System on a chip? Duck à l'orange, more like

WE HAD THE CHANCE to meet up with the boss of Fairchild Semiconductor earlier this week - you know, the outfit that spawned Intel and the entire semiconductor industry. As so often happens, talk turned to Systems on a Chip [What a strange world you live in - Ed]. Fairchild's CEO, Kirk Pond, dismissed systems on a chip by saying "I only know of two true SoC devices - one of them's a watch and the other's a calculator."

Pond went on to say that, despite CPUs packing around 50 million transistors these days, little of the mobo functionality is finding its way onto the processor die. As a result, Fairchild - which makes practically all the other components found on motherboards - reckons it's making more money on supplying components for Pentium 4 mobos than it did back in the days of Pentium III - "We sold between $2-3 worth of parts per system with Pentium III, now it's up to $8-$9 for Pentium 4 boards."

Pond puts this largely down to the need for more sophisticated power management, but it begs the question of whether SoC - once the Holy Grail for Intel and Transmeta - is a dead duck.

A couple of years ago, Intel was working on Timna, an SoC chip for notebooks. First it was delayed. Then it was delayed again. Then it was scrapped. Writers not a million miles from this place had been warning it was a stupid idea in the first place - it was designed as a cheap chip, but only supported the then staggeringly-expensive Rambus memory, so a 750MHz Timna notebook would have cost more than an 800MHz Celeron one. Sheer brilliance.

At the time, we said: "The total saving for a notebook manufacturer by using a Timna rather than a Celeron plus separate graphics and audio will be no more than $30, tops. How will Chipzilla price Celerons against Timna? The SoC chip should cost less than Celeron because it's entry-level, but the Celeron has less functionality, so it should cost less than Timna. If Celeron costs less than Timna, there goes a sizeable chunk of that $30 saving. If Timna costs more than Celeron, Celeron becomes the entry level."

It was almost as if Intel had been listening, for three short weeks later it pulled the plug. Problems with the memory translator hub (MTH) which would have enabled the entry-level chip to use SDRAM memory rather than the overpriced Rambus RIMMs for which it was originally designed, meant that by the time a workable solution had been found, the chip would have missed its market window. Not long afterwards, Transmeta canned its SoC design, the TM6000.

The chip giant conceded that positioning Timna as an entry-level product, while Celerons were available at a lower price, would have been extremely difficult. On the power front, Timna would also have looked distinctly underpowered compared with Duron and Celerons around at the time of its Q1 2001 launch target.

Intel's next generation notebook chip, Banias, isn't an SoC product either. Due for launch early next year, it will be supplanted by yet another non-SoC chip, Dothan, before the end of 2003.

System on a Chip? It was a nice idea whose time has passed. µ