Proprietary FPGA Chips and a New Gen ACAP Platform
American chip company Xilinx is an industry pioneer, with 34 years of experience and counting. And yet the company remains a niche supplier lacking the star power of chip giants like Intel, whose processors are found in most PCs; or Nvidia, whose GPU are the choice for most AI applications.
Xilinx is however now moving into the spotlight; targeting AI applications from autonomous vehicles to data centers with its proprietary FPGA chips. First invented by Xilinx founder Ross Freeman in 1985, FPGAs are often referred to as “universal chips” as they can be reprogrammed and customized with specific software after manufacture. The chips are mostly used to complement Intel CPUs and other central processing units, handling specialized computing tasks such as video compression and data encryption.
The chips market is undergoing a tectonic shift. Upgrading multi-core CPUs is becoming more difficult as Moore’s Law tops out; while highly specialized ASIC chips cannot keep pace as AI algorithms iterate rapidly. This has created space for FPGA chips, whose steady growth in popularity has created a market now worth US$5 billion.
Xilinx, Altera, Lattice, Microsemi, and Microsemi are the four Silicon Valley companies that dominate the current FPGA chip market. Xilinx and Altera hold nearly 90 percent of the total market share, along with patents in more than 6,000 related technologies, constituting a solid technical entrance barrier. After the acquisition of Actel and Altera by Microsemi and Intel respectively, Xilinx is the main independent in the group.
One drawback for FPGAs is that while they are widely used in product research and prototyping, developers tend to replace them with non-reprogrammable, dedicated semiconductors once a product is ready for market. Xilinx CEO Victor Peng is working hard to change this.
This March in Beijing, Peng unveiled his company’s new Integrated Adaptive Compute Acceleration Platform (ACAP), an integrated multi-core heterogeneous computing platform that flexibly changes hardware layers based on application requirements. At the heart of the ACAP is a next-generation FPGA architecture that combines distributed memory with hardware-programmable DSP blocks and programmable and hardware-flexible computing engines, all interconnected via a network-on-chip (NoC).
Peng, who was named Xilinx CEO in January, remarked that “after putting one billion dollars in R&D, we expect ACAP to become a mainstream computing platform with the same importance as CPU and GPU, although this may take many years to realize.”
Making Acquisitions and Withstanding Competition
Xilinx’s current market capitalization has reached US$21.3 billion, while its share price has risen 8 percent in the past year — outperforming Intel. The company’s second quarter financial report for the 2018 fiscal year shows record revenue of US$746 million, a 19 percent year-over-year increase. Xilinx has now achieved positive growth for 12 consecutive quarters.
This July Xilinx announced one of the biggest acquisitions in company history, nabbing Chinese startup Deephi Tech for US$300 million. With 200 employees, Deephi is involved in FPGA chip applications for security, and other terminal and cloud computing markets. The move was seen by many as Xilinx preparing to step up and enter bigger markets.
When Altera was acquired by Intel for US$16.7 billion back in 2015, Xilinx found itself in competition with a chip giant that has annual revenue almost 30x its own US$2.54 billion. One of Altera’s largest customers is Microsoft, which uses Altera chips on data center computers to accelerate the compression of data and run AI algorithms for tasks such as image recognition and other functions.
To stay competitive, Xilinx is now aggressively entering new markets. Amazon’s live broadcast service platform Twitch, the largest and fastest-growing online live video service provider in North America, announced earlier this month that it will use Xilinx chips to create the industry’s first broadcast-quality live streaming media platform in VP9 video encoding format. In June, German automaker Daimler announced that it would use Xilinx chips to run AI software in its cars. According to sources familiar with the matter, Google intends to integrate Xilinx chips into the Google Cloud data center.
Amazon Web Services (AWS) new F1 instance acceleration application service is based on Xilinx FPGA chips — pointing to growing interest in Xilinx among software developers. Applications such as video processing and data analysis in the cloud can benefit from the low-cost hardware acceleration provided by FPGA chips. Tencent Cloud, Alibaba Cloud, and Huawei have followed suit and launched their own FPGA acceleration application rental services.
Source: Synced China
Localization: Meghan Han | Editor: Michael Sarazen
0 comments on “Can 34-Year-Old Chipmaker Xilinx Challenge the Big Players in AI?”