EDA's cold war: real-life intrigue rocks sector's oldest, most vital technology - Simulated Program with Integrated Circuit Emphasis - Industry Trend or Event

Electronic News, Nov 4, 2002 by Gale Morrison

Mad Russians, stolen code, Newton-Rophson and Gaussian elimination methods and defecting developer armies. It sounds like the stuff of pulp fiction, but in the world of chip design engineering's oldest, hottest property, known as SPICE, this couldn't be more real.

So vital is this $100 million software business, in fact, that the vastly larger $200 billion microelectronics market wouldn't survive without it.

There's not one story here, but 10. At turns they are ironic and contradictory. The Simulated Program with Integrated Circuit Emphasis (SPICE) was born at the University of California at Berkeley in the 1960s and remains a pillar of the university's electrical engineering and computer science department to this day. UC-Berkeley continues to develop and hone SPICE models, and they are free to download for noncommercial use.

Things change dramatically when that code makes its way into EDA chip design tools. The world's largest and most successful semiconductor companies' internal teams have used those UC-Berkeley VSIM models and spin versions of software for real engineering schedule usefulness. As a result, SPICE has become the lifeblood of the most leading edge, billion-transistor designs at an annual cost of $100,000 per seat.

According to the 2001 Gartner Dataquest numbers--ripped from the ticker last week and here in print for the first time--Nassda Corp. has rocketed to near supremacy in IC SPICE simulation with 35 percent market share. That's up from 13 percent share the year before, which is a nearly unprecedented jump in a single year. The trouble is that Nassda is embroiled in a lawsuit with Synopsys Inc. over whether or not CEO Sang Wang and the former EPIC team of Synopsys, who are the brains of Nassda, misappropriated trade secrets and defrauded Synopsys in defecting to form Nassda. For its part, Nassda denies any wrongdoing.

Synopsys became the standard bearer in IC SPICE simulation with its acquisition of Avant! because Avant! had pummeled this market in the 1990s with a tool called HSPICE. In fact, HSPICE held 62 percent market share as recently as 2000. In real dollars, one software tool originating from free software models brought in at least $40 million to the Avant! coffers that year.

Nassda brought the Synopsys HSPICE share down to 38 percent share in 2001, albeit with some minor help from Cadence's Spectre line.

It's not nearly so simple, though. Nassda's HSIM competes directly with Synopsys' NanoSim, not HSPICE, in the area usually dubbed "fast SPICE" simulation. The predecessor products were developed by the very same Nassda team, when it was EPIC Design, which Synopsys bought. In fast SPICE simulation, the mathematical computations of transistor-level simulations are reduced so that one can achieve a reasonable interpretation in days instead of weeks, which is how long true SPICE simulation, as in HSPICE, takes. The trade-off is accuracy for speed.

"It's a gray area really. Nassda can do about 70 percent of what HSPICE can do. The two are complementary," says one CEO, who requested anonymity.

In one of transistor-level simulation's most bitter ironies, true SPICE is ever more critical as finer geometries cause thornier signal integrity issues, yet finer geometries make the designs so big that SPICE takes so long as to be at times useless.

"There's a lot of people saying that true SPICE simulators are dead, but I don't agree with that at all," says Les Spruiell, a product marketing manager for Cadence Spectre, Spectre RF and the Agilent alliance for mixed signal design. "You have to have a really sophisticated analysis engine. To look at 90 nano effects and extract deep analog stuff ... you can't treat that as a number in a digital simulator. At 90 nanometer and 65 nanometer, those have stopped being digital designs. Those are deeply analog designs. You really need the ability to get that level of accuracy."

Nassda has taken hold because it handles these fast simulations hierarchically, so the full chip, including memories, can be encompassed and in a reasonable time.

But Spruiell says that people forget there is no free lunch. "With these SPICE-like [or fast SPICE] simulators, you have to give up something. The math in these simulators is 300 years old. It's good old Newton-Raphson algorithms. That's not Richard Newton [the UC-Berkeley electrical engineering chair], that's our friend Isaac. It's fundamentally a Gaussian limitation. You either take short-cuts and get the speed, or you do the math and wait. There's all kinds of tricks you can play, but you got to give up something," says Spruiell.

So what's the big deal about a worldwide EDA segment of $70 million this year? The answer is in the challenges of those finer process geometries as well as serial I/O. Intra-electronic communication moving forward needs to get to rates of several gigabits per second, and that takes an analog block on-chip for serializing and deserializing the data. Intel has 140 companies lined up behind the multigigabit serial I/O protocol PCI Express and most of them promised OEMs they would have silicon to deliver it by the end of next year.