involution, 16/28nm competition: high-end domestic chips still need 3-5 years
On May 18, Guangdong GaoYun Semiconductor announced the completion of a heavy B+ round of financing with a total size of RMB 880 million; at the end of May, CSCYH also completed a B round of financing with a total size of RMB 300 million; in June, JingWei QiLi, which focuses on general-purpose FPGA chips and a new generation of heterogeneous programmable computing chips, also officially announced the completion of strategic financing of tens of millions of RMB.
So far, following the listed Fudan Microelectronics, Anlu Technology and Chengdu Huawei to be IPO, as well as the head enterprises such as ZiGuang Tongchuang, a number of domestic mainstream FPGA companies, including Zhidocrystal and GaoYun Semiconductor, have now been tied to the chariot of capital.
Under the helm of capital, coupled with the domestic alternative and the wave of chip shortages, domestic FPGAs will open up a new boom and bust.
After the AI chip fever, FPGA regains capital favor
In the past few years, the AI chip boom stirred, so that the FPGA track as one of the core processors of arithmetic power also experienced a wave of growth peak. At the time, FPGAs and ASICs, GPUs, XPUs and other arithmetic chips were in a fierce competition, and there was a lot of debate in the industry about who was the best route for AI chips.
But as the fever fades, the capital fever of AI chip track has long been gone, and industrial investment has gradually returned to rationality, which also directly affects the once hot FPGA field, investment and financing has naturally cooled down. However, under the impetus of chip local substitution, the real high-quality targets can still get a good amount of financing, and this is naturally the case in the FPGA field, such as the 300 million RMB B round of financing of CSCYH and the 880 million B+ round of financing of GaoYun Semiconductor.
The reason why capital remains bullish on the FPGA track is closely related to the high retention rate and strong life cycle of FPGAs.
Especially in the years after the outbreak, FPGAs not only retain their value but also rise outrageously. The head company Xilinx XC6SLX16-3CSG225I (product model), for example, according to the set of micro network from the channel, the current product unit price of this FPGA has been as high as 5,000 (Huaqiang North price) to about 10,000 yuan, while in the epidemic has just begun (early 2020), the FPGA offer only in 800-1200 yuan, which can be seen the alarming rate of increase.
Naturally, there is no shortage of agents and channel speculation, but the overall lack of stock market, such fluctuations, but also let the market once again feel the FPGA compared to other categories of chips in the price of the firm degree.
From the domestic brand of low-end FPGA products in recent years, a glimpse of the price changes can be seen. Compared with the head company Ceres, domestic brands such as Anlu Technology’s FPGA products price changes are not obvious, but also still show a high rate of retention.
According to the IPO prospectus disclosed before the official IPO of Anlu Technology, in 2018, the average unit price of ELF low-power series FPGA products was only $7.37 / piece; in 2019, the price rose directly to $12.96 / piece; in 2020, it was reduced but still maintained at the price of $12.23 / piece, fluctuating; but in the first half of 2021, it rose again to $13.84 / piece. However, in the first half of 2021, the price rose again to $13.84 per unit. Despite the low-end positioning of Anlus’ FPGA products, the price has continued to increase rather than decrease over the past four years, which is certainly influenced by the lack of core wave, but the high retention rate of FPGA chips is also one of the most critical elements.
China’s FPGA track intensifies the inward volume, 16/28nm competition: the high-end of domestic chips still need 3-5 years
After all, from the point of view of cost and cycle, it is not easy to develop an FPGA, and some industry insiders have told Jiwei: “FPGA life cycle can reach 15-20 years of products, and behind to pay a huge investment. For example, the R&D cycle of an FPGA product with a new process is about 3-5 years, and it takes 1-2 years for the product to mature and mass production after successful R&D. It takes 1-2 years for mass production to land on customer projects, and it takes about 3 years for the product to really become mainstream in the market after landing.”
Therefore, to successfully implement an FPGA chip has a very long lead time, roughly estimated that it takes about 5-8 years. Not to mention the flow of more than tens of millions of dollars, which determines the FPGA from the beginning to the end is an extreme test of endurance, technology and financial resources and other aspects.
All want to be the next “Celeris”
Every Chinese chip company has the dream of becoming an industry leader or even a world-class chip giant, especially for FPGAs, which are a high threshold track. Even though they are facing increasingly brutal market competition, capital and profitability pressures, any domestic FPGA company is continuing to raise funds and dispatch resources through various channels to become the next “Xilinx”.
At the moment, only the fastest-running companies can really eat the wave of lack of cores, domestic substitution and foreign supply chain discontinuity dividends. All companies are striving to enlist as many customers and industry resources as possible within the shortest possible time, so that the products can be imported into various available projects to maximize the benefits.
In order to compete for customers and project resources, the domestic FPGA companies have opened up an open and dark battle, the most direct is to play the price war. Some industry vendors say: “Even now there are only ten domestic FPGA manufacturers, but for the domestic FPGA circuit has been quite a lot, which will inevitably lead to the enterprise with a price cut to enlist the market, and now, the FPGA industry price war has begun to fight.”
This is also in line with the “blood” and “greed” of the capital market, especially at a time when most domestic civilian FPGA products are mostly positioned in the low-end market, the price of crushing rivals to grab customers has become the consistent way of many companies. With the subsequent domestic FPGA track competition continues to intensify, some industry insiders said that this price war may be more and more fierce, which will also accelerate the arrival of the “reshuffle period.
The reason behind it can be imagined, because in the eyes of all kinds of crazy capital, China’s “Celeris” can only have one in the end. The final winner of the crown must be able to get the most customer resources in the domestic FPGA market “brutal growth” price war period, the fastest commercialization of profitability and the most potential to enter the civilian high-end FPGA market and large-scale release of the enterprise. The remaining followers will either become China’s “Lattice” or wait until the market dividend period is over and be “killed” by the recovering overseas FPGA brands.
The battle for 16/28nm node, high-end domestic FPGAs still need 3-5 years
Capital boost, even though the domestic FPGA track added enough gunpowder flavor, but the chip iteration must be on the ground, especially in such high threshold areas as FPGAs, from low-end to high-end is not a one-time success. For the time being, specific to the process node, according to the set of micro network understanding, most domestic shipments are concentrated in 40nm, 55nm process level, but the most profitable foreign is still 28nm process node, medium and large scale devices, the process gap between domestic and foreign at least 2-3 generations.
However, there are industry views that, with the domestic improvement in technology and the expansion of talent, 2022 to 2023, the entire domestic FPGA will have greater competition in the 28nm process node.
A domestic FPGA chip company insider told me: “16nm and 28nm will be the next mainstream nodes of the domestic FPGA market competition, these two nodes alone, it is enough for domestic FPGA manufacturers to eat for several years. And at present, the domestic FPGA to higher process such as 7nm direction to do is actually not very meaningful, because the high-end FPGA field of customer volume is too small, it is difficult to make up the chip from research and development, flow to mass production and delivery of the entire process cost, for the current domestic rely on financing a large number of FPGA chip company to live more than the loss, not to mention that in the current domestic civilian FPGA field no A company really profitable situation.”
Therefore, such as 16nm and 28nm such relatively mature process products, will be the home of the domestic FPGA companies in the next few years of the battlefield. And domestic FPGAs want to 14nm billion gate level or even 7nm and other higher-end areas have been built, the industry insiders said frankly less than 3 years, long will take 5 years.
Behind this, also depends on a variety of factors, the FPGA chip company insider said: “specific including R & D cycle, hardware, software design, and customer cycle, etc.. For example, the software algorithm, the current domestic FPGA original factory large-scale EDA tools focused on 100k logic scale, more than 300K, the software is not a high rate of cloth, which is a greater challenge for each original factory.”
In addition, in terms of process, domestic FPGAs also need to domestic such as SMIC such as chip foundry giant in such as 7nm and other more advanced process iterations are no longer subject to external restrictions. Only by truly breaking through the market and technology blockade in the entire supply chain, domestic FPGAs can gradually come out of a high-end road.