For ASIC design, in-house rules

For ASIC design, in-house rules
By Crista Souza, EBN
March 26, 2001 (3:22 p.m. EST)
URL: http://www.eetimes.com/story/OEG20010323S0033

Despite a growing need to quickly cram more functions on a chip, ASIC designers have been slow to adopt a model that relies on preverified intellectual property blocks from third parties.

IP quality and support have always been a major concern among OEMs and semiconductor makers, but ease of use has recently grabbed more attention. While standards groups and design automation tool vendors touted their efforts as the missing link in effecting more efficient IP trading methods, IP users at last week's IP/SOC 2001 conference in Santa Clara, Calif., told a different story.

Surprisingly, the wide commercial availability of functions such as USB, PCI, and UARTs do not make system-on-a-chip design any simpler, according to some industry executives. "When it comes to what I call jelly bean IP, it's not an automatic decision of make vs. buy," said Steve Schultz, worldwide ASIC CAD strategy manager at Texas Instruments Inc., Dallas. "It's a much more comp licated decision-and the cost/benefit trade-offs, in the majority of cases, do not favor an external solution."

As a result, developers of commodity IP are simply biding their time, said Gary Smith, an analyst at Dataquest Inc., San Jose. Smith has long contended that, except for "star IP"-critical system elements such as microprocessors and DSPs-people will ultimately discover that it makes more sense to design IP in- house, even if the technology in question is not their main strength.

The reason stems from a shift in the way chip design is being done. As circuit designers start to work at the system level, the form in which most IP is provided-register transfer level (RTL) code-is hard to manipulate, he said.

Some contend that new EDA tools on the horizon will make it easier to evaluate the characteristics of RTL blocks and see upfront how the IP will interact with other blocks. Others say that-new tools or not-enabling the proliferation of commercial IP will require deep, cultural changes.

Despite the challenges, Dataquest projects the IP market will grow to nearly $1 billion in 2001, doubling its size from 1999. Most of the revenue, analysts said, can be attributed to sales of embedded microprocessors.

Not everyone has written off store-bought IP. Sun Microsystems Inc. uses cores developed outside the company to improve its time-to-market. The Palo Alto, Calif., computer OEM still develops most of its own IP, but increasingly purchases standards-based bus interfaces.

But the company places strict requirements on the IP it buys, said Satya Simha, hardware manager at Sun's Information Appliances Group. "Quality to us is absolutely zero bugs," he said. "We will not work with vendors that can't guarantee their IP works to our timing needs and quality expectations, and conform to industry standards. One thing the IP company never does is pay for our re-spins."

Joe Pumo, director of the SoC Design Technology Group at Motorola Inc.'s Semiconductor Products Sector, sees valid need for t hird-party cores, but acknowledges that in large organizations the "not-invented-here syndrome is very real. If it's a performance-driven IC, there may be certain reasons you have to do your own development," he said. "But if it's cost-driven, you have to look for and utilize what's available beyond simply what's in your own shop. You can't do everything inside anymore."