The art of the deal: what it really took to get Axis' new verification tool out the door - Case Study - Axis' Xtreme-II

Electronic News, May 20, 2002 by Ed Sperling

In the greater scheme of things, the decision by Axis Systems Inc. to base its flagship verification system on Xilinx Inc.'s FPGA technology was about as important as business decisions ever get.

It was a case of one company betting its entire future on another's, and because of that the top brass of both companies sat face to face in a succession of nervous meetings.

For Axis, the process began 16 months ago when it was looking to develop its next-generation verification tool, called Xtreme-II, which the company publicly announced last week. The company began its due diligence process to figure out what technologies were available on the market and how responsive the companies that made those technologies would prove to be.

From a technology standpoint, that decision was relatively easy. In the programmable chip world, Xilinx had just released its Virtex II product, so it was one step ahead of the competition in what has always been a game of leapfrog in programmable chips. Axis also had evaluated products from Altera Systems and Cypress Semiconductor, too.

"Every time you go through a new generation of product, you have to look at what's available today," says Michael Young, director of product marketing at Axis. "We normally go through a review that lasts three or four months, but on a continual basis we keep abreast of what's going on in the market.

From a purely technological perspective, Young says the decision involved the large number of multipliers in the Virtex II family, which are the fundamental building blocks for chips, as well as a decoding/encoding algorithm and embedded memory.

"The goal was to leverage the best technology available at the time so we would minimize our own risk," Young says.

But technology was only part of the issue. It was the prerequisite for choosing an FPGA vendor, but it wasn't the deal-clincher. Nor was the price. In the design and testing world of semiconductors, tools and support structure are far more relevant than price.

"There is some price sensitivity if you're creating very expensive parts," says Steve Wang, Axis' founder and its current VP of marketing. "Price is an issue, but it's not the primary issue. If you're paying $1 million for a mask and your yield is not great, that costs a lot more. So the accompanying tools are important.

"You really have to justify all of this based on the value of the system. It's more important to have high performance and ease of use. You can't just justify it based on the components themselves because the majority of the cost is software. The real justification is your overall productivity."

That has proven particularly true in the semiconductor market, where time-to-market is critical. Getting a design right the first time is a key part of that equation-particularly one that contains all the functionality it was supposed to have.

"The value-add is additional verification or getting to a tapeout earlier," Wang says. "Time-to-market alone was a major selling proposition two years ago. Now people are paying more attention to quality."

Behind Closed Doors

In the boardroom at Xilinx, the real issue was the comfort level between the two companies. Xilinx sent out its CTO, VP of engineering and its CEO to make the deal happen.

"[Axis] wanted to make sure Xilinx could deliver a product that was core to their value proposition," says Erich Goetting, VP of Xilinx's advanced products division. "They were basing the entire company on a programmable logic supplier. They got to know the field sales organization and our management. But because our production system was going to be the heart of their system, they needed to feel good about the company.

The process took several more months. At that point, Axis was ready to set its engineers to work on the emulation technology that last week turned into its Xtreme-II product. And the engineering wasn't just done by Axis.

"As we've supplied Axis and other companies, we've taken a lot of input that has appeared in the chip already," Goetting says. "One is the speed of reconfiguration. A chip has to be 100 percent perfect, and an emulation application has to work over the life of a product and use more of the chip than currently is needed, so the need for perfection is higher than the general market."

That becomes increasingly difficult, however, as programmable logic becomes more complex. In previous generations, Xilinx says that reasonably proficient designers could get up and running in one to two weeks of training. Today, it takes an average of six to 12 months to achieve that level of proficiency, which is what keeps engineers from swapping from one vendor's programmable logic technology to another's. It's also what makes a decision by companies such as Axis so important to Xilinx.

As such, any deal involves showing companies such as Axis a detailed roadmap for years to come.

"Programmable logic technology is a key enabler for them," Goetting says. "They have to see our technology roadmap years in advance and sync up their technology with ours. A lot of our discussions are NDA (non-disclosure agreement). They need to know how much it's going to cost for memory and I/O. We do the same with silicon and packaging. It's a long-term strategic alignment process."