Any SoC design manager can relate stories about the pains of using third-party IP. Surprisingly, many of the pains are not due to the problems of integration: they involve IP quality, the compatibility of cores within a tightly-integrated subsystem, the challenges of negotiating licenses, or simply the hassles of finding and qualifying cores in the first place. A new start-up in the IP space, ChipStart, might very well help with a number of these problems.

It is really easy to mischaracterize this new company. It's tempting to call it an IP distributor, or maybe a consolidator. But that doesn't describe the full value proposition. Nor is ChipStart an independent IP company in its own right. Nor is it a design-services company, although there are services involved. Maybe the best glib term would be IP facilitator.

Here's the proposition, according to ChipStart president and CEO Howard Pakosh. There are certain subsystems that occur repeatedly across a wide range of SoC designs. Among these are error correction and repair functions for embedded RAMs, DDR DRAM interfaces, and security subsystems, for example. Each of these subsystems is available as a project from the big integrated IP providers, for a price. But each could be implemented—perhaps more cost-effectively and with more exciting features—by assembling the latest IP from a variety of smaller IP vendors.

The problem with that approach, of course, is that it exacerbates the costs of finding, qualifying, and dealing with third parties, and places the risk for the subsystem squarely on the shoulders of the SoC team. And those costs multiply. Pakosh relates that one recent design he's aware of required the design team to identify, qualify, and negotiate for 28 separate IP licenses in one calendar quarter. "You can imagine that given the number of small IP vendors out there, that probably meant the design team had to evaluate close to 60 separate pieces of IP in the course of three months," Pakosh said.

This doesn't mean all the work is done. Support of the individual IP blocks is still provided by the IP vendors, not directly by ChipStart. And at this point, the customer still needs to negotiate a license with each vendor individually. But ChipStart is working to solve both of these problems as the company's resources grow and vendors become more familiar with this business model.

The goal, according to Pakosh, is to create a global sales and marketing resource that can save SoC teams much of the front-end work of finding and qualifying IP. At the same time, the goal is to create a resource to which smaller IP companies—or even big companies that want to focus their sales and marketing budgets on their major accounts—can outsource global marketing and sales. This should give smaller IP companies global reach and a more realistic chance of being found and considered by customers. And it could offer the big IP vendors some considerable savings on the income statement, which might also be welcome in 2009.

Initially, ChipStart will focus on four subsystems. The first will be embedded-memory test and repair—not a surprise given that many of the people involved came from Virage Logic. The company is working with SRAM vendor Dolphin and design-for-test vendor LogicVision to offer an integrated package for SoC designers. Pakosh said that the LogicVision has some key advantages in the area of memory diagnostics that have not been appreciated as widely as they should be. And ChipStart is working to develop a technique for test-and-repair optimization, so that given an SoC design and memory topology, the company can find the organization of test-and-repair structures that optimizes the area-vs.-yield trade-off across the chip design.

The second subsystem will be DDR—and in particular DDR-3—memory interfaces. Pakosh points out that these interfaces involve a rich mix of logic and high-speed analog IP, standards-based I/O, and proprietary scheduling algorithms, and that achieving anything like ideal memory bandwidth is far from a trivial problem. ChipStart is working with a number of IP suppliers in this area to create a complete tunable subsystem offering.

The third area is security processing. This product is still in the works, but Pakosh did say that the ballooning number of security stacks that connected SoCs must support has pretty much forced a migration away from hard-wired security black boxes to more flexible processors based on an embedded CPU core plus accelerators. Again, ChipStart's approach will be to pull together a team of disparate IP vendors to integrate a solution.

Finally, the company hopes to lever this last effort to field an embedded CPU core of its own, aimed either at security and/or at the Android platform. The particular need here, according to Pakosh, is to make the processor deliverable in a form that is virtually proof against reverse-engineering in the Asia-Pacific market, where much of the SoC design activity for handsets and other mobile devices will be, but where IP vendors are loath to provide enough design-level data to permit integration of a really good processing subsystem. This project is currently planned for completion in the third quarter of 2009.

In addition to the four subsystem areas, ChipStart is working on a software product to assist managers in tracking their use of licensed IP. Pakosh says that vendor auditing provisions in IP licenses, royalty estimation, and the demands of Sarbanes-Oxley all require managers to be able to very quickly estimate just what their royalty liability is today, and what it would be for a given new-product proposal. By the end of the first quarter ChipStart plans to have a software package in place that can tell them just that.