Globally, the main market share of FPGA industry has been firmly grasped by Xilinx and Intel for many years. With the accumulation of time, the two companies have gradually built a “moat” composed of tens of thousands of core technology patents, which has also successfully prevented the further development of many FPGA companies. Some surviving companies, such as lattice and MICROSEMI, often can only rely on the sword to take the edge, specializing in special application fields such as low power consumption and aerospace.

However, at the end of May this year, an American company called achronix released an FPGA product called speedster7t, focusing on high-speed network transmission, machine learning acceleration and other fields. These application scenarios not only belong to the traditional advantage projects of FPGA, but also include the new applications of FPGA that are booming in recent years. Obviously, achronix is going to take advantage of this product to face off with Xilinx and Intel and other giant FPGA companies.

It has been 15 years since the company was founded. In these years, achronix has gradually developed from scratch and from small to large, and today it has the confidence to compete with the giant FPGA company. In the following article, Lao Shi will specifically analyze the performance characteristics of achronix’s latest FPGA and what enlightenment its development process has for our domestic FPGA industry.

Who’s achronix

Achronix was founded in 2004 and headquartered in Santa Clara, Silicon Valley, USA. It has an R & D center in India and a branch in Shenzhen. Although it has been 15 years since its establishment, achronix is still a private company. It is speculated that its total financing amount is about US $180 million to US $200 million.

It is worth noting that achronix was profitable in the first quarter of 2017, and its quarterly revenue jumped from $5 million in the fourth quarter of 2016 to $40 million a year later, becoming one of the fastest-growing semiconductor companies in the world at that time.

Achronix’s rapid development in recent years benefits from the continuous and wide application of FPGA technology on the one hand, and its own technological progress on the other hand. Interestingly, as a fabless chip company, achronix was OEM by Intel and released the world’s first FPGA based on FinFET technology, even earlier than Xilinx and Altera, the second largest FPGA manufacturer later acquired by Intel. Achronix’s FPGA is also the first FPGA manufactured by Intel.

In addition to “traditional” FPGA chip products, achronix is also the main IP provider of the main embedded FPGA (efga), as shown in the figure below. As mentioned in previous articles, some large Internet companies, including Facebook, are believed to be major or potential customers of achronix’s efgas IP.

Characteristics of achronix's latest FPGA and Its Enlightenment to domestic FPGA

Compared with the traditional FPGA, the main feature of efga is that it can integrate the programmable logic unit and other ASIC IP into the same SOC, reducing the chip area and reducing the system power consumption, as shown in the figure below. For more information about efgas, interested readers can check Lao Shi’s previous article “what is efgas”, which will not be repeated in this article.

Characteristics of achronix's latest FPGA and Its Enlightenment to domestic FPGA

Speedster7t: new FPGA benchmarking agilex and ACAP

The speedster7t series FPGA newly released at the end of May this year should be regarded as the technology integrator of achronix, which also gives achronix the confidence and courage to challenge Xilinx and Intel. In fact, this FPGA does have many dazzling technical features that deserve our attention.

Characteristics of achronix's latest FPGA and Its Enlightenment to domestic FPGA

In terms of process, speedster7t will be manufactured based on TSMC’s 7-nm process, which is the same manufacturing process as Xilinx’s “ACAP” chip. Since TSMC’s 7Nm process is the same generation as Intel’s 10nm process, speedster7t is not inferior to Intel’s flagship FPGA “agilex” in the process.

In terms of chip architecture, one of the biggest features of achronix’s FPGA is the adoption of 2D on-chip network technology all over the chip, which coincides with Xilinx’s ACAP again.

Network on chip (NOC) is an important technology rising gradually to deal with the bandwidth bottleneck, which has been mentioned many times in previous articles. Especially in high-speed network applications, such as 5g, data center network and other fields, FPGA line speed has been required to process Ethernet traffic up to 400gbps. If the traditional design method is used, it needs a 1024 bit wide data bus and runs at the clock frequency of 724mhz to meet the bandwidth requirements. In addition, programmable logic units often need to transmit data from different locations of the chip, including on-chip and off-chip memory, as well as specific hardware acceleration modules, such as machine learning accelerator and so on. In order to transmit and transport a large amount of data on chip, a new NOC technology must be adopted.

The speedster7t adopts the horizontal and vertical 2D on-chip network, and uses the 256 bit wide Axi bus to operate at the frequency of up to 2GHz, so as to provide a single 512gbps and a total of 20tbps on-chip network bandwidth.

In addition to meeting the bandwidth requirements, the on-chip logic unit can be divided into relatively coarse-grained areas by using the on-chip network, which can greatly reduce the complexity of the layout and routing algorithm and improve the speed of timing convergence. Lao Shi believes that network on chip technology will be an important direction for the development of FPGA Architecture in the future.

In addition, speedster7t integrates multiple machine learning processors (MLP), which can support the operation of different fixed-point numbers and floating-point numbers, as shown in the figure below. In addition, this FPGA can integrate up to 300MB memory units on the chip to meet the implementation of artificial intelligence and machine learning related applications. Although the capacity of on-chip memory is still a certain gap from that of Xilinx ultrascale + series up to 500MB, it is almost the same as that of Intel stratix10 series FPGA.

Characteristics of achronix's latest FPGA and Its Enlightenment to domestic FPGA

Characteristics of achronix's latest FPGA and Its Enlightenment to domestic FPGA

In addition to on-chip network and machine learning processor, speedster7t also integrates many high-speed interfaces, such as PCIe gen5, gddr6, 112gbps SerDes and so on, which will not be described in detail in this paper. For more technical information and materials on this FPGA, please check in the “old stone talk core – advanced version” of the knowledge planet.

Important enlightenment of achronix’s development process to domestic FPGA

Achronix has developed for 15 years, from scratch and from small to large. Until today, it can compete with the two FPGA giants, which has a lot of enlightenment and experience worthy of our domestic FPGA companies.

First of all, there is no shortcut to the development of technology. Don’t be superstitious about “overtaking on Curves”. Achronix’s technology accumulation is not achieved overnight, but requires continuous precipitation and innovation for more than ten years. Especially for the FPGA industry with extremely high patent barriers, whether it is chip architecture, key IP, high-speed interface and other chip hardware design, or FPGA design tools and ecosystem construction, it is unlikely to be realized through “corner overtaking” again and again. Only by continuously accumulating and accelerating on the right track can we compete with those major players on the mainstream stage one day, like achronix.

Second, embrace new technologies and new directions. As mentioned above, many new technologies such as on-chip network technology and machine learning processor are adopted on the new generation FPGA of achronix. These technologies are only used in the most high-end FPGA of Xilinx and Intel. This shows the importance of FPGA companies embracing and using the latest technology. The development of FPGA has always followed Moore’s law, and these new technologies and new directions also make FPGA continue Moore’s law.

Third, leave the “comfort zone” and have the courage to fight. Unlike other FPGA companies such as lattice and MICROSEMI, achronix chooses to be hard and hard with traditional FPGA giants in fields such as artificial intelligence and high-speed network processing. Its confidence and courage are really admirable. For China’s FPGA companies, they should also strive to go out of their traditional business fields and launch an impact on the emerging FPGA application fields.

epilogue

A small step makes a thousand miles. Today, when people are impetuous, technology is empty, and everyone makes cores, but he doesn’t know what the cores are, Lao Shi believes that the company’s more than ten years of persistence and struggle in this article should give us some enlightenment.

Tagged:

Leave a Reply

Your email address will not be published. Required fields are marked *