Moore’s Law not a barrier to progress

1999-10-12

According to Paul Pakan, an Intel semiconductor scientist, we are approaching the limits of Moore’s Law which states that the number of transistors on a chip will double every 18 months. This should not, at least in the short term, provide much of a hurdle for progress. How can we say this? Well, for several reasons.

First of all, as an obvious point, Pakan notes that the limitations will start to bite in 2001 and beyond. However this is not the crux of the debate. Hardware technologies have been far and away in advance of software technologies ever since Bill Gates determined that 640K of RAM was sufficient for the fledgling PC operating system DOS. Software, in general, is inefficient and burdensome, using up chunks of hardware resource for the simplest of manipulations. Should software vendors discover that the hardware resources upon which they rely are not infinite, then they might be tempted to develop more efficient designs. Embedded software developers have worked within their bounds with excellent results, and it could be argued that the top end of chips required for next generation mobile devices will be sufficient for most needs.

This leads to the second point. To make the best use of limited resource requires a well-designed architecture. Thin client is becoming the approach which is recognised as best for both application partitioning and consequent manageability. The capabilities of the end user device may require to be limited not only because the hardware has reached its limits but also because the application architecture requires it.

Finally, we have so far neglected to mention the huge leaps forward in hardware design, not only concerning the physical construction of chips but also the designers’ abilities to create on-chip components which exploit the underlying hardware to the fullest extent possible. We only have to look at how processors such as those from Intel, AMD and ARM have been constructed, with features such as intelligent instruction look-ahead and multithreading, to see that the transistor size is only one (albeit important) facet of chip design.

As noted on the Register, “the sooner they can get quantum and/or optical processors to work, the better”. In the meantime, however, there is plenty to be keeping both hardware and software designers busy, and plenty of progress still to be made, with the technologies we have available today.

(First published 12 October 1999)