Exponential Technology Inc, as reported (CI No 2,809), is claiming to have a revolutionary BiCMOS design process that will enable it to fabricate PowerPC chips far faster than anything that IBM Corp or Motorola Inc have in the pipeline. The San Jose, California-based company, backed to the tune of $14m by Apple Computer Inc and a collection of venture capitalists and private individuals, has negotiated the right to build its own PowerPC chip and says that its first product will tape out early next year with volume production following in early 1997. Exponential says its PowerPC processors will offer twice the performance of CMOS for the same die size – possibly better. The first processor will be a functional equivalent of the PowerPC 604, and though the company is being careful not to quote clock speeds, ‘twice the performance’ would imply a 300MHz to 400MHz part. One source outside the company was talking about 700MHz parts last month. The large majority of the chip’s logic will be made in bipolar circuitry, with CMOS used for the onboard cache.
Burn up
Despite early rumours, the company says that it is not involved in building iAPX-86-emulation hardware into its PowerPC clones. Bipolar technology has traditionally been the province of mainframe manufacturers – it switches faster than CMOS, with a consequent improvement in processor clock speeds, and the more you increase the power, the faster the thing switches. Unfortunately, there has to be a trade-off between the power you apply and the circuit density you can achieve because if you design the thing to design rules that are too fine, the chip simply burns up; CMOS is not so fast, but the components can be packed much more closely together without the thing burning up. And as the technology has advanced in leaps and bounds over the past decade – it was scarcely used before 1980, and only started seriously supplanting NMOS in around 1985 – even mainframe manufacturers have switched to CMOS-based processors. CMOS adds an extra component to every switching element to enable it to switch at low power, and until fabrication processes caught up, the penalty of the extra components made it much slower than NMOS. BiCMOS, as used by Intel Corp in its initial Pentium designs, aims to bring the best of both worlds by introducing relatively small amounts of bipolar logic into the speed-critical areas of a CMOS processor. However, according to Exponential, this approach has its drawbacks. Exponential claims that this conventional bipolar-on-CMOS approach results in a relatively weak form of bipolar logic suitable only for building small-to-medium size logic functions and driving long wires.
By Chris Rose
An alternative approach – placing small amounts of CMOS onto a predominantly bipolar chip has languished, mainly because of the size and energy drawbacks. This is a pity since a fast bipolar chip with onboard CMOS cache and support circuitry could be a winner. Exponential claims to have cracked the problem by discovering a way to design bipolar circuitry that is roughly equivalent to CMOS in density and energy consumption, in particular, it contends, since CMOS elements tend to grow in size anyway as clock speeds increase. The company has applied for 18 patents covering bipolar design methodologies, tools and also chip packaging. So far eight patents have been granted. There is no reason why the process shouldn’t be applied to create BiCMOS variants of any microprocessor and if the company manages to deliver on its promises, its technology will be attractive to a number of chip makers. However, Exponential chief executive Rick Shriner (formerly vice-president of core technology at Apple) says that there are no plans to license the technology out, and anyway we’ve still got to prove it. No, instead Exponential intends to sell its products on the merchant market, in competition with IBM and Motorola at the high end. The company does not have its own chip fabrication plant, and is unwilling to divulge who will be making them, other than to say that the fab will be off-shore and the company isn’t a US one. Hitachi Ltd already has some ill-defined deal to build PowerPC processors and has also been talking about its innovative BiCMOS fabrication technology. Ivonne Valdes, vice-president of sales denies, however that the company is unduly tied to its nameless manufacturing partner – Motorola and IBM, among others have similar manufacturing capabilities already, she suggests. Exactly how IBM and Motorola will react to the competition will be interesting to watch: as we went to press they had no comment. Rick Shriner says that relations are very good, but acknowledges that if he had the time he could talk all year about the negotiations required to get the rights to build the PowerPC variant. Neither IBM nor Motorola are being forthcoming about their thoughts on Exponential. The third partner in the PowerPC triumvirate, Apple, is obviously delighted at the prospect of competition. It was an original investor when the firm was founded in the summer of 1994, and came back for second bite of the cherry in a subsequent funding round.
Flag in the ground
They were looking for someone to put a flag in the ground in terms of PowerPC performance. Exponential’s financial footing is nearly as hazy as its initial product plans: $14m has been raised in two rounds of funding – the first round saw Apple accompanied by venture capitalists, Venrock Associates and a group of undisclosed private investors. The second round saw the two companies joined by Itochu Corp, Innotech, Nazem & Co and Woodside Fund. One investor is Jean Louis Gassee, formerly of Apple and now head of Be Inc. But whether Gassee is involved on Be’s behalf, privately or through the Innotech venture company, was not clear as we went to press. To date Exponential says it has spent half of its funds and expects to start a second round of funding after tape-out of its first product next spring. So can it do it? Exponential itself points out that the technology is untested, but the names behind it are impressive. The company was founded by George Taylor and Jim Blomgren: Taylor had been director of experimental architecture at Sun Microsystems Laboratories, Blomgren also worked at Sun Labs on ECL Sparc processor design and at Chips & Technologies Inc on 80386 and 80486 processors. Exponential’s chairman Gordon Campbell was also at Chips & Technologies – as chief executive.
