Logic devices (digital chips) can be roughly divided into two types: standard devices and custom chips. Generally, the more customized, the more advantages the performance (speed), integration (number of gates) and design freedom of logic devices have. Manufacturing-related development costs are higher, and the turnaround time from order to shipment is longer.
Among them, a type of logic device among the standard devices is called a programmable logic device (Programmable Logic Device, PLD). FPGA is a kind of PLD, which has a higher degree of freedom in design than previous PLDs, and has a structure similar to a gate array, so it is named FPGA.
The first FPGA product was first launched by Xilinx in 1985, and its manufacturing process has reached 7nm
FPGA chip principle and composition
FPGA consists of a programmable logic unit (Logic Cell, LC), an input and output unit (Input Output Block, IO) for signal interaction with the outside, and a switch wiring array (Switch Box, SB) connecting the first two elements. composed of parts.
The logic unit of the FPGA implements different circuit functions through the binary data stored in the data look-up table (LUT). LUT is essentially a static random access memory (SRAM), whose size is determined by the number of signals at the input end. Commonly used look-up table circuits are four-input look-up table (LUT4), five-input look-up table (LUT5) and six-input look-up table (LUT6). The more input terminals of the look-up table, the more complex the logic circuit can be realized, so the logic capacity is larger.
According to Boolean algebra theory, for a logic operation with n inputs, no matter what kind of gate operation, there are only 2n results at most. Therefore, if the corresponding results are stored in the storage unit in advance, it is equivalent to realizing the logic circuit function.
A general K-input lookup table consists of 2k SRAM cells and a 2k-input data selector. The input of the lookup table is the address signal of the memory table, and the output is the 1-bit data of the word selected by the address. The K-input lookup table can realize 22k logic functions. FPGA configures the content of the lookup table by programming files, so as to realize different logic functions under the same circuit.
FPGA chip programmable features
The biggest feature of FPGA is field programmability. After the CPU, GPU, DSP, Memory and various ASIC chips are manufactured, their functions are fixed, and users cannot modify their hardware functions. After the FPGA chip is manufactured, its functions are not fixed, and users can According to the actual needs, the designed circuit is used to configure the function of the FPGA chip through the special EDA software, so that the blank FPGA chip is converted into an integrated circuit chip with specific functions, and multiple different function configurations can be performed.
Flash memory, antifuse, and static memory are commonly used programmable technologies in modern FPGAs. FPGA realizes programmability by controlling the circuit structure through programmable switches. Such "programmable" switches are usually implemented using a variety of semiconductor technologies. FPGAs have historically used EPROM, EEPROM, flash memory, antifuse, and static memory (SRAM), among which flash memory, antifuse, and static memory are modern FPGAs. Commonly used programmable technology.
The principle of FPGA chip programmable features - static memory as a programmable switch
The static memory is composed of flip-flops communicated by two CMOS inverters and two pass-transistors (Pass-Transistor, PT). The static memory uses the bistable state (0 and 1) of the flip-flop to record data, and the writing of the data is carried out through the PT, and the PT uses an nMOS type transistor.
The static memory usually drives the word line according to the address signal, and the data is read through the PT. Therefore, the high potential between VDD and Vth output by the memory cell can be amplified by the sense amplifier and then output. But because the FPGA needs to read data all the time, in the FPGA the data is read directly from the flip-flop instead of through the PT.
Most FPGAs that use static memory as programmable switches use look-up tables (LUTs) in logic blocks, and use data selectors, etc. to switch wiring connections. What is stored in the memory of the lookup table is the truth table of the logical expression itself, which is composed of multi-bit static memory.
In addition, the selection signal controlling the connection of the data selector is also connected to the static memory. This kind of FPGA is generally called SRAM type FPGA, which is the current mainstream type.
FPGA high capacity brings high performance
The FPGA chip uses the programmable logic unit LC including LUT as the basic logic unit to design the circuit, and the CPU/GPU/ASIC and other chips use the dedicated data device library to design the circuit. To complete the same code, the FPGA requires more transistors, resulting in a lower main frequency. From the data comparison, there is a big gap between the main frequency of FPGA (<500MHz) and CPU and GPU (1-3Hz).
The parallel computing performance of the FPGA chip is improved by its capacity, and a high-capacity FPGA allows the deployment of more processing circuits, thus bringing higher processing performance. In terms of capacity, the number of LUTs, the number of DSPs, the number of RAMs, and the number of User IOs are important technical indicators, and the number of LUTs is the basic indicator of FPGA chip capacity. Secondly, manufacturing process, DSP operating frequency, dynamic power consumption, SerDes rate and DDR3/DDR4 rate are important technical indicators of FPGA chips.
Advanced technology contributes the most to FPGA capacity improvement
Since its birth, FPGA has followed the development route of the process, and has continuously launched new products using the most advanced process technology. This is mainly because more advanced processes can achieve lower power consumption, faster response speed, and more transistors per unit area. Make the chip performance better.
After 2005, the process gap between FPGA and ASIC has gradually widened. FPGA, like general-purpose processors, is updated every two years following the pace of process development. However, in the past ten years, except for some applications such as game consoles, most products of ASIC are still 130 -90nm process, the process used by FPGA is three or four generations ahead of ASIC.
Xilinx's FPGA products include Spartan series, Artix series, Kintex series, and Virtex series. The product processes are mainly 45nm, 28nm, 20nm, and 16nm. Comparing the capacity of different Xilinx products, it is not difficult to find that for the same series of products, the more advanced the technology, the higher the capacity of the corresponding product.
FPGA covers a wide range of downstream areas, with communication and industry as the main application markets
Communications and industry are the largest downstream markets, and the market share of automobiles keeps rising. From the perspective of the downstream application market, the communication and industrial market share ranks first and second in FPGA chips, and the communication market share is expected to continue to increase, while the industrial market shows a slight downward trend.
The global military and aerospace FPGA application market share is stable at around 15%. Automotive is the fastest growing downstream market, and Gartner expects its market share to increase from 5.9% in 2020 to 12.3% in 2026.
Consumer electronics is expected to be the downstream market with the smallest share of FPGA chips, mainly because although FPGA is suitable for the rapid iteration rhythm of consumer electronics due to its flexibility, in the consumer electronics market with significant scale effect, FPGA has obvious cost disadvantages compared with ASIC . Therefore, in terms of life cycle, the application of FPGA has certain limitations.
Communication is currently the largest downstream market for FPGAs, and will maintain a higher-than-average growth rate in the future
Communication is currently the largest application market for FPGAs, and with the continuous iteration of communication technology, the advantages of FPGA applications are gradually expanding, and the market demand will continue to maintain high growth. In the next few years, the compound annual growth rate of the domestic and foreign communication FPGA market will be higher than that of the FPGA as a whole, and the application share of the communication market will also increase steadily. According to Gartner data, the global communication FPGA chip market will reach US$3.23 billion in 2026, with a CAGR of 10.4% from 2020 to 2026.
In the intelligent upgrading of industrial manufacturing, FPGA may become an iterative means for the rapid development of the industry
In the industrial field that is accelerating transformation, FPGA has a tendency to replace ASIC. Industry 4.0 is the future blueprint for the manufacturing industry formulated by Germany and the industry, and it is also a guiding light for industrial transformation and upgrading. Most application scenarios of Industry 4.0 put forward new requirements for hardware acceleration of edge intelligence, requiring lower latency data processing capabilities. On the other hand, new devices need to be reprogrammable to cope with the evolution of standards or protocols. In addition, the establishment of a smart factory is inseparable from the application of big data processing and AI technology.
FPGA improves flexible development solutions, so it has a great tendency to replace ASIC, especially in application scenarios such as industrial communication, motor control, machine vision, edge computing, and industrial robots.
Status quo of FPGA enterprise layout
Xilinx and Intel dominate the market. Major global FPGA suppliers include overseas chip design companies such as Xilinx, Intel, Lattice and Microchip, among which Xilinx has the most obvious advantage. According to Gartner statistics, in terms of revenue, the market share of Xilinx, Intel, Lattice and Microchip will reach 51%, 29%, 7% and 6% respectively in 2021.
