Like it or not, IP's here to stay

In the early days, many believed that the chipless chip companies would take a starring role, supplying a steady stream of intellectual-property cores to be soaked up by an insatiable chip industry that couldn't possibly make use of all the transistors at its disposal. Investors poured in money, and companies were born.

But the industry never achieved that nirvana. Integrating third-party IP proved much harder than anyone had anticipated. Only a handful of providers managed to make the chipless model work.

It's time to get over it.

It may fall shy of the grand plan of a decade ago, but the IP trade is real and has become a vital appendage to the chip industry. Chip companies are indeed buying blocks of circuitry from outside

vendors, and some IP companies are thriving. And those who have a stake in IP are more willing today to have open and honest discussions about problems and how to fix them.

One persistent problem on the business side is that buyers and sellers of IP still haven't found much common ground. For the most part, there is still no pricing framework or any standard legal clauses to which they can agree. This leads to drawn-out negotiations that can fritter away months of precious design cycle and amount to lost revenue on both sides.

"Quite often, it's been my experience that it takes more time to negotiate a license than to do a design," said John Weekley, director of business development at design automation giant Synopsys Inc., which in recent years has become increasingly involved in IP development.

The division of labor is often a point of contention between the IP suppliers and the designers — and not knowing who's responsible for what can be disastrous when problems crop up. Finger pointing ensues, and the project gets thrown off track until cooler heads prevail.

And despite early promises of re-usability and seamless integration, the adoption of third-party IP almost always takes more engineers and more money to make it work in the design. Users blame providers for sloppy code and lack of design support; providers blame customers for not reading the design notes and for trying to do things on the cheap.

A strained IP relationship is like a bad marriage, said Steve Deeley, director of IP for Zarlink Semiconductor Inc. In the seduction stage, IP vendor and chip developer come together and expectations run high. The doubts surface when it becomes clear that the IP must work with the customer's tools, manufacturing guidelines and other components, such as on-chip CPUs and buses. "That's when it all starts to fall apart," Deeley said.

Vicious circle
Designers start asking for more support, just when IP design teams are trying to develop their next core. Chip designers find bugs and blame the IP vendor; the IP vendor accuses the designers of not reading the documentation. Meanwhile, the IP is aging and losing value with each passing week.

It can become a vicious circle: The chip vendor, to avoid getting burned again, will be more cautious when engaging with the next IP company. Believing that it won't get the support it needs, it will play hardball on pricing, perhaps demanding that the IP provider waive the licensing fees and accept royalty payments only.

But IP providers won't easily budge on licensing fees. With the exception of some of the larger IP vendors, like ARM Ltd. or Virage Logic Corp., most don't have substantial royalty streams and couldn't survive without up-front licensing fees.

So negotiations reach a stalemate — and everyone loses.

Some sympathize with the IP companies. Kurt Wolf, director of library and IP management at Taiwan Semiconductor Manufacturing Co. Ltd., says it's short-sighted for users to try to squeeze an IP supplier for another 5 percent discount when that might mean missing out on vital support services. Users will request that support whether or not they've paid the extra $20,000 for it, he said.

"There have been times when customers publicly say that support is important but then privately devalue it, and then they wind up needing it later on," Wolf said. Exhibiting a willingness to pay a little more for service, he said, "is really the healthy way for the IP industry to evolve."

Many first-time users of third-party IP don't understand that licensing and royalty fees represent just a fraction of the true cost of IP. They may underestimate or fail to consider the cost of making the design reusable, and of support and maintenance. Moreover, they often overlook the risks, such as the price for mask respins and lost profits.

"Everyone focuses on the obvious cost. But when you do the math it's the other costs that are dominant," Weekley said. "People really need to consider these [other costs] explicitly and not just look at the tip of the iceberg. There are some things that can kill you if you're not watching out."

Perhaps the biggest cost component is validating the IP in silicon. It's a multistep process that involves a review by the foundry, production of test chips and yield reports that can take two to three months, depending on the maturity of the process technology. But validation removes much of the risk and is the best way to ensure quality.

"We have a good flow, but silicon speaks at the end of the day," said Alex Shubat, chief technical officer at Virage.

Sometimes chip companies don't want to wait for an IP core to make a first pass through the foundry before getting started on a design. This is the case for about 20 percent of Virage's customers, Shubat said. Some, however, don't realize the additional cost they will have to bear.

"The less experienced will ask you to design some IP that is custom to their requirements," but they will not have budgeted for the necessary silicon validation, Shubat said.

Dabblers need not apply
IP vendors need a wake-up call too, observers said. Some IP vendors see IP as a sideline business or as an opportunity to exploit a market need, but chip vendors have become wary of opportunistic plays.

Deeley said he works only with those IP companies that have demonstrated a commitment to the business."You either need to get big, get a niche or get out," he said of IP players.

There are opportunities for startup IP companies, but providers should be able to back up their claims, Deeley said. "If you come to me and tell me you're a startup and have a great idea, that helps. What helps even more is when you tell me you have done this 15 times before and have an army of people ready to support me."

IP companies can also take some common measures to avoid conflict with their customers. First, they should expect that their customers will examine their code with a critical eye. Errors come in many forms, and IP vendors need to put procedures in place to root out the most egregious. Deeley recalled one incident when his company found a comment, beneath a command line, that read: "For God's sake, don't let the customer see this."

Synopsys, for one, said it learned the hard way how to minimize gaffes. Back in 1997, the company fielded one of the first PCI cores to hit the market. But it was forced to withdraw the product from the market after discovering that the core had to be customized each time it was used.

Synopsys subsequently adopted an approach that would make design-for-reuse a requirement for future cores, said Phil Dworsky, director of marketing for Synopsys' IP business unit.

It's tempting to look at the IP industry's many missteps and deem the experiment a failure — but that would be a mistake. The reality is that chip companies are benefiting from not having to make, say, JPEG and 32-bit RISC processors themselves. Moreover, companies like ARM, Artisan Components Inc., Rambus Inc., Synopsys and TTLcom have shown that the IP model can work.

Indeed, said Zarlink's Deeley, "When it succeeds, it can be absolutely huge."

Getting the IP business started may have been like trying to roll bricks. But the business is moving forward nonetheless — and dragging the naysayers kicking and screaming along with it.