Programmable Notes

Where Does the Fitter Deal Fit? The new AMD-Minc agreement for Minc to make fitters for all AMD's MACH line of programmable devices looks good for both sides. Given that, what does it really mean? by Larry Waller

On the surface, the new all-encompassing design software development link between Advanced Micro Devices (AMD, Sunnyvale, CA) and Minc (Colorado Springs, CO) looks like one of those too-rare deals that deliver nothing but upside for both parties.

The two firms themselves surely regard it that way, as reflected by details revealed enthusiastically in March and by their subsequent glowing statements. The ensuing comments by some industry peers, however, underscores once again that no event is taken at face value; the unannounced nuances are at least as important to most observers as the open facts. And they probe incessantly along the word-of-mouth network to discern any underlying meanings.

In essence, the agreement is a simple and straightforward one. From now on, Minc will develop and support all the design software for AMD's entire line of MACH programmable devices. While the deal includes all logic compilation tools, what most sparks programmable peer interest is the part that provides for Minc to develop and enhance the functional "fitters."

These critical software tools must be written specifically for each device family, then integrated into a comprehensive EDA design environment, to place and route new designs onto the chip. Because each new family requires a different fitter, the software-writing load is often backbreaking for silicon companies. Minc is thus assuming important responsibilities by taking on the whole AMD fitter development assignment, from the earliest stages of architectural definition and design of MACH products.

It is no wonder, then, that the vice president of AMD's programmable logic division, Frank Barone, says that "this kind of agreement between a silicon vendor and a universal software tool supplier is unprecedented in the semiconductor industry."

Minc has an excellent record in fitter work, performing this task on a specific contract basis for numerous device companies, Cypress Semiconductor and Atmel (both in San Jose, CA) among them. It also supplies other universal software design tools for integration into EDA platforms for most leading software houses. Leveraging off Minc's design software strengths clearly frees AMD to concentrate on its own silicon design specialty, as Barone points out. It also reduces the heavy investment in internal software development engineering required as the pace of introducing new programmable architectures quickens.

This AMD-Minc announcement hits a collective competitive nerve for a predictable reason, though: it challenges a bedrock article of faith held by most programmable industry officials. They believe strongly almost to a person that these fitter tools are super critical in bringing out new products on a successful and timely basis. Experience has taught them that a misstep in fitter performance can doom an otherwise promising product family. Developing them properly is regarded as a tricky and demanding job that requires a significant investment of software man-hour resources. To be done right, it involves virtually continuous hands-on internal engineering support that closely intertwines both silicon and software designers, according to industry lOR

Accordingly, leading device firms mostly have been dead set against taking a chance on letting outsiders, no matter how skilled, get advance looks at device data, much less assume the entire software development responsibility. The reluctance of Xilinx (San Jose, CA) to make its new device parameters available in advance of introduction to NeoCAD (Boulder, CO) when this firm was perfecting its FPGA Foundry open design platform is one example of this attitude, sources say.

It is not surprising that device competitors couldn't wait to launch some barbed questions and comments about the AMD-Minc linkup. Some key words that stand out in most of their statements, it should be noted, deal with "control" and "commitment" in relation to AMD's side. In fact, AMD takes most of the verbal flak; Minc's part of the deal generally is seen as simply a sound business expansion that makes sense for a third-party software house.

"A fairly gutsy move, but a strange move," is how Xilinx's Scott Lewis sees it. As product marketing manager for the firm's EPLD division, he goes nose-to-nose with AMD's MACH line. What really puzzles him, and others, is AMD "relinquishing control of critical fitter software to an outside source." As a practical matter, Xilinx has found that software development can be even more difficult and expensive, thus in many ways more important than silicon work. So when "control [of fitters] goes to someone else, that says something about commitment, and is not a good situation," Lewis says. Customers in particular may end up somewhat confused, he adds. Lewis is one of the observers who speculates there must be more to the deal than yet announced.

San Jose-based Altera--which usually finds few things on which it can agree with arch-rival Xilinx --in this case is strongly in the same camp. "The downside for AMD," notes strategic planning manager Bob Beachler, "is some of its destiny is now out of control, in the hands of a third-party vendor."

There have been some earlier, less sweeping fitter development attempts along these lines that didn't prove workable, he points out. EDA supplier Data I/O (Redmond, WA) among others offered fitter development services, but found it was difficult to keep pace externally with swift device design changes.

Along with Xilinx , Altera contemplates no changes in the way it does software, maintaining about a 50/50 split in software-hardware engineering staff. "It's very expensive, but crucial. Long-term it's just not a viable option to rely on a small company (for software)." At Lattice Semiconductor (Hillsboro, OR) marketing manager Stan Kopec regards fitter development as an internal matter, as do his two bigger rivals. "It's a significant investment," he says, "but we'll just keep on doing it the same way."

Lattice has previously said that getting design software right has been the toughest task in establishing its ispLSI and pLSI family of complex devices.

For Minc's part, company president Don Van Dyken doesn't see much difference except for doing 100 percent of AMD's development instead of a portion. Minc already supports some 23 IC manufacturers and about 5,200 active device designs, of which AMD's is only a small part. "We're ready, willing and able to do a lot more," Van Dyken says. He predicts other programmable outfits will also jump on the MINC fitter bandwagon. It is no secret that one of the knottiest problems in programmable software is getting internally designed fitters to mesh smoothly with universal design platforms. To beat this problem, he says, "a number of [the device vendors] are saying, 'let's avoid the grief'."

The intense give-and-take aroused by the AMD-Minc pact does not exactly surprise officials at the larger EDA companies, who have a broader perspective on the issue. "It's part of the push and pull between semiconductor vendors and tool vendors as to who should do what," observes Steve Eichenlaub of Mentor Graphics (Wilsonville, OR). He says some believe that "silicon and tools are locked together, and others say silicon needs (outside) tool help to form a higher silicon dimension."

Eichenlaub gives kudos to Minc in particular for being able to walk this narrow line between the two camps, without offending either one. It is a diplomatic feat for Minc to maintain the "wall" that separates the projects of its very competitive chip-firm clients who must supply proprietary device data to Minc. This bodes well for the AMD-Minc alliance, in the view of the software official. Mentor's EDA design environment is one of those on which the Minc fitters will be integrated.

Most likely, more such controversies will be fanned as the issue resolves itself, most parties predict. In the meantime, the present lively dialogue is, in Eichenlaub's opinion, "a healthy one, since it illustrates the [EDA] evolutionary process."

Larry Waller is a contributing editor of Integrated System Design.

To voice an opinion on this or any Integrated System Design article, please e-mail your message to: michael@asic.com.