Not long ago, Intel announced through its official website that it would spin off the Programmable Solutions Group (PSG), which is responsible for developing Intel's Agilex, Stratix and other FPGA products, and operate it as an independent business. The goal is to sell part of the business in an IPO in two to three years.
Affected by the recession in the personal computer market, Intel's recent revenue performance has been slightly weak. The first quarter of fiscal year 2023 hit the lowest quarterly revenue in the past 13 years and suffered continuous losses. The revenue performance of this division has been good in the past.
Although in the first quarter of 2022, Intel stopped treating PSG as an independent financial department. According to previously disclosed data, the average revenue of this department per quarter is very stable, around US$500 million, and even set a record of US$700 million in the last few quarters before the announcement.
The reason why Intel divested PSG was not because the business unit was underperforming or in a commoditized, low-margin market, but because it believed that PSG might perform better without Intel's huge business and pressure.
Intel believes that PSG has been unable to meet the real high-growth, high-profit market demand for FPGAs, such as industrial, automotive, defense and aerospace. According to data provided by Intel, the global FPGA market may have a compound annual growth rate of more than 9% in the next five years, expanding from US$8 billion in 2023 to US$11.5 billion in 2017.
FPGA
FPGA (field programmable gate array) is a reconfigurable computer chip that can be programmed to implement any digital hardware circuit. FPGAs can be reprogrammed after fabrication to simulate digital circuits, making them ideal for prototyping new features before volume production, or to serve rare use cases where custom chips would not be economical.
In the history of integrated circuits, FPGAs are not the only devices that can achieve this function, but FPGAs are by far the most commercially successful.
In the 1970s, if you wanted to prototype and test the logic functionality on actual hardware, you basically had two options
One is a TTL chip, but TTL chips will be affected by circuit board size and power consumption limitations
The other is custom chips (ASICs), but custom chips have high upfront costs. There’s the cost of a photomask and then months of waiting for manufacturing. If you make a mistake in the design, you won't be able to accurately edit the JS file and recompile it
In 1980, Altera was established. Altera pioneered complex programmable logic devices that can take a stack of smaller PALs and connect them using cross-connects to increase scalability. Altera's complex programmable logic devices take a step forward
After that, Xilinx created and released the first FPGA, Xilinx XC2064. The XC2064 is a large chip with approximately 85,000 transistors arranged into 64 configurable logic blocks and 58 input/output blocks. Although not surprising today, the first FPGAs gained a foothold in the market
After that, with the significant drop in transistor costs and the emergence of independent foundries, FPGA stood at the intersection of technology and business and began to flourish.
The monopoly era of Altera and Xilinx
Xilinx and Altera continue to launch new products to improve logic density and performance, and actively work with customers to develop customized solutions to continue to occupy market share. The global FPGA market once formed a monopoly between two giants. These two companies accounted for nearly 90% of the market (Xilinx ranked first with a share of 56%, and Altera ranked second with a share of 31%).
In the past 40 years, more than 60 companies have been engaged in the research and development of FPGA technology and products, including large and powerful companies such as Intel, IBM, AMD, TI, GE, AT$T, Motorola, Samsung, Toshiba, and Philips. , but most of them fall short after spending hundreds of millions of dollars.
Around the data center, the era of "acquisition" of FPGA giants
Intel spent $17.6 billion to acquire FPGA maker Alter. Based on Altera, Intel established the Programmable Division. Over the next several years, Intel built an impressive FPGA business
An important reason why Intel acquired Altera is to make up for the impact of slowing demand in the personal computer (PC) field by increasing high-profit chip products in data centers. Three years after the acquisition, Intel announced that its FPGAs have been officially used in mainstream data center OEM manufacturers
In 2022, AMD officially announced its acquisition of Xilinx. Through these acquisitions, AMD can further expand some key markets, such as the data center market, as well as 5G communications, automotive, industrial, aerospace and defense markets. AMD also hopes to rely on Xilinx FPGA to supplement data center capabilities
Looking at it this way, both Intel and AMD have set their sights on the data center market early.
Currently, there are two main trends in data centers:
First, the increase in data traffic
Second, the increase in compute-intensive applications
The main challenge is that data centers must not only be able to handle increased data and more rigorous calculations, but also need to use lower power consumption and higher performance
Data centers have always been a market where FPGA has high expectations. The reconfigurability and parallel processing capabilities of FPGAs can be seamlessly integrated into existing HPC infrastructure, complementing traditional CPU-based clusters and GPU-based systems. By offloading specific tasks to FPGAs, HPC systems can achieve higher performance, lower power consumption and increase efficiency
What’s next for FPGAs?
Separating FPGAs from Intel's core business raises questions about AMD's $49 billion acquisition of Xilinx. Intel, with its compact portfolio, and AMD, with its comprehensive chip portfolio, are heading in opposite directions
However, focusing on AI has become a consensus between the two companies. Currently, the three giants Intel, AMD, and NVIDIA have their own rich AI software and hardware solutions. CPUs and GPUs can do it, but they are different.
Intel CPU has a long history and has always occupied a leading position; NVIDIA GPU has great advantages, but the CPU is limited and uses the Arm architecture. AMD has the strongest comprehensive performance, and its products include CPU, GPU, FPGA and adaptive SoC products.
FPGA has been around for many years. With the rise of artificial intelligence, FPGA is a suitable choice for artificial intelligence reasoning needs that require frequent changes in the underlying model.
It remains to be seen whether FPGA will be split and independent, or whether merger and development will better adapt to the trend of the times.
