Xilinx has larger capacity FPGA chips. Alibaba Dharma Academy also published FPGA chip technology for AI voice. So is there anything special about FPGA chips? What does the hot 5G have to do with the AI ​​field?

According to media reports, the world‘s largest programmable chip (FPGA) manufacturer signed the official release of the largest capacity FPGA chip to date, the Virtex UltraScale+VU 19 P.

VU 19 P adopts 16nm process and is based on Arm architecture. It has 35 billion transistors, has the highest logic density in history, and has a large number of I/Os (9 million system logic units, 2072 user I/O interfaces). The simulation and prototyping of advanced AI chips, 5G chips, and automotive chips will be available in the fall of 2020.

On the 21st, Alibaba Dharma Academy published Ouroboros on FPGA chip technology for AI voice. It is a technology that improves the computational efficiency of speech generation algorithms by more than 100 times. Is there anything special about FPGAs? What‘s the outlook for that market? What other investment opportunities are there?

FPGA (Field Programmable Gate Array) is an integrated circuit with programmable characteristics pre-designed and implemented on a silicon chip, which can be configured in a specified circuit structure according to the needs of the designer, and can be configured on a chip chip without any dependence on chip customers. Because of the ASIC chips designed and manufactured by manufacturers, they are widely used in prototype verification, communication, and automotive electrical. electronics, industrial control, aerospace, data center and other fields.

For ASICs, FPGAs have three advantages:

(1) Flexibility: By programming the FPGA, the FPGA can perform any logical function that an ASIC can perform. The unique advantage of FPGAs is that the function of the chip can be changed at any time. This feature can reduce the cost and risk of the product at the stage of immature technology. This feature is especially important in the early days of 5G.

(2) During the release period: Because the FPGA can be directly programmed and used, the FPGA solution does not need to wait for the chip flow cycle of 3 months to 1 year, and obtains the release time of the product for the enterprise.

(3) Cost: FPGA and ASIC are mainly the latter with fixed costs, and there is almost no FPGA way. When the usage is small, the FPGA method does not need to pay a one-time $1 million chip cost, so it does not need to bear the risk of streaming failure. The cost of the FPGA approach is lower than that of an ASIC. With the increase in usage, the cost advantage of the FPGA method gradually shrinks. If the usage exceeds a certain amount, the ASIC method generates economies of scale through a large number of streaming transmissions, which becomes advantageous in terms of cost.

While the FPGA market is small, 5G and AI bring increased reliability.

According to WSTS data, the global integrated circuit market reached US$468.8 billion in 2018, and the world FPGA market in the same period was about US$6.3 billion, accounting for about 1.34% of the integrated circuit market. Although the market is small, it will benefit from the global layout of 5G infrastructure and the continuous development of AI technology in the future, and the increase in demand in the FPGA industry is reliable.

In the 5G era, the FPGA market faces an opportunity for price increases. Since FPGAs are mainly used for the baseband of transceivers, in the 5G era, as the number of channels increases, the computational complexity increases, and the scale of the FPGAs used increases. The pricing of FPGA is positively related to the resources on the chip, so the unit cost of FPGA in the future communication field will also rise. The unit price of FPGA in the current base station transceiver is usually several hundred yuan. The scope is expected to be further improved in the future. The main cost and power consumption of the transceiver is contributed by the baseband part, and the complexity of the future technology will again increase the cost of the transceiver and further lead to the increase in the price of the base station AAU.

The increase in FPGA supply for 5G comes from both sides. One is to increase the usage of FPGA components by increasing the number of communication base stations. In the early stage of 5G, the cycle ratio of base station laying has increased. On the one hand, due to the early signal attenuation of 5G, the demand for small base stations is huge. It is expected to exceed 10 million units in the next 10 years, which is a significant increase compared with the 4G period. Second, by increasing the usage of a single base station FPGA, the usage of the FPGA communication market will be increased as a whole. The high concurrent processing requirements of 5G Massive MIMO increase the FPGA usage of a single base station from 2 to 3 blocks in the 4G period to 4 to 5 blocks in the 5G period, promoting the overall usage of FPGAs.

At the same time, FPGA is widely used in the field of AI accelerator card based on its flexibility and high-speed computing power. According to Semico‘s survey, the FPGA market size in the AI ​​field is expected to reach US$5.2 billion in 2023. Compared with the current FPGA market of US$6.3 billion, the application in the AI ​​field should not be underestimated.