According to a report from Electronic Enthusiasts (by Huang Jingjing)Xilinx launched its first FPGA chip, the XC2064, in June 1985, featuring 600 gates, 64 configurable logic blocks, and a clock frequency of 70 MHz.This field-programmable gate array (FPGA) enabled engineers to flexibly customize chip functions without waiting for chips to return from the foundry, significantly accelerating chip development and reducing time to market.Kirk Saba, Vice President of Products, Software, and Solutions at AMD, stated that the 40th anniversary of FPGA is a very important milestone, demonstrating the significance of FPGAs as an indispensable technology that supports a wide range of applications and innovations.
Edge intelligence
Various computing engines, including FPGAs, CPUs, GPUs, and ASICs, are being utilized in edge intelligence scenarios, and Kirk Saba believes that FPGAs are best suited for edge applications, especially in situations requiring real-time processing. In contrast, edge GPUs often have higher power consumption.The key value of FPGAs lies in their ability to be programmed and upgraded on-site, and FPGAs can be flexibly deployed in terms of both power consumption and performance. He further pointed out that to fully leverage the capabilities of the underlying hardware, it is essentially a system-level design issue, and it is also necessary to make full use of the storage architecture of the devices to realize their maximum potential. “FPGAs have customizable storage architectures that can meet different algorithmic needs of customers.“ Kirk Saba said. AMD offers a wide range of FPGA products, including the smallest and lowest power FPGAs, such as Spartan UltraScale+, as well as Spartan 7, Artix 7, and Artix UltraScale+, which can provide different customization needs for customers. If customers require embedded processing capabilities, they can choose from the “Zynq 7000” and “Zyng MPSoC” product lines. In summary, the programmable capabilities of FPGAs can provide solutions comparable to ASICs or ASSPs. At the edge, we are seeing an increasing number of smart devices being deployed and interconnected, such as numerous sensors, cameras, and other devices that will incorporate edge AI. FPGAs can help customers decide how much AI to deploy on the devices, Kirk Saba stated, and in the future, we will see many unique use cases for AI deployment at the edge, a trend that is continuously evolving.
Heterogeneous Computing
Years ago, AMD proactively established a heterogeneous computing platform. Starting with the Versal product family, it features embedded computing capabilities along with multiple ARM processing cores, while the hard NPU Virtex can serve as an AI engine. Programmable logic FPGAs have many hardened IPs, such as hardened Ethernet cores and hardened security cores, which can support heterogeneous computing. They also provide flexibility in combining programmable logic, AI engines, and specific applications, offering users an embedded computing assistance method, and run corresponding embedded operating systems, such as Linux. “ We have already seen this disruptive change happening, including the division of roles between different chips. Our long-standing value proposition has been to integrate different technologies into a single chip.“ Kirk Saba concluded. AMD has a long history in small chip technology, which will also be widely applied to FPGAs. This technology originated from AMD’s collaboration with TSMC to develop CoWoS with Virtex-7. Subsequently, GPUs also adopted small chip technology and memory integration. AMD will continue to develop products based on advanced process nodes, and the latest products will be launched in the near future. Over the past 40 years, FPGA development tools such as Vivado and Vitis have continuously optimized to adapt to the needs of hardware engineers. Kirk Saba stated that the Vivado software tool primarily focuses on hardware developers, supporting developers in streamlining workflows, shortening development cycles, and achieving higher performance through high-level synthesis, machine learning optimization, and seamless IP core integration. Vitis, on the other hand, targets software developers. Whether writing embedded C code or HLS code, it can perform compilation and more. For AI engines, the Vitis AI tool focuses more on integration with AMD CPU and GPU software. Under an integrated AI software framework, it enables better support for customers in training models on AMD hardware and multi-platform deployment. Additionally, AMD is incorporating generative AI technology into the FPGA development process, introducing AI-assisted development tools to make complex hardware programming smarter. In terms of the AI software ecosystem, AMD is also continuously refining and enhancing. Earlier, AMD acquired the software company Mipsology, whose core technology is the Zebra compiler. This move will help improve AMD’s compiler capabilities in the AI field, enhancing software efficiency on hardware and strengthening its AI inference software capabilities. Kirk Saba believes that investment in software will better unleash AMD’s potential in the hardware domain. AMD has always strongly supported and advocated for open-source ecosystems, becoming a significant contributor to the open-source community through collaboration with Yocto. This greatly helps developers accelerate development efficiency and ensures they have access to robust, world-class software tools. Furthermore, AMD is promoting the open-source ecosystem of ROCm in the GPU software domain. In summary, open-source will continue to be a very important differentiator for AMD.
FPGA Embraces the AI Era
AMD has a rich and comprehensive product matrix, offering CPUs, GPUs, FPGAs, and other chips to meet customer needs for different computing performance in consumer, enterprise, and data center sectors. The next wave of semiconductor technology belongs to AI, and after 40 years of development, FPGAs will play a crucial role in the AI era. Kirk Saba stated that industries such as AI, automotive, and robotics are developing rapidly, and these technologies require FPGAs. For example, we see the rapid development of new energy vehicles in China, where modern cars are like computers on four wheels, requiring a range of electronic components, and FPGAs have significant potential in areas such as ADAS, intelligent driving, and in-vehicle entertainment systems. Additionally, the automation and intelligence of edge devices like drones and robots will present tremendous opportunities for AMD’s technology and capabilities in edge AI.
Disclaimer：This article is originally from Electronic Enthusiasts and must be cited when reproduced. For group discussions, please add WeChat elecfans999, and for submission of interview requests, please email [email protected].
More Hot Articles to Read
-
“Presidential Phone” priced at $499! Trump’s American-made, challenging the iPhone?
-
Reports of a certain chip team facing rapid layoffs! This market competition is too fierce!
-
Nezha Auto releases restructuring statement! Once topped the new power list, how did it fall into trouble?
-
Is it the only one? IBM updates quantum computing roadmap: delivery in 2029!
-
Performance skyrockets! AMD’s new product challenges Nvidia’s GB200, competing in the $500 billion AI accelerator market.
Click to Follow and Star Us
Star us to not miss any updates!