FPGA 탄생 40주년을 맞아 살릴 라지(Saili Raje) AMD 어댑티브 및 임베디드 컴퓨팅 그룹(AECG) 총괄 수석 부사장이 엣지 AI에서의 FPGA의 지속적 혁신에 대해 이야기한다.
.jpg)
▲Saili Raje, Senior Vice President and General Manager of AMD Adaptive and Embedded Computing Group (AECG)
“FPGA 40th Anniversary, Leading Continuous Innovation in Edge AI”
FPGA·Adaptive SoC, Real-time Sensor Data Processing, Edge AI Inference Acceleration
AMD Continues to Invest in Tools for Improving User Productivity and Efficiently Responding to Edge AI
This year marks the 40th anniversary of the introduction of the first commercial field-programmable gate array (FPGA), which introduced the idea of programmable hardware.
The FPGA's reprogrammable architecture, born from the goal of "hardware as flexible as software," has changed the landscape of semiconductor design.
For the first time, developers could design their own chips, and then redefine their functionality to adapt them to new tasks as specifications or requirements changed during development or after manufacturing.
This flexibility allows for faster development of new chip designs, reducing the time it takes to get new products ready, while establishing itself as a new alternative to application-specific integrated circuits (ASICs).
> The impact FPGAs have had on the industry is amazing.
FPGAs have created a massive industry worth more than $1 billion.
Over the past 40 years, we have shipped over 3 billion FPGAs and adaptive SoCs (devices that combine FPGA fabric with system-on-chip and other processing engines) to over 7,000 customers across a broad range of industries.
In fact, we have been the number one programmable logic market share leader for the past 25 consecutive years, and we expect to continue to maintain this market leadership based on the strength of our product portfolio and roadmap.
■ Acceleration of Innovation
FPGAs were first invented by the late Ross Freeman, co-founder of Xilinx (now a subsidiary of AMD).
He was an engineer and innovator who believed there had to be a better, more technologically and cost-effective way to design chips beyond the limitations of standard ASIC devices, which are fixed in function once designed.
FPGAs give engineers the freedom and flexibility to change the chip design at any time, allowing them to develop and design custom chips in as little as a day.
FPGAs also helped pioneer the "fabless" business model, which transformed the entire semiconductor industry.
FPGAs have accelerated the hardware development process by creating custom mask tools and innovations that reduce one-time development costs over repeatable ones.
This approach proved that a company doesn't need to own a foundry to create groundbreaking hardware; all that matters is vision, design skills, and FPGAs.
enter;">
▲Ross Freeman (right) examines the XC2064 layout.
Since the introduction of the world's first commercial FPGA (XC2064) 40 years ago, FPGAs have been widely used across the electronics industry and have become deeply embedded in our daily lives.
Today, adaptive computing devices, including FPGAs and adaptive SoCs or System-on-Modules (SOMs), are used in a wide range of applications, from cars, trains, and traffic lights to robots, drones, spacecraft, satellites, wireless networks, medical and test equipment, smart factories, data centers, and even high-speed stock trading systems.
▲The world’s first commercial FPGA, XC2064, consists of 85,000 transistors, 64 configurable logic blocks, and 58 I/O blocks. By comparison, today’s most advanced AMD FPGA-based device, the Versal Premium VP1902, has 138 billion transistors, 18.5 million logic cells, 2,654 I/O blocks, up to 6,864 DSP58 engines, and a wide range of hardware IP for memory, security, and interface technologies.
■ Key innovations and product journeys
AMD's innovation and evolving market demands have led to numerous remarkable advancements in FPGA technology over the past 40 years, including:
- 1985: XC2064 - First commercial FPGA released
- 1990s: XC4000 and Virtex™ FPGAs - First with embedded RAM and DSP for wireless infrastructure
- 1999: Launch of the Spartan family of products - providing a cost-effective option to replace traditional ASICs for high-volume applications
- 2001: First FPGA with integrated SerDes launched
- 2011: Virtex-7 2000T - Industry's first mass production application of Chip-on-Wafer-on-Substrate (CoWoS) packaging. Pioneered cutting-edge 2.5D integrated technology that became the foundation of high-performance computing (HPC) systems, and is now used as a core technology driving GPU innovation for AI.
- 2012: Zynq family - first adaptable SoC combining Arm CPU and programmable logic
- 2012: Vivado™ Design Suite - Making FPGA design more accessible to software developers
- 2019: Launch of the first Versal adaptive SoC - Introducing a dedicated AI engine and programmable on-chip network (NOC)
- 2019: Vitis™ unified software platform - providing pre-optimized AI tools and abstraction layers to accelerate inference
- 2024: Versal AI Edge Series Gen 2 - The first end-to-end AI acceleration that integrates programmable logic, CPU, DSP, and AI engine on a single chip, supporting next-generation applications that demand heterogeneous computing, low latency, and low power performance.
- 2024: Launch of the Spartan UltraScale+ FPGA family - Expanding the broad portfolio of cost-optimized FPGAs and adaptable SoCs that deliver cost- and power-efficient performance and are ideal for I/O-intensive applications at the edge
The introduction of Vivado and Vitis software has contributed significantly to the expansion of the FPGA market. Vivado enables streamlined workflows, shorter development cycles, and improved performance through high-level synthesis, machine learning-based optimization, and IP core integration.
The Vitis development environment helps accelerate AI inference with pre-optimized tools and abstraction layers.
The latest version (2024.2) adds a standalone tool for embedded C/C++ design and features to simplify use of the AI engine on AMD Versal adaptive SoCs.
AMD continues to invest in relevant tools to increase user productivity and efficiently respond to new data types and AI models.
■ AI at the Edge
Most AI today runs on data center GPUs.
But more and more AI processing is happening at the edge.
FPGA technology is at the forefront of supporting the rapid growth of AI-integrated applications across a wide range of industries.
FPGAs and adaptive SoCs are accelerating AI inference at the edge, processing sensor data in real time and without latency.
With the advent of small generative AI models, we will witness a new ‘ChatGPT moment’ at the edge.
These new AI models can run on edge devices, including AI PCs, vehicles, factory robots, space, and any other embedded application.
Next up is AMD's adaptive Here are some examples of how computing technologies are enabling edge AI workloads today.
- NASA - AMD Virtex FPGAs are powering NASA’s Mars rover, expanding its AI capabilities to support image detection, alignment and correction, and filtering out useless data before transmitting it back to Earth. Additionally, the latest space-grade Versal AI Edge Adaptive SoC accelerates AI inference in space with an enhanced AI engine optimized for machine learning applications.
- Subaru - Subaru has selected AMD's 2nd Generation Versal AI Edge Series adaptive SoC to bring AI capabilities to its next-generation ADAS, the EyeSight driver assistance safety system.
- SICK - AMD Kintex UltraScale+ FPGAs and the FINN machine learning framework are driving SICK’s factory automation through fast and accurate parcel inspection.
- Radmantis - AMD Kria adaptive SOM device enables real-time AI inference for sustainable aquaculture.
- JR Kyushu, one of Japan’s largest high-speed train operators, is leveraging AMD CREA SOM-based real-time image processing technology for its AI-based track inspection system.
- Clarius - AMD Zynq UltraScale adaptive SoC enables AI to identify regions of interest in portable ultrasound devices.
■ Outlook for the future
Looking ahead, FPGA-based adaptive computing will continue to drive innovation in edge AI applications for autonomous driving, robotics and industrial automation, 6G networks, climate change, drug discovery, scientific research, and space exploration.
As we celebrate the 40th anniversary of the FPGA, we are incredibly proud of its invention and reflect on how far this technology has come and how much more impactful it will be in the next 40 years.
FPGA technology continues to be used to develop cutting-edge and market-leading products. We design innovative chips, enhance hardware-based verification, and reduce time to market.
AMD will continue to lead the advancement of this amazing technology in the future.
.jpg)
