The applicability of programmable semiconductors is opening up, thanks to lower prices, more convenient form factors, and expanded capabilities. As the practicalities of the chips increase, so should the universe of product designers considering their use.
To what extent will these engineers switch from the conventional ASICs (application specific integrated circuits) to adopt programmable logic devices (PLDs)? Individual customers will still need to do their homework to make that call.
Field programmable gate array (FPGAs) chips, the most sophisticated type of PLDs, are already finding their way into markets not practical before, including plasma screen televisions, digital set top boxes, and automotive telematics systems. FPGAs are sometimes characterized as chameleon chips because the off-the-shelf devices can be programmed to handle a variety of duties. The chips can be reprogrammed during the product design process and even after deployment to meet changing performance requirements. They can provide differentiation within product lines. This flexibility encourages innovation, can slow product obsolescence, and lower development costs.
FPGAs found their calling in the 1990s among sectors where speed-to-market and operational flexibility could make or break a company. They were particularly welcome in telecommunications , where the devices help manage data transmission on high-end network equipment.
While the limited markets for such devices made the chips pricey, many customers found them worth every penny. Cellular operators, for example, found that the operation and maintenance advantages gained by the ability to modify equipment remotely with software programs provided substantial return on investment even if each application required dozens of the chips costing several thousands of dollars apiece.
Telecom is now stalled, but FPGA vendors claim readiness to serve markets that wouldn't consider the technology before. "They're branching out. It's not to say they haven't done that in the past. You have to look farther afield in different applications to generate new revenue," says Richard Wawrzyniak, senior ASIC and system-on-chip analyst at Semico Research Corp.
FPGAs also offer time-to-market advantages compared to ASICs, which are custom-built for each application. ASICs, on the other hand, can be dirt cheap on a per-unit basis due to high volume production, and they're generally faster, smaller, less power hungry than FPGAs. But ASIC customers rack up exorbitant up-front non-recurring engineering costs—particularly mask charges. FPGAs, made without a mask, don't put that burden on their customers.
BlueArc, a start-up that makes silicon servers, says it could not have designed its products without FPGAs, which it purchased from Altera. Geoff Barrall, executive VP and chief technology officer at BlueArc, says the ASIC development costs were prohibitive despite the fact that the FPGA approach required more chips, at higher per-unit cost than the ASIC alternatives. "For a company in our stage of growth, it doesn't make any sense to be shelling out big amounts of money," he says.
While FPGAs got their start in the 1980s to provide simple "glue logic" functions, today the devices often integrate some system-level functions such as memory, input/output tasks, and processors to achieve both performance and efficiency advantages. Suppliers are working to add on more functions, like digital signal processing.
Some ASIC suppliers themselves have also begun incorporating FPGA capabilities so that the once-unchangeable devices be reprogrammed. IBM, for example, recently licensed a Xilinx FPGA core and will embed the technology in forthcoming ASIC product lines. The move is an "excellent indication of the future importance" placed on high-end embedded FPGA technologies, says Jerry Worchel, a senior semiconductor analyst at In-Stat/MDR.
IBM foresees broad applicability for the products, from networking and IT segments to pervasive computing and consumer products. "This is something fundamentally that was driven by our customers," said Peter Gasperini, manager of IP sourcing and embedded FPGA.
While the respective FPGA and ASIC camps borrow from each other, FPGA makers believe they've already made gains on traditional ASIC advantages in per-unit cost, size, and speed. Both Xilinx and Altera will soon produce FPGAs using a next-generation 90-nm process on 300-mm wafers, keeping pace with the latest ASIC manufacturing trends. The processes will cut chip size by half or more while greatly increasing yields and performance and substantially cutting costs.
Still, not all companies are buying. Motorola, for example, says it does not use FPGAs in its cellular phones because of power consumption and cost. "Certain applications can't tolerate any increase in cost, no matter what benefits are brought to the table. But with lower price points you start looking at applications that are on the fence," Wawrzyniak adds.
As Robert Bielby, senior director of standard solutions marketing at Xilinx sees it, FPGAs will challenge ASICs and customers will switch if the alternative gives them quick results and comparable quality at an equivalent price.