This article is more than 1 year old
Intrinsity to sample Intel-beating CPU late 2002
Performance leader... or Exponential Technology 2.0?
As chip makers like Intel, AMD and Motorola struggle to drive up their products' clock speeds while ensuring they are still economic to manufacture and sell, a little-known semiconductor company believes it can outshoot the big guns with faster and - more importantly - cheaper chips of its own.
The company, Intrinsity, this week demonstrated a test processor that not only runs at 2.2GHz - faster than anything yet shown in public by Intel, which currently leads the clock-speed race - but can be punched out in high volumes using a bog standard chip fabrication process that's rivals wouldn't consider using for a moment.
That's just a generic, proof-of-concept chip - Intrinsity promises faster, more tightly optimised products down the line. It will begin sampling its first commercial product during the latter part of 2002, the company told The Register today.
By then, Intel will be nudging up to or may even have exceeded 3GHz, but the cost of each chip will be significant. Yields won't be great, and the semiconductor giant will have to subsidise its products by a hefty margin to make their price appealing to PC makers and buyers. It will also have to rely on expensive cutting-edge chip production processes, including 0.13 micron technology, 300mm wafers, silicon-on-insulator and copper interconnects.
Intrinsity, by contrast, reckons it can deliver comparable performance with yesterday's processes, rather than tomorrow's. Using these newer techniques can only improve performance even further.
Dynamic not static
The basis for the company's claims is its Fast 14 technology - its new take on an old design methodology called Dynamic Logic. Before the arrival in the 1980s of automated design software and the CMOS manufacturing process that helped make it possible, DL was how all chips were designed. Today's Static Logic methodology - essentially chip-building by numbers - made for slower, less efficient processors (by between 50 and 80 per cent) but ones that could be designed, tested and put into manufacturing for more quickly and cheaply than DL.
DL has never gone away because it's ideal for the development of highly performance-sensitive circuitry. But since you need highly skilled staff - and plenty of them - to do DL properly, it has become uneconomic for all but a few on-chip units.
Intrinsity claims Fast 14 puts DL design on a par with SL. The company admits DL remains harder to do, but the cost of the extra work is lowered sufficiently for the extra performance it brings. Fast 14 comprises enhancements to the basic building blocks of a DL processor that not only make automated DL design tools effective but have the knock-on effect of reducing circuit noise - a major barrier to higher clock speeds - and increasing gate speeds.
The proof that this all works, says Intrinsity, is its test chip. But then that's what Exponential Technology said in the mid-1990s. Exponential was the last semiconductor start-up to promise it could bring radical performance gains to mainstream microprocessors by utilising new takes on old methods. Exponential claimed it could offer way faster PowerPC chips using a manufacturing technique developed for mainframe processors.
Unfortunately, what Exponential could demonstrate in the lab it couldn't bring to volume production, and was quickly abandoned by its most vociferous supporter, Apple. Apple gave its backing to Motorola's PowerPC 750 - aka the G3 - which gave a big leap in performance but a year or so back left the company languishing at 500MHz. Only now has it been able to reach 867MHz, still a long way off Intel's peak of 1.8GHz, which will become 2GHz the week after next.
Exponential's mistake, perhaps, was too talk up its technology too soon. Intrinsity - which has a number of ex-Exponential staffers on its pay-roll - is taking a more cautious approach. It's goal, it says, is simply multi-gigahertz processors that consume around 15W of power (about average for modern CPUs, but less than top-speed P4s) - it doesn't hand out hugely inflated clock speeds the way Exponential did.
Seeing out the slump
And with the semiconductor market in such a depression, it's arguably not the best time to run around making wild claims about new processors. "It's exactly the right time for development," an Intrinsity engineer told us. Expect the company to keep its head down while it continues to perfect its technology and work on its first commercial product.
And Intrinsity does have work to do. It can demonstrate a chip running at 2.2GHz, but it will have to show it can solve the power disspation problems inherent at that and higher speeds. Power consumption is an issue too, and it's telling that the company says its product will target applications "that need a high level of performance" rather than those whose main concern is saving power.
Routers, desktop PCs and servers seem to be Intrinsity's areas of interest right now, as well as more performance-oriented embedded applications. Transmeta, for one, needn't worry then. Other chip companies should - especially if a concerned Intel gets out its cheque book... ®