This is an excerpt from a January 10 article published by
Microprocessor Report (copyright issues, so I'm only pasting a few portions relevant to the ATI relationship).
Intrinsity Takes Its IP on the Road
First Licensee Was ATI; More to Follow
By Kevin Krewell
Intrinsity has been the subject of numerous Microprocessor Report articles. In the past, we've reported on Intrinsity's proprietary and patented Fast14 logic design and the processors (FastMath and FastMIPS) the company produced using this technology that outperformed the competition. Intrinsity's design even won our Analysts' Choice Award for best extreme processor of 2002. (See MPR 2/18/03-05, "Extremely High Performance.") Fast14 allowed the FastMath processor to run at 2.5GHz in 130nm process while most of its competitors were struggling to get to 1GHz.
But Intrinsity is a small company populated mostly by engineers. Building processors is one thing; selling them, quite another. Although the founders had visions of creating a fabless processor company much like SiByte or QED, those other companies were subsumed by larger companies with a broader product base while Intrinsity was still slugging it out on its own. Eventually, Intrinsity's business model of selling processors became untenable. Still, the company had a design technology that could give any chip company an unfair advantage over those using traditional design methodologies and binary logic design. Intrinsity has now transitioned from fabless semiconductor manufacturer to licensable IP and consulting company and is offering the market its Fast14 technology and design chain.
In February 2004, Intrinsity announced it had licensed the Fast14 process to ATI. (See MPR 2/23/04-01, "Intrinsity Licenses Fast14 to ATI.") At the time, ATI claimed it expected Fast14 would allow ATI's synthesized graphics cores to run about four times faster than those created using the traditional design process. This situation would offer ATI a significant advantage over its competition. First, however, ATI must incorporate the Fast14 design flow into its tool chain, something that still requires considerable support from Intrinsity. To date, ATI has not commented on when or where it will use the Fast14 technology.
Before that announcement, we wrote that "If it's as easy to design and build a 2GHz microprocessor using Fast14 as Intrinsity claims, there should be a huge market awaiting the company in the system-on-a-chip business. Whether Intrinsity will pursue this market, or simply use its technology and tools to create standard parts, remains to be seen." As we saw, Intrinsity tried to pursue the chip market, but its effort appears to have fallen short.
----------------------
Speed Comes With a Price
As noted earlier, we've written extensively about Intrinsity and Fast14. The technology has great promise and has been demonstrated to work in real chips. Therefore, one might ask, what is preventing greater acceptance? As we noted, Intrinsity was using the Fast14 technology to develop its own chips. With the change in business model, Intrinsity must do more to educate the design community on the way to migrate to a Fast14 design flow.
Fast14 works best in computational and control circuits but is less well suited for load and store buffers. Therefore, real-world designs will likely require a mix of dynamic and static logic. The use of NDL logic does lead to more interconnect lines, which could create a design problem. Transistors continue to scale down, but it's more difficult to scale copper interconnects without increasing wire resistance.
Before Intrinsity can support tons of designs, however, it must port its design library to a greater variety of semiconductor processes. Unfortunately, although the design methodology process has been automated, Intrinsity has not automated the development of its library on new semiconductor processes. Each library development effort requires three to six months, which could slow new design efforts.
The biggest problem for Intrinsity is probably that of education and the inertia of the chip-design community. Adopting Fast14 requires a new design methodology for circuit designs, and it requires training and handholding from Intrinsity (at least in the early phases). Intrinsity is still a relatively small company and can support training only a limited number of engineers. Opening up Fast14 for wider acceptance will require more time, people, and money.
Did'nt NV also license the the technology?
