For decades, semiconductor and IT analysts have debated whether the end of Moore’s Law might be nigh. But recent investments and advances in the field-programmable gate array (FPGA) suggest that the debate may be missing the point. IT developers now see in the programmable chipset an alternate route toward ever-greater performance.
Recent moves by Microsoft, Intel, and the Defense Advanced Research Projects Agency (DARPA) suggest that many IT strategists now see FPGAs as a key factor in their business strategies and future growth. Indeed, the FPGA market is projected to grow to $10 billion by 2024, up from about $6 billion in 2015, according to Grand View Research.
According to Satwant Singh, senior director of strategic planning at Lattice Semiconductor, a Portland, Ore.-headquartered FPGA product and developer tools supplier, FPGA developments are focused now on two general areas. The first concerns servers in data centers and other high-performance computing architectures. The second encompasses the chipsets that support automotive and mobile applications—e.g., smartphones, notebooks, and the emerging Internet of Things (IoT)),
Although FPGAs have been around since the 1980s, they have grown logarithmically more useful since then. Over the last 30-plus years, FPGA capacity has increased by a factor of 10,000, performance has done the same by a factor of 100, and cost and energy per operation have decreased by a factor of 1,000, according to a 2015 IEEE overview.
Unlike conventional chip architectures, FPGAs are reprogrammable. This means that a designer or a customer can configure the array after it is printed, making it possible to tailor the chip so it focuses only on the pathways needed by the application.
“FPGAs are most commonly used for system design challenges which have no ‘standard’ solution from other semiconductor components in the market,” explained Singh. “FPGAs bridge incompatible interface formats, expand the number of inputs or outputs to a processor, or custom-control how you initialize and power your system.”
FPGAs are helping developers of both small and large systems keep up with their ever-rising demands. In mobile-related applications, Singh noted, the pace of innovation is accelerating, even as the complexities and costs of semiconductor development continue to increase and component design cycles keep getting longer. “In this environment, the ultra-low power and small size of FPGAs is becoming increasingly important,” he said.
Data centers are also finding FPGAs useful, because “it takes a long time and a lot of investment to design and tune dedicated components for emerging applications,” Singh continued. “Large FPGAs provide a convenient programmable hardware engine that enable rapid experimentation with overlay architectures to provide many orders of magnitude improvement.”
On the data center side, major industrial users of computing power are finding that FPGAs can act as useful components that help knit their data centers more closely together. Microsoft, for instance, has now rolled out what it calls a Configurable Cloud, which places a layer of FPGAs between network switches and servers all across its production data centers worldwide, providing a sharp boost in processing speed and system efficiency.
By enabling the FPGA nodes in the network to communicate with each other in a few microseconds, Microsoft engineers can link those arrays together in a way that allows them to act as a single computer—one focused on reducing latency and increasing the efficiency of resources used in these mammoth multi-continental operations. Microsoft engineers have said this “acceleration fabric” enabled its data centers to process data twice as fast as the company could before it rolled out the new architecture while using only10% more power.
At the same time, FPGAs are also being used to pack more computing power into tiny devices. “FPGA-based architecture will allow further distribution of algorithms into smaller form factors, with considerably less electrical power in the device mesh,” explained David W. Cearley, a vice president at Gartner, quoted in a Gartner research note on key 2016 technology trends, “thus allowing advanced machine-learning capabilities to be proliferated into the tiniest IoT endpoints, such as homes, cars, wristwatches, and even human beings.”
It’s this latter capacity that seems to have caught the eye of DARPA planners most. On Jan. 17, the agency announced that it plans to work with Flex Logix Technologies, helping to develop the latter’s EFLX embedded FPGA technology for use by any company or government agency designing integrated circuits for the government.
The three-year-old Mountain View, Calif., company’s EFLX system provides an IP core that can be used to build nearly 50 different arrays, thus meeting the needs of a wide range of applications.
The DARPA announcement was a signal to any government supplier that they have an opportunity to work with embedded FPGAs, according to Geoff Tate, president of Flex Logix. “Making the agreement public is the easiest way for DARPA to communicate this to all of the suppliers to the U.S. government,” Tate explained. “DARPA believes embedded FPGAs will become a widely used building block in government ICs.”
Already, Flex Logix executives say, developers have used Flex Logix’s embedded FPGA to:
- extend battery life in IoT applications by offloading monitoring algorithms from the less-power-efficient processor
- enable customers to customize chips by programming an EFLX-embedded FPGA rather than printing multiple variations
- control high-speed large arrays (500 MHz-plus) for wireless base stations.
Tate said having the option of using an embedded FPGA gives chip designers more flexibility. Embedded FPGAs allow developers to change the resistor-transistor logic (RTL) at any time after fabrication—even in-system.
“As every chip designer knows, having to change the RTL blocks at any point in the design process could easily cost multiple millions of dollars and add 3-6 months to the design schedule,” he noted. “With embedded FPGAs, this risk can now be eliminated, providing designers with a higher level of confidence when undertaking a new [system-on-a-chip or microcontroller] project.”
Other FPGA makers are making similar gains. For instance, a little more than a year after Intel completed its $16.7-billion purchase of FPGA maker Altera, the company has introduced its first new FPGA, the Intel Stratix 10. This offering boasts twice the core performance the previous generation array (the Stratix V) while using up to 70% less power, according to an Intel whitepaper.