Reconfigurable Computing Reconfigures How We Compute

The era of reprogrammable hardware is upon us. And it’s quietly becoming big business. To keep you up to date on this hot high-tech movement, the Trends research staff has ferreted out the latest developments in this revolutionary field.

With traditional computing chips, the tedious process of design and fabrication can take months or even years. And once it’s done, that chip can never do anything other than what it was designed to do. Moreover, because it’s expected to perform so many different applications, it is wasteful in its use of power. Most people will never use much of its potential.

In essence, programmable hardware completely changes the game of chip design and manufacturing and further blurs the line between hardware and software, according to an article in the January 31, 2005, issue of Linux Journal.1 Programmable logic devices, known as PLDs, are essentially generic chips that are like a blank slate on which an engineer can write his own parameters, tailored to a specific application. If the application changes, the chip can be reprogrammed in a fraction of a second to perform the new functions.

Based on rewritable memory technologies, such as SRAM, they can be reprogrammed on the fly, even after they’re shipped and are in the customer’s hand.

When Intel shipped a flawed Pentium chip a few years ago, the company stumbled badly, first by trying to minimize the severity of the problem and then by having to replace millions of chips anyway. With PLDs, a company in such a fix might simply post a few lines of code on its Web site to fix the problem.

That’s one reason why programmable logic chips are poised right now to overtake traditional, or fixed logic, chips. The former make up a $3.5 billion business right now, while the market for fixed-logic chips is about four times that size. But PLDs are rapidly catching up. Their sales growth has outpaced that of fixed logic for several years, and the newest PLDs are now stealing market share from the traditional vendors.

In addition to their ability to fix problems, even after shipment, there are a number of other reasons for this trend toward reconfigurable computing:

First, PLDs cut time-to-market for new generations of existing products from months or even years to days or weeks.

Second, their ever-increasing power allows functions (that would previously require many traditional chips) on a single chip which would result in decreasing size, cost, and power consumption, even while increasing performance and reliability.

Third, in a matter of seconds or minutes, an engineer can configure or reconfigure the chip at a workstation or on the system assembly line. This unheard-of flexibility allows companies to react to last-minute design changes and to quickly prototype ideas before implementation.

There are two main types of PLDs, field programmable gate arrays ? or FPGAs ? and complex programmable logic devices, known as CPLDs. FPGAs have the best performance. They have the highest logic density and the most features.

The leading manufacturer of those devices is Xilinx, which promises to be the long-term winner in this game. Xilinx FPGAs are much more than simple microprocessors. They contain built-in hardwired processors, to be sure, such as the IBM Power PC chip, but they also have on-board memory, clock management systems, and they support the hottest device-to-device signaling technologies. FPGAs are used in a wide variety of applications ranging from data processing and storage, to instrumentation, telecommunications, and digital signal processing.

CPLDs carry thousands of logic gates, rather than the millions of logic gates the FPGAs have. However, they do have predictable timing characteristics that are ideal for control applications, and they use less power and they are cheap. That makes them perfect for small, battery-powered devices that are price-sensitive, such as cell phones and PDAs.

While traditional chip manufacturers are locked into the “end-game” of Moore’s Law, as they try to squeeze the last bits of speed and power from that fixed-logic technology, the performance of reconfigurable chips from Xilinx and others is surging ahead. Within a span of a few years, FPGA technology has gone from having only a few thousand logic gates running at 40 megahertz and costing more than $150 per chip to having millions of gates, running at 300 megahertz, and costing just a few dollars per chip.

To those familiar with two- and three-gigahertz fixed-logic chips, that may sound slow, but even at these clock speeds, today’s FPGAs are actually faster that today’s fixed-logic processors. The reason is that fixed-logic chips execute software instructions one at a time. On the other hand, FPGAs can be readily configured to run in parallel, doing many jobs at once. In a recent benchmark of 300-megahertz FPGAs against Intel’s 2.8-gigahertz Xenon chip, the programmable logic won hands down, according to Linux Journal. At the same time, their added memory and other functionality put them in a class apart from traditional chips.

Perhaps most important is the growing library of so-called “intellectual property” available for them. These are software bundles that can instantly reprogram the chip to perform the functions of a wide range of traditional chips, such as complex digital signal processors, memory controllers, bus interfaces, and fully-functioning microprocessors. It could take months for engineers to design the same functions in fixed logic, so being able to download and incorporate these pre-designed functional designs speeds up the development process tremendously.

Another huge advantage of PLDs is that Xilinx and other manufacturers can potentially keep them on the shelf for years to serve the needs of a wide range of customers, without worrying about new models making them obsolete. The new chip model comes into being when engineers design a new piece of software. For manufacturers using their chips in products for end-users, there are no worries about running short of chips or holding big inventories that will suddenly become obsolete. Since these manufacturers can use the same PLD version for years, they can order what they need, when they need it.

Xilinx has also developed a new business model for the chip industry, which the Trends editors believe is going to change the way companies like Intel function. Xilinx has no foundries of its own for chips. It farms out that work to IBM and others, who are already in the business of making chips. This means Xilinx can concentrate on its most important contribution: New chip architectures, software for those chips, and other intellectual property.

The implications of these developments in electronics design are far-reaching. For one thing, programmable logic scares the pants off Microsoft and Intel ? or at least, it should. With IBM out of the PC game, those are the two leading companies that benefit from the clear split between software and hardware. But PLDs promise to erase that distinction, not to mention changing the way the game is played. In light of these trends, we offer the following six forecasts for your consideration:

First, within the next two years, PLDs will continue their spread through the telecommunications infrastructure as the ever-increasing need for cheap bandwidth makes them the technology of choice for phone companies, cable operators, and satellite systems. They will replace existing chips and come to dominate all the switching, routing, and processing functions. On-the-fly upgrades will accelerate the rate at which those services improve. Those companies that can gain a foothold in this market will be poised for the next stage.

Second, in the next five years, programmable chips will explode into the mainstream, as they replace fixed-logic sets in every kind of consumer device in existence. Devices like laptop computers and iPods will receive “hardware upgrades” the way they now receive software upgrades ? via small packages of software sent over the Internet.

Third, during this timeframe and throughout the next 10-year window, the nature of the time-to-market battle will change dramatically. Instead of rushing to get the next super-chip developed, companies such as Intel will gradually have to morph into suppliers of PLDs to companies like Xilinx and Altera, the other main competitor. Microsoft’s business model will have to shift dramatically or else face marginalization. The big time-to-market push will be for the PLD makers to get the latest software and intellectual property on line. Chip foundries will churn out neutral platforms.

Fourth, consumers will enjoy a new generation of devices, from cell phones to cameras to pacemakers and even cars that are constantly being altered or upgraded. While the era of electronics junk won’t completely come to an end, product lives will be extended. Consumers will no longer have to make decisions about what functions they want in a phone, for example. They’ll simply buy a platform for a cell phone and download the functions ? camera, PDA, GPS, and so on ? as they need them, then discard them when a better version comes along.

Fifth, manufacturers of consumer electronics will increasingly have to face what the auto industry already faces: Their products have a much longer life, and people won’t need to throw them away every two years. That will change the evolutionary trajectory that the industry is on, and require sweeping adjustments to the way business is conducted.

Sixth, one of the leading PLD vendors of today, probably Xilinx or Altera, may well become the next Intel, dominating the business of providing combined software-hardware solutions to every sector of the electronics marketplace. Xilinx has already penetrated deep into the wireless infrastructure,2 and its chips are being used in everything from Gibson’s new electronic guitars to NASA’s Mars Rover. Even though its stock has taken a beating since the tech meltdown, Xilinx could be a huge winner.

References List :
1. Linux Journal, January 2005, “Application Defined Processors,” by Dan Poznanovic. ⓒ Copyright 2005 by SSC Publishing. All rights reserved.2. Nikkei Electronics Asia, November 2004, “FPGA-Centric Approach in Tomorrows Wireless Systems,” by Narinder Lall. ⓒ Copyright 2004 by Nikkei Business Publications Asia Ltd. All rights reserved.