FPGAs in the 2020s – The New Old Thing

Authored By:
Bob SIller
Director, Product Marketing

Posted On: Jan 09, 2020

FPGAs are the new old thing in semiconductors today. Even though FPGAs are 35 years old, the next decade represents a growth opportunity that hasn’t been seen since the early 1990s. Why is this happening now?

There continues to be a data explosion in the world, with IDC predicting over 175 zetabytes of data will be generated annually by 2025. With this much data, there is a tremendous opportunity to analyze it for insights that can change and influence the world. AI will play a huge role in this data mining operation, and companies are growing their workforce with deep skills in machine learning and data analytics to meet the challenges of the future.

From a workforce perspective, the 2020 emerging jobs report from LinkedIn lists the #1, #3 and #8 fasting growing jobs as AI specialists, data scientists, and data engineers. This trend is not surprising given the value that can be generated from data. The only way to unleash this data potential is through AI-driven insights and having people skilled in using technology to analyze the data.

The growth in these jobs depends on the hardware technology gains that only semiconductor companies can achieve. It's generally agreed that CPU-based solutions will not be able to keep up with the growth in data, which means that there is a need for hardware-based acceleration solutions to supercharge servers. FPGAs are the clear accelerator of choice with Amazon, Microsoft, Baidu and Huawei all offering or using FPGAs as custom accelerators. FPGAs are now the mainstream accelerator used across a wide variety of data acceleration applications, and unlike traditional GPU techniques, FPGAs can accelerate diverse workloads in addition to AI. Some examples are network function virtualization and high-performance compute.

This year, Achronix will ship its new, high-end, 7nm Speedster7t FPGAs that will be evaluated alongside high-performance FPGAs from Xilinx (also on TSMC 7nm) and Intel (who uses their own 10nm process technology). We have seen tremendous customer excitement about our new architecture, and now the industry sees Achronix on a level playing field, delivering much needed competition to the incumbents. Speedster7t FPGAs include a new network-on-chip, or NoC, capable of delivering over 20 Tbps bandwidth and provides many design benefits compared to traditional FPGAs. Additionally, our 112 Gbps SerDes, GDDR6 memory interfaces and support for PCIe Gen 5 all lead the industry.

In addition, today's FPGA's capabilities are now available in a variety of form-factors that have not previously been supported. Traditionally, data centers use FPGAs as sidecar accelerators to a CPU. Now companies, such as Microsoft, are using FPGAs in the data path where high-speed Ethernet connects directly to the FPGA instead of the CPU. However, both approaches require an additional standalone FPGA device, which increases the solution form-factor and are, therefore, not necessarily suited for all deployment models. The emergence of embedded FPGA (eFPGA) allows a region of programmable logic to be integrated into an ASIC/SoC, providing flexible acceleration but with a minimal increase in cost and form-factor size.

Furthermore, eFPGAs reduce latency, improve memory access and make communications between devices much more power efficient. Our Speedcore eFPGA technology offers the same proven look-up-tables (LUTs), memory, DSP and machine learning processor IP as in our Speedster7t FPGAs for customers to design their own devices. Alternatively, certain applications favor a non-monolithic approach and are using FPGA chiplets that act as an in-package accelerator, as is being developed by the Open Compute Project. In this case, Achronix offers the same high-performance FPGA core fabric as chiplets built with Speedcore eFPGA IP.

These different FPGA form-factors enable exciting new deployment models, for example, where data analytics are performed closer to the network edge, which could be co-located with sensors or within a mobile edge compute implementation. Achronix is unique in the industry in supporting a full portfolio across high-performance standalone FPGAs, embedded FPGA technology and FPGA chiplets, as well as the ability to migrate designs between these different deployment models. This range of options is not available from any other high-end FPGA supplier and allows for new innovation in data acceleration applications.

The next decade holds great promise for a resurgence in FPGAs as data continues to grow at an exponential rate, and Achronix is uniquely positioned to lead the way with our innovative FPGA and eFPGA architecture optimized for data acceleration workloads.