The search for semiconductor IP intensifies

But arithmetic doesn't capture the enormity of finding, qualifying and reusing IP. In reality, by many accounts, the industry's experience was initially horrendous and has since improved to the merely dreadful.

"The situation is so bad that we consider all IP defective by definition," Naveed Sherwani, president and CEO of ASIC venture Open Silicon, stated during a panel at the recent International Symposium on Quality Electronic Design. "We have an internal IP quality group that not only investigates the IP and the vendor, but surveys previous users to try to anticipate the problems."

Nor is Sherwani unusually pessimistic. Colin Harris, vice president of worldwide operations at PMC-Sierra Inc., said his organization has made a policy of acquiring IP for supposedly stable blocks, such as standard interfaces, where PMC's engineers could add little customer-visible value and where one would think IP quality should be reasonable. But "we typically spend about a third as much on IP evaluation as we do on the purchase itself," Harris said.

"Clearly, buying IP today is still risky, despite all the internal approaches to evaluation that large companies have put in place," lamented Ralph von Vignau, director of technology and standards at Philips Semiconductors.

One of the most formidable undertakings for any SoC design team today is the identification, evaluation and selection of IP. If the choice is poor, at best the team may overlook a product that would have saved it much time and effort. At worst, a single IP choice destroys the project.

Today nearly every large-scale user of IP has its own approach for dealing with this risk. The process often involves market searches, checklists, interviews and internal evaluations. It may extend to audits of the vendor's claims, mandatory test silicon and virtual attachment of vendor engineers to the user's design team. Despite all that support, the relationship may still end in a failed product.

In general, the IP acquisition process can be broken into a few distinct phases. The first is awareness. With hundreds of IP vendors-many of them small and transient-and with some companies offering broad product portfolios, just finding candidates that might meet the design requirements is a challenge.

The second phase is screening. With so many possible candidates-especially for IP that implements industry-standard functions-there has to be some way of narrowing the field before spending the time and money on more detailed analysis. So shoppers apply simple yes/no criteria to weed out the least likely to succeed.

The third phase is evaluation. This involves a number of dimensions. One, certainly, is verifying that the IP meets its functional specifications and the design's functional requirements. Another, equally important, is that the IP can be implemented in the current design since it is compatible with the user's flow and the intended process.

But there are less technical aspects of this evaluation as well. Can the vendor provide enough support to the design team during the IP integration process? What about during ramp-up, when process understanding is vital? Does the vendor have a viable business, or will the IP be orphaned sometime during the project?

The problem in finding an IP vendor for most functions is a superabundance of riches. Several major companies have made it their business to gather up huge libraries of IP, any of which can contain almost any industry-standard function. But especially in emerging areas or for unusual functions, small players may have important offerings. The problem for vendors and users alike is just getting noticed.

"There are way too many small providers of IP," said Michael Kaskowitz, president of the Virtual Socket Interface Alliance (VSIA) and general manager of the Inventra division at Mentor Graphics Corp. "You need some kind of metric to help sort through the offerings.

Thus, most IP users write a checklist-or several of them-to help do a quick estimation of IP suitability. The problem with that solution from the vendor's perspective, particularly for small vendors, is that each prospective customer has its own checklist. All the checklists may be superficially similar, but each requires a separate effort on the vendor's part during an evaluation.That has led to calls for a uniform IP evaluation checklist.

Early this month, the VSIA announced just such a list, QIP Metric version 2.0. The concept presents a standard set of questions that can lead a user through a thorough first-pass inspection of any piece of IP. From the user's point of view, the initial phase becomes a routine, instead of a one-off research project for each new piece of IP being considered. Many offerings can be disposed of quickly, narrowing the search to IP that could be appropriate to the project.

From the vendor's standpoint, the benefit is at least equally strong: Answer a single set of questions once, and you have on hand the data necessary to begin the discussion with almost any potential licensee. "This can really minimize the work for smaller suppliers," said Saverio Fazzani, technology director at Cadence Design Systems Inc.

Victor Berman, Cadence group director for language standards, added, "It's more than just convenience. Without a standardized process such as this, the finite resources design teams have to devote to the IP search would mean that some vendors would not get a positive hearing. And a situation in which only the large vendors get heard stifles innovation. It defeats the whole purpose of a third-party IP market."

One noteworthy feature of the QIP approach is its scoreboarding capability. The user can assign weights to the answers to each question in the spreadsheet and, after answering the questions, get a cumulative score for a particular vendor and IP offering.

Small differences in scores can be the beginning of valuable conversations between user and vendor. "A big advantage to the QIP Metric is the communications it creates," said Kathy Werner, reuse manager at Freescale Semiconductor Inc. "Two vendors may end up with similar overall scores on the metric but with strengths and weaknesses in different areas. With the metric, you have a quantitative way of discussing what the vendor needs to do to better meet your needs."

A manager at one IP vendor commented that the QIP Metric can also serve as a diagnostic for one's own marketing process. Walking through the QIP process, the vendor gave itself a very good score on a particular piece of IP, while the prospective user gave it a mediocre score. "It showed us very clearly that we had failed to communicate a lot of what we knew about the IP to our prospect," the manager said.

Gross differences in score can be a basis for eliminating a piece of IP-or a vendor-from consideration. So there is a motivation for vendors to tweak either their processes or at least their answers to the questions to score well. This latter possibility serves as a warning that just asking a vendor to answer some questions is not going to be enough.

"This is a multistage process," affirmed Philip Dworsky, director of marketing and DesignWare IP at Synopsys Inc. "First, it's 'Do they have what I need?' Then starts an examination of the vendor's design and verification process. Everybody starts with a different list of questions, but if you pass the initial questions everybody has to go much deeper than paper can take them."

Digging deeper

In fact, the questions on the first-pass checklist may guide further, in-depth assessments as well. "After we have made the initial decision that a piece of IP is a good candidate for evaluation, we will repeat in-house everything the vendor says he has done," said Jean Bou-Farhat, vice president of the CoWare group at LSI Logic Corp.

"We do a detailed audit to confirm the answers to the questions," said PMC's Harris. "Nobody's going to buy a piece of IP off the Web these days."

For many, the only real proof lies in shipping silicon-and sometimes, even that isn't enough for experienced IP evaluators. "We try to get silicon validation as early in the process as possible, but sometimes it just isn't available," lamented Cadence's Fazzani. Berman added, "Even when it is available, if the IP block is at all complex, it's unlikely that any one silicon implementation will exercise all of it."

So auditing-in effect, doing a design review with the vendor, retracing its steps and comparing them with your own standards-is vital. This auditing process can be quite invasive, with members of the user team reviewing design flows and-especially-verification flows in detail with the vendor's engineering team. It can, as in the case of Open Silicon, involve discussions with previous users of the IP. And it may involve data that the vendor is unwilling to share or questions the vendor's engineers may find offensive.

But increasingly, as design teams take measure of the risks they are assuming in licensing third-party IP, they are becoming more willing to ask the hard questions. Even with such aids as the QIP Metric, this means an increasing flow of resources into the evaluation process. For this, there may be no solution short of an independent evaluation service-an Underwriters Laboratories for IP, as it were. The VSIA, for one, would love to see it happen, but Kaskowitz is not holding his breath.

"Auditing processes are a first step," Kaskowitz said. "In the long run, it would be great to have an independent certification process. But right now it's hard to see who would fund it."

 See related image