What Is FPGA?
FPGA, which stands for Field Programmable Gate Array, is a device that integrates logic circuits that designers can program in the field.
While dedicated logic ICs have fixed circuits and require re-design/re-manufacturing of masks when altering a portion of the circuit, FPGAs are characterized by their logic circuits that can be freely changed by the designer.
FPGAs were developed by Xilinx in the U.S. FPGAs are large-scale PLDs that can be modified countless times by writing the circuit configuration into SRAM.
Applications of FPGAs
FPGAs find applications in various sectors, including automotive devices, data sensors, and deep learning. Large-scale logic circuits are employed to perform high-speed logic operations that cannot be programmed by the CPU. One solution is to design and manufacture dedicated LSIs (such as ASICs). However, dedicated LSIs are difficult to change circuits.
On the other hand, FPGAs allow circuit designers to freely design application circuits and easily change circuits, thus significantly reducing the development cost of logic circuits. These features have made FPGAs widely used in a variety of fields.
1. Automotive Equipment
Reasons for the adoption of FPGAs in automotive equipment include shortened development cycles, flexibility for modification, and the emergence of devices that meet quality requirements. A specific example is video analysis for driver assistance systems.
Driver assistance systems need to instantly analyze real-time video signals from in-vehicle cameras to assist the driver’s operations. This requires low latency and high precision algorithms. FPGAs are suitable for this purpose because they require high-speed arithmetic processing and the electronic control functions in the device can be changed as needed.
2. Data Centers
FPGAs are increasingly being used in data centers. In particular, FPGAs are replacing CPUs to handle AI, security, authentication, real-time analysis, deep learning, and other processing. FPGAs are also being used to improve the performance of large data systems. They provide high-bandwidth and low-latency connectivity to network/storage systems to accelerate data processing. It also supports functions like data compression and fill processing, among others.
3. Deep Learning
In the world of deep learning, the flexibility of FPGAs to change circuits is extremely useful because optimal modeling is constantly evolving. FPGAs are ideal devices for applications that require frequent system upgrades, such as this application.
FPGA Principles
FPGAs are LSIs based on a structure in which programmable, relatively small-scale logic blocks are arranged in a grid with vertical and horizontal wireways between them. Although each logic block is small in scale, many blocks can be combined to form a large-scale circuit.
The basic logic block consists of a LUT (Look Up Table), flip-flops, and additional circuits. Logic blocks can be connected arbitrarily by means of a switch matrix (transfer gate) provided in the wireway.
The LUT uses SRAM for its operation. The ON/OFF of the switch matrix is also controlled by the data written to the SRAM. Since the data in the SRAM is lost when the power is turned off, the FPGA reads circuit information (configuration data) from the outside when the power is turned on.
The internal structure of an FPGA includes various components such as basic logic blocks, internal wireways, dedicated clock routing, multiplier (DSP), I/O section, PLL, and block RAM. These are arranged in a mesh pattern for easy routing of any circuit pattern.
Other Information on FPGAs
1. Design Tools
Traditionally, RTL (Register Transfer Level) has been used as the design language for FPGA design. Based on the designer’s RTL, a download file to be written into the FPGA was generated from the tools provided by the FPGA vendor.
In recent times, however, FPGA vendors have released tools called high-level synthesis compilers. By using this high-level synthesis compiler, efficient design is possible, and at the same time, circuit verification time is reduced. As a result, it contributes to shortening the product development time.
Presently, FPGA vendors offer the following three high-level synthesis compilers.
- Model-based (DSP) compilers.
- HLS compiler
- OpenCL compiler
Evaluation boards are usually used to study circuits using FPGAs. These are sold by a variety of companies, including semiconductor vendors, evaluation board manufacturers, and contract design companies. Therefore, there are a wide array of evaluation boards, and it is necessary to select the one that best suits your technical level and purpose. The following six are representative manufacturers:
- HiTech Global
- BittWare
- TUL
- IOxOS
- Portwell Japan
- ANVENT
2. Market
According to an April 2020 report by Global Information, Inc., the FPGA market is projected to reach US$8.6 billion by 2025, a significant increase from the US$5.9 billion recorded in 2020. This growth is anticipated to be driven by a compound annual growth rate (CAGR) of 7.6%. While specific figures for each technology node are not provided, it appears that in 2019, the majority of FPGA products fell below the 28 nm technology node in terms of market share.
Furthermore, the forecast indicates that the market for sub-28nm products will continue to show high growth in 2025 due to the emergence of low-power products, etc. The applications that will drive the FPGA market from 2020 to 2025 include high-performance computers for cloud computing and 5G networking. The applications driving the FPGA market from 2020 to 2025 include high-performance computing for cloud computing and 5G networking.