The ASIC market is not going to go away anytime soon, however. Custom products will continue to meet the needs of customers in many ways and will continue to offer cost-effective strategies for those looking for point solutions. The PLD vendors are not trying to take on the ASICs on that front, according to Patel. But, in the increasingly frantic market within which semiconductor products live and die, PLDs are emerging as an excellent solution for those who can't "see" out there-two or three years in the future-when the latest, greatest products will be in demand. In the current market, "You just don't have the luxury of casting an ASIC solution. You either miss the market window, or you've got to have a very, very clear crystal ball," adds Patel.

The reasons are fairly straightforward. The on-chip device sizes are shrinking and the metal layer count increasing. All of this only helps to push the popularity of the PLD/FPGA products in the market, says Patel, because of the increased ability to program (and reprogram) complex circuitry on-chip. "Connectivity trends are in favor of FPGAs. All of the gates can be easily connected up with lots of fast wires, the results being that implementation wins and performance wins."

Is bigger better?

Patel points out that ASICs are growing in size as well. Ten years ago, the ballpark figure was 10,000 gates per square inch; today, that figure is closer to an average of 1 million gates per square inch. ASICs can be made more dense, but they are distinctly "pad limited." You've got lots of devices on-chip, but there are only a finite number of pins with which to access the innards. "It's one thing to pack in more transistors and more layers of metal. It's another thing, entirely, to connect them up for something useful," he adds. On top of that, you may be able to produce a huge (ASIC) product chip, but even if you only etch five die onto a wafer, you've only got one type of die. There is no flexibility there.

All of these limitations obviously evaporate when you take a look at the FPGA/PLD option. The number of metal layers now available work to the advantage of these products. The design capacity is fully utilized when the on-chip devices are programmable-either in the factory or out in the field. For FPGAs, Patel says, half in jest, "Even were we to get up to 5 trillion transistors," the full-chip would be available for design on a PLD.

And there's more good news coming for the fans of PLDs. There's all of the advantages that they bring to system-level design because of the wealth of on-chip system I/O features potentially available-particularly important in this era where so many design starts are destined to land in the network infrastructure. The data bandwidth demands of that infrastructure are enormous and, according to many, a perfect match to the growing capabilities of emerging product lines in the PLD category.

Additionally, customers of the PLD vendors are pounding at their door with demands for a product that will help them meet their very real time-to-market demands that overwhelm designers today. So, where as in the past, PLDs were the princes of prototyping, now they are providing enough essential flexibility and adaptability as to become proud and crucial products in their own right.

And, finally, there's the newest buzz in development paradigms-platform-based design. PLD vendors see a whole new world of opportunity opening up here as the mentality around this newest in design strategies goes mainline. As Patel puts it, "A microprocessor is always a microprocessor. It's not called logic or a state machine. Similarly, memory is always memory." He argues that, although currently, programmable devices travel with a variety of monikers-including FPGA, PLD, and so forth-that's changing. Just as microprocessors and memory are unambiguous terms, "FPGAs will be platforms." They will sport enough embedded microprocessors, DSPs, connectivity capacity, and switching fabric to fully function as generic platforms upon which design differentiation can be parked.

Webster beware

Like so many realms within technology, the semantics can be very confusing. What, in fact, is the difference between a programmable, a reprogrammable, and reconfigurable device? Xilinx's Patel has a straightforward answer. "Programmables are either one-time programmable, or programmable in the field or factory multiple times. However, a reconfigurable, while it's programmable, also has other capabilities. It has the capability to be reprogrammed while within a functioning system. This is what gives the reconfigurable it's value and potential to be a true 'killer app.' Even though it still remains to be proven, the industry is moving to remove that doubt," he says. In other words, in its full, functional glory, a reconfigurable could have its personality reprogrammed on the fly without having to fully shut down the system within which it resides. This is a far cry from a programmable, which will always require the system to grind to a complete halt before it can be re-programmed to meet a newly discovered need.

A different spin

Ralph Zak, vice president of marketing for Adaptive Silicon, puts a slightly different spin on the lexicon and definitions. Zak says that a programmable chip is one which can be progammed to perform different functions. A reprogrammable is one that can be programmed remotely to perform different functions. A reconfigurable chip is a design concept-typically viewed as a reprogrammable design that can be reconfigured in real time. Based on changes in the environment, the reconfigurable, therefore, has the ability to perform different or multiple functions at different points in time. Zak summarizes by saying that the reconfigurables are a subset of the reprogrammables which are a subset of the programmables.

According to Zak, it's important to note that some FPGAs are reprogrammable, and some are not. Therefore, these distinctions, for the time being, are actually quite important-especially if you're a customer trying to sort out the various product offerings on the market today. It's probably a safe bet to assume that an FPGA, once programmed, is set for life unless you are specifically told otherwise.

Buried in the minutia here is a profound message about where we are headed in the next era of chip design. Per Zak, the two ends of the technology spectrum-all custom or all FPGA-are coming together quickly to establish a versatile, highly useful chip, chock- full of programmable logic and other key components, which will usher in the ultimate product for our time-the system chip that can be all things to all people.

Click here for this month's focus tables.

To voice an opinion on this or any other article in Integrated System Design, please e-mail your comments to sdean@cmp.comd