Feedback A forum for readers to speak their mind on issues that are important to the design community

ASICs vs. FPGAs redux

To the Editor:

In response to Bob Sandler's Viewpoint column ["The New 'ASIC Replacement' FPGAs," August, p. 72], I'd like to clear up a few points for readers interested in the rest of the story.

Sandler states that my own Viewpoint ["Cost or Performance--ASICs Still Beat FPGAs," March, p. 95] ignores today's most important factors: fast time to market and a high quality-to-cost ratio. There's no debate over the usefulness of FPGAs for quick-turn prototyping: once a design is simulated and laid out, FPGA protos can be programmed immediately. That's the basic benefit of an FPGA. It's also why FPGAs generally cost two to four times as much as an ASIC.

Once an FPGA has begun to work in the system, the customer typically switches to gate arrays to reduce costs and profit on production runs. Gate array production lead times range from 6 to 10 weeks, not 8 to 16 weeks. The average product life cycles when looking at all markets--communications, EDP, industrial, automotive, consumer--lie in the 24- to 36-month range. Only some of the EDP and consumer products suffer 12- to 18-month product life cycles. Furthermore, the argument that ASIC lead times create a 2- to 4-month sales delay is erroneous because most customers are smart enough to keep a minimum amount of FPGA inventory on hand until the ASIC is in production.

The reality is that many designers can easily prototype using an FPGA and concurrently convert the design into an ASIC, achieving the tight product development time needed to get to market quickly. That's particularly true because scan insertion and vector generation are becoming easier to deal with, now that "vectorless" services that automatically generate vectors are available from some ASIC suppliers.

Quality and performance also favor ASICs. Today's FPGA conversions still generally process gate arrays with two or three layers of metal for interconnect. Mask and wafer fab costs are considerably cheaper for an ASIC process than for FPGAs, since the programming is done at the wafer level. Most ASIC processes use 12 to 14 masking levels, whereas many FPGA processes require over 20. Therefore ASIC technology more readily achieves lower wafer cost with higher gate density than do FPGAs. The bottom line tells all: For any digital designs that go into production volume, gate arrays are still very competitive.

Vince Hopkin
Director of translation ASICs
American Microsystems, Inc.
Pocatello, Idaho

NT for the masses?

To the Editor:

Thank you for further confirmation that EDA purchase decisions are made with little influence from users of those tools. Not surprisingly, your editorial ["Who's Got the Power?" September, p. 6] mirrors the sentiments of Jerry Burwell expressed in the article "Windows NT vs. Unix: The Road to Abilene Is Paved with Customization" in Electronic Design, Oct. 13, 1997. However, by means of similar arguments, you arrive at different conclusions.

I think the problem is a matter of perceptions of accessibility. To engineers outside of workstation-based design departments, the PC represents access to the world of EDA. Some years ago, as companies changed, replacing secretaries, clerks, and typists with PC-based office tools, PC use grew exponentially. At last, engineers outside the mainstream of computer-based design finally enjoyed access to EDA tools on their own desktops. This access was in some cases in defiance of corporate design philosophy or policy. Certainly, authority was being challenged and, in some cases, domains were threatened.

If PCs had not proliferated, promising empowerment to such vast quantities of engineering professionals, the dominance of workstation-based EDA would continue. Microsoft would have had no base of users eagerly anticipating truly world-class access to EDA on something similar to their own PC. Microsoft took the formula of workstation-based EDA--"sell the tool at the highest possible corporate level"--and applied it in a new direction: They sold it at a high level to the EDA companies.

To survive, workstation EDA suppliers must now encourage mass adoption of Unix-like operating systems in order to provide hosts to propagate their tools. Probably the biggest surprise is that the promise of accessibility would be made on the basis of a cheap, reliable, self-supportable operating system that runs on "your PC."

Allan Goodman
Senior staff engineer
Raytheon Systems
Newport Beach, Calif.

PCs: The Sun also rises?

To the Editor:

I'm weighing in late in the Linux versus NT debate, but I feel the discussion to date has missed several elements of the bigger picture. I've watched with amusement the expression of strong biases, as I've always thought being objective was in the job description of every engineer.

I've worked in both the Unix and NT environments and have no doubt that the various Unix flavors have the edge in stability and flexibility. The debate is over the size of the edge and whether it's large enough to make NT completely unusable (as many have asserted). Today I run Model Technology and Exemplar tools under NT and routinely do 50,000-gate designs with overnight synthesis and placement and routing, as well as multiday simulations, without ever having a crash of any sort. I concede that million-gate designs may prove to be a different matter, but when stability problems do arise, are the problems properly isolated? Instability can result from deficiencies in any number of areas, from the application and operating system to the hardware platform or environment.

In the application space, I've used tool revisions from several of the big EDA vendors under Unix that would routinely core-dump. An app that's stable and bug-free under Unix may not be equally bug-free in its NT version.

The wild and woolly hardware side of PCs can also lead to stability issues. Many smaller corporate environments purchase their PCs from resellers who build or sell off-brand clones. The varying brands and revisions of motherboards, memory, and CPUs used can greatly undermine stability. Even the big names offer--within a single model--an ever-changing mix of motherboards, graphic cards, disk drives, memory, and power supplies.

Unix workstations generally see much more controlled design, assembly, and testing. At the same time, I've used Unix workstations that routinely generate a memory parity error once a month. Both platforms benefit greatly from proper maintenance and service. PC systems have a tendency to gather more service neglect than Unix workstations because of their lower initial cost, greater number, and greater disbursement.

With all those variables present in the PC platform, how can so many users assign stability problems to the OS? Running Linux on the same box that was previously running NT isn't conclusive; the app is a different version. Even when the answer may not be NT, why is Linux the answer? The same users apparently have been well served by their HP and Sun workstations, so when moving to a PC platform (by choice or not), why not run Solaris x86? EDA vendors may be more receptive to offering applications in that space rather than Linux.

Some proponents argue that Linux enjoys infinite resources for support from the newsgroups, but no one seems to take support as an issue with Solaris on his workstation. Of course, Solaris x86 costs more per seat than Linux or NT, and that's a central issue. It seems that to many people the main attraction to Linux is its low cost (free, with no maintenance). That mind-set doesn't captivate EDA vendors.

So it seems that the EDA PC platform landscape is three-pronged: NT, Linux, Solaris x86. As first noted, NT may be lagging a bit, but it will continue to get better. If NT can't meet your current needs, fine. But why Linux? If users are happy with Solaris on their Sun workstations, why isn't it a viable option on the PC? Any dearth of applications for Solaris is no different than for Linux, but Solaris is mainstream--thus capable of being loved by your IT department and your EDA vendor--and it integrates your PC seamlessly with your Sun workstations.

Gary Stein

Senior hardware engineer
Switched Network Technologies
Plymouth, Minn.

James Lee replies:
I hope you've had a chance to read my article, which agrees with many of the issues you raised regarding stability ["Linux Has What It Takes for EDA," September, p. 60]. As for Linux versus Solaris x86, I can only offer the following observations: Many Sun users are still back on SunOS and think the Solaris/SunOS issue is an obstacle. I saw early versions of Solaris x86 and don't know how much it has improved since then. I haven't seen Solaris x86 used outside of Unix administration classes and don't personally know of many applications that are available for it.

The positive things about Linux are that it works, it's stable, it installs easily, and under the gnu development environment most applications port and run quite well. I guess I shouldn't care which Unix-like OS is used on PC hardware if my EDA software runs as well as it does on traditional Unix platforms; I just think that there are perhaps more Linux boxes out there than Solaris x86 machines.

64 bits, but not yet--and not for Pentiums

To the Editor:

The "Caveats" section of the simulation article ["EDA Platform Benchmark: Simulation," March, p. 84] states that "Solaris is already a 64-bit operating system." It won't be one until Solaris 2.7, which isn't going to production until this fall. Also, the 64-bit Windows NT 5.0 OS can only help 64-bit hardware like Merced and Alpha.

Jim Shallman

Staff engineer
Mentor Graphics Corp.
Wilsonville, Ore.

To voice an opinion on this or any Integrated System Design article, please email your message to miker@isdmag.com.

integrated system design  November 1998