Santa Cruz, Calif. - Despite promises of breakthrough technology in IC physical design, EDA startup AmmoCore is going into liquidation. The company fell victim to a difficult marketplace with unexpectedly long sales cycles, president and chief executive officer Gary Larsen said.
After five years of operation and $33 million in venture capital, AmmoCore has restructured in order to sell the company or its assets. It retains a small crew to support its Fabrix netlist-to-routing product until a home is found for the technology, but it is no longer seeking new business.
It wasn't supposed to turn out this way. One of the more notable EDA startups of this decade, AmmoCore had technology that promised to speed time-to-tapeout for large ASICs and ICs. It employed more than 50 people at its peak and had attracted such EDA luminaries as Jim Hogan, Bruce Bourbon and Glen Antle to its board.
But the board finally decided not to invest any further, said Larsen, who joined AmmoCore in 2003 after serving as president and CEO of Circuit Semantics. Aside from the long sales cycle-"a lot longer than we ever anticipated," he said-a second problem was that AmmoCore's software is best suited for chip designs of more than 5 million gates, and "there weren't as many of those designs starting up as we had anticipated."
AmmoCore's Fabrix could have initiated a new methodology for chip design. The tool takes synthesized netlists, converts the logical hierarchy into a physical hierarchy and partitions the design into "SBlocks" that contain 1,000 to 10,000 standard cells each. It then optimizes the placement of the SBlocks for timing and power, reportedly while eliminating signal-integrity problems within the SBlocks.
It's a complete netlist-to-GDSII solution except for one thing: detailed routing. Larsen said the absence of routing wasn't a big problem for potential AmmoCore customers, although it could have been over time.
Publicly announced in 2002, Fabrix didn't ship until late 2003-much later than expected, Larsen acknowledged. In February 2004, AmmoCore announced that Fujitsu had taped out a 4 million-gate, 0.11-micron chip using Fabrix. A Fujitsu engineer confirmed that the chip took only four weeks from netlist to tapeout, a process that would have taken a few months using a traditional flow.
However promising its technology, AmmoCore couldn't have chosen worse timing. The company got its start, and its initial funding, right at the end of the tech boom and then ran headlong into the downturn.
"Last year, things started to open up a little bit for us," Larsen said. "The year before was tough sledding-people didn't want to talk at all. We had three or four cases where companies were very interested in using our software for large designs, and [then those projects] either got postponed or canceled."
Another difficulty was competing with the big EDA vendors, which offer bundled packages that included all the software a customer might need, Larsen noted. For customers, adding software from a startup adds cost that can't be recouped by deleting an item from the bundle.
Larsen said there are "about 100 lessons learned" from his experience with AmmoCore. One was that the company did too much selling and benchmarking before the product was ready. When he joined the company, Larsen said, the marketing effort was detracting from the R&D work needed to complete Fabrix.
"If I had to do this all over again, I would probably have bitten the bullet and stopped all selling for six months, to tune the software and reduce some costs," he said.
But the main lesson was the harsh reality of the marketplace.
"I think we were too optimistic about the sales picture," Larsen said. "We didn't know how long it would take to bring things to fruition."