Lattice Blog

Share:

Different Types of FPGAs

Bob O’Donnell: Different Types of FPGAs
Posted 02/26/2020 by Bob O’Donnell

Posted in

As discussed in the first part of this blog series, An FPGA Primer, FPGAs (field programmable gate arrays) are a flexible type of semiconductor chip architecture that incorporate the ability to be changed or updated once they have been designed into products or even put into the field (hence the name).

But, not all FPGAs are the same. Some FPGAs are designed for high-performance, data-intensive workloads in places like cloud data centers and can require as much as several hundred watts to operate. Others feature lightweight, low-power designs and can draw as little as one milliwatt (1/1000th of a watt). Still others sit somewhere in between and can run on just one watt of power across a wide array of applications. While they all share the same fundamental characteristic of reprogrammability, FPGAs are inherently flexible enough to be adapted into an enormous variety of different environments.

In cloud data centers, for example, Microsoft has been using high-power FPGAs to accelerate searches on its Bing search engine. In that application, the FPGAs work alongside CPUs to accelerate certain portions of the search algorithms that Microsoft has developed for Bing. The result has been not only a speed up in performance, but a big reduction in power consumption because of the efficiency of FPGAs versus CPUs or GPUs in performing some of those tasks.

Within the same types of servers hosting the high-power FPGAs (made by companies like Intel and Xilinx), you can also find tiny, low-power FPGAs (like those made by Lattice Semiconductor). These specialized, low-power versions are optimized to run a few specific functions within a device, like system control, power management, and/or ensuring system firmware is secure. While the job may not be as glamorous, these tiny FPGAs play a critical role in the functioning and security of both cloud and enterprise data centers.

Even more interesting are larger sized, but still low-power FPGAs that efficiently perform cutting edge applications like computer vision, or other types of AI inferencing. From consumer applications such as drones, home security cameras and wearables, to industrial applications including predictive maintenance, motor control and machine vision, modest-sized FPGAs are increasingly being tasked with doing unique tasks, such as running artificial intelligence-based software algorithms on low-power Edge computing devices.

For many years, FPGAs have also been playing a hugely important, though little-known role in telecommunications infrastructure equipment that powers cellular networks. Given the ongoing rollout of 5G networks and the enormous amount of attention they are receiving, awareness of network infrastructure, and FPGAs that help power it, is bound to rise. The flexibility, time-to-market benefits, programmability and parallel processing capabilities that FPGAs offer are particularly well-suited to the rapidly evolving world of 5G network deployments. Telecom carriers around the world are scrambling not only to install some the new type of infrastructure equipment that 5G millimeter wave services require and that FPGAs can power (such as small or “pico” cells), but also to update their existing infrastructure. Complicated new functions such as Dynamic Spectrum Sharing, Carrier Aggregation and more that can be used to help carriers update and modernize their existing 4G infrastructure to make them 5G compatible via software updates are only possible because of the flexibility of FPGAs.

But FPGAs aren’t only flexible because of their internal structure, they can also match well with certain hardwired capabilities, particularly connecting to a variety of different signal inputs, and the combination offers some intriguing opportunities. In the automotive market, for example, FPGAs are starting to take on new roles as a tool for “fusing” the feeds from multiple sensors, such as cameras, lidar and more—in addition to roles that they’ve played for things like car infotainment systems. This sensor fusion functionality, sometimes also called bridging, is an absolutely essential capability for assisted and autonomous driving features in connected cars. In addition, there are additional opportunities in electric cars for controlling and protecting certain aspects of the electric motors that power them.

Most all of these new functions demand smaller, lower-power chips that are able to function in a variety of different environments. Not all of them may be as well known as the higher-profile, higher power drawing FPGA applications in the enterprise, but collectively they represent a huge opportunity that’s ideally suited to the rapidly changing, highly connected world around us.

In the next blog, I’ll take a look at a key application for low power FPGAs: embedded vision.

Bob O’Donnell is the president and chief analyst of TECHnalysis Research, LLC a market research firm that provides strategic consulting and market research services to the technology industry and professional financial community. You can follow him on Twitter @bobodtech.

Share:

Like most websites, we use cookies and similar technologies to enhance your user experience. We also allow third parties to place cookies on our website. By continuing to use this website you consent to the use of cookies as described in our Cookie Policy.