Since Moore’s law was put forward, it has been accompanied by the semiconductor industry for more than half a century. This law reveals the rapid progress of information technology. It makes people believe that IC process technology can show linear development. Through advanced technology, IC products can continuously reduce costs and improve product performance. However, this year, such an idea may be broken, and there are continuous doubts about Moore’s law in the industry. First, lattice core announced to abandon the 7Nm FinFET project, and then Intel delayed the research process of 7Nm process. These actions highlight the new views of enterprises on the industry. Will Moore’s law really disappear in the next few years? Will it change the current industrial pattern? It’s hard to say, but the slowing pace is undeniable. The change of Moore’s law has brought a lot of uncertainty to the semiconductor industry, which also brings development opportunities to efga.

Efga ushers in a good opportunity for development

For the industry, efga is not a new technology. It has appeared more than ten years ago. It is an embedded FPGA technology. We are more familiar with FPGA technology. Both FPGA and ASIC chips are widely used. However, it seems that ASIC chips are the real mainstream of the market all the time. It is not the reason for FPGA technology, but their manufacturing costs and application scenarios. Relatively speaking, the cost of FPGA or efga is higher. Unless there are specific scenario requirements, customers are not willing to adopt more expensive technologies.

With the development of chip technology, ASIC chips also have development bottlenecks. The limitations of materials and architecture make it difficult for them to meet some scenarios, such as the computing power requirements brought by AI, machine learning applications and cloud computing. The progress of big data and the Internet of things has brought greater computing requirements. Efgas has once again become a hot pastry in the market. This new, heterogeneous The new technology with programmable hardware accelerator has become the key driving force to promote the semiconductor industry.

It is learned that at present, there are not many companies providing complete efgas technology solutions on the market. Achronix is one of the best. They have just been shortlisted for the 2018 most respected private semiconductor company award of the global semiconductor alliance. They have the ability to embed efgas technology into ASIC chips to achieve mass production. Achronix speedcore is the first efgas IP Product in the industry to ship to customers.

Advantages and applications of efga Technology

With the higher requirements of artificial intelligence and machine learning for big data processing ability, the increase of the number of processor cores can not bring the doubling of computing power, while the SOC of efga can not only process data faster, but also consume less power. However, it is not a simple thing to embed FPGA into SOC. We need not only FPGA experience, but also IP integration experience. Achronix was the first vendor to offer a high-density FPGA with embedded system level IP, so it finally launched the efga product using the same, proven technology.

“Achronix is pleased to be at the forefront of the embedded FPGA market. This technology is rapidly becoming the first choice for many applications that need hardware acceleration,” said Steve mensor, vice president of marketing of achronix. “These new efga accelerator application acceleration projects will enable innovative companies and research institutions to use our IP and tools to build the next generation of programmable chips to meet the growing data and computing needs of AI / ml and other computing intensive applications.”

Accelerator of artificial intelligence and machine learning

According to the application characteristics of artificial intelligence, efga technology provides a new solution. Using its advantages in custom processor and memory management technology, embedded programmable logic reduces the total area of the chip by eliminating the demand for I / O circuits with large area and high power consumption. Luo Weiliang, general manager of achronix Asia Pacific region, said that many FPGA architectures such as speedster of achronix provide a fully customizable mixed structure of logic and digital signal processing engine, which can support fixed and floating-point operations. DSP engine can adopt a construction module composed of 8-bit or 16 bit units. This method can combine them to support higher precision data types, By using the look-up table in the logical array to realize the related logic, it can also be suitable for low precision. Achronix’s speedcore can be integrated into the embedded FPGA Semiconductor Intellectual Property module on the chip.

In addition to the general requirements of computing and network infrastructure, AI / machine learning also has a significantly increased demand for high-density and targeted computing. MLP module is a highly flexible computing engine, which is closely coupled with memory, so as to provide a solution with the highest performance / power consumption ratio and the lowest cost for AI and machine learning applications.

Steve mensor said that compared with the previous achronix FPGA products, the new achronix machine learning processor takes advantage of the specific attributes of artificial intelligence / machine learning processing and improves the performance of these applications by 300%. This is achieved through a variety of architectural innovations that can simultaneously improve the performance and number of operations per clock cycle. He mentioned that the fourth generation speedcore efgas IP products launched by achronix for artificial intelligence, machine learning and network hardware acceleration applications, compared with the previous generation of speedcore efgas products, speedcore gen4 has a 60% faster performance, 50% lower power consumption and 65% smaller chip area; The new machine learning processor unit module provides 300% higher performance for AI / machine learning applications. These machine learning processors support various fixed-point and floating-point formats, including bfloat16, 16 bit, half precision, 24 bit and unit block floating-point. Users can balance precision and performance by selecting the best precision for their applications.

Efga technology is called the best hardware accelerator

Big data and network bring huge data processing requirements. Traditional multi-core CPU and SOC need programmable hardware accelerator to preprocess and unload data, so as to improve its computing performance. For low to medium capacity applications, independent FPGA chip is a convenient and practical solution. For high-capacity applications, efga is the best solution. Steve mensor said that to embed FPGA into SOC, we first need to solve the area problem of FPGA chip. The area ratio of standard FPGA core to programmable IO and controller is close to 1:1. Achronix’s speedcoree FPGA is directly connected to SOC, which can not only reduce the area of FPGA chip by half, enable FPGA to be embedded into SOC, but also reduce the size of CB, reduce the number of layers of PCB and improve signal integrity.

Steve mensor continued that speedcore is directly connected to SOC by internal connection, which eliminates the large programmable input-output buffer visible in external independent FPGA and reduces energy consumption. Speedcore eliminates the need for all supporting components around independent FPGA. These components include power regulator, clock generator, level shifter, passive components and FPGA cooling devices, and the cost will be reduced accordingly.

For FPGA developers, the importance of software tools is self-evident. Achronix provides customers with ACE design tools. Steve mensor said that achronix’s research class efgas accelerator application acceleration project and test chip class efgas accelerator application acceleration project will support research institutions and test chip developers to easily obtain the license of achronix’s speedcore efgas technology. The license includes docking the pre configured and streamed verified speedcore efgas IP and the company’s industry-class ace design tool.

Finally, Steve mensor mentioned that achronix has been able to provide speedcoreip products based on TSMC’s 16 nm finfetplus (16ff) process, and is developing IP products based on TSMC’s 7 nm process.

Summary:

With the development of machine learning and artificial intelligence, the market’s requirements for chip computing power are increasing day by day, and the bottleneck of manufacturers’ improving performance in architecture and process is gradually emerging. New FPGA technology can help them solve these problems. Achronix’s efga technology finds a balance between performance and power consumption, and solves the problems of enterprises. High performance computing will become the mainstream application. In the future, efga technology will shine in the market and its market share will gradually increase.

Leave a Reply

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