Although it may be hard to tell from the explosion in chip speeds in the past year, an end to Moores Law is in sight. Manufacturing chips economically at 0.1 microns and below is going to be a challen
Although it may be hard to tell from the explosion in chip speeds in the past year, an end to Moores Law is in sight. Manufacturing chips economically at 0.1 microns and below is going to be a challenge for the semiconductor industry, yet one that it has to face if chips are going to keep getting faster. Currently, the fastest commercial chips are manufactured using 0.13-micron design rules.
With this end in sight, chip manufacturers have been looking to challenge the old assumptions about chip manufacturing and find new techniques that will make chips go faster without the designers having to tackle some of the physical problems that will crop up in the next ten years.
Surprisingly, some real discoveries have been made that should keep Moores Law on track for longer than even he expected. Some of the physics hurdles have been overcome, such as sub-0.1-micron manufacturing using extreme ultraviolet and infra-red technology. Currently, even 0.05-micron (thats five nanometre) manufacturing looks within reach.
More importantly, perhaps new ways of building chips have been developed. Intel, for instance, has developed new packaging that requires fewer layers in a chip, allowing more transistors to be packed into a smaller processor. IBM earlier introduced copper interconnects and silicon-on-insulator technology, which have boosted the potential operating speeds of its processors.
One of the most interesting developments is Motorolas gallium arsenide technology. Silicon is not the best semiconductor, but it is one of the cheapest and most readily available. Gallium arsenide is a far better conductor - up to 40 times better - but was too expensive to be used in a large number of chips. However, Motorola has developed a hybrid of silicon and gallium arsenide that it believes can boost processor speeds dramatically without corresponding increases in manufacturing cost.
These developments should lead to chips with speeds and transistor counts not even dreamt of today. Intel has already predicted to have a processor with one billion transistors (the Pentium 4 has 43 million) running at 1,500GHz within six years.
And what happens when conventional semiconductors run out of steam in, say, 15 to 20 years? There are plenty of other computing types waiting in the wings: notably optical computers (which use light rather than electricity), neural nets (which emulate the massive parallelism of the human brain) and quantum computers (which well get into a bit later). So Moores Law has a long way to go, and it will produce applications that weve never even dreamed of, although it might not refer specifically to the number of transistors on a piece of silicon, but to computing power in general.
Artificial intelligence
Alan Turing, one of the founding fathers of computer science, once said that by 2000 computers would be so advanced as to be indistinguishable from real people during an interrogation. He was definitely wrong on the timeline, but the potential still exists for this to happen.
Artificial intelligence (AI) was originally defined in terms of a computer performing a task that could normally only be performed by an intelligent person, although this definition is now considered to be too limiting. This is because programming a chess computer, or even a simple calculator, to perform better than an intelligent human in that area is not so hard, while the major stumbling blocks in AI continue to be the simple things like recognising faces, walking or carrying on a conversation, let alone things like abstract thinking, art and humour.
Conventional AI is really a system of stimulus and response; a long series of IF-THEN statements using Boolean logic. It can be as complicated as the computing platform allows, and as the programmers want to develop. Another approach is to use neural networks that do not operate along straightforward logical lines, but utilise the variable and relative strengths between neurons to react to input and learn.
Functional AI is around today, going even as far back as Eliza in 1966 (which wasnt very intelligent, but was designed to trick you into thinking it was). Theres expert systems, Net agents, game AI, medical diagnosis, fuzzy logic processors, heuristic programs (handwriting, speech and even facial recognition), chess playing, data mining and so forth. Stronger AI applications are even self-learning, adding to their own knowledge base through interaction and response.
To get Spielberg-like, full-human AI with both inductive and deductive reasoning, however, you need to muster up more power than we can imagine today. The processing power required to even come close to replicating the functions of the human brain is truly stupendous. Even the worlds most powerful supercomputer, IBMs ASCI White, which is capable of 12.3 trillion calculations per second (and is the size of a basketball court), would not be capable of replicating the full functions of the human brain. But then, personal computers today are thousands of times more powerful than the supercomputers of 50 years ago, and have yet to come close to reaching their potential - in fact, Ray Kurzweil, CEO of Kurzweil Technologies (and developer of a number of recognition systems) has said that he believes that by 2019, a $1,000 personal computer will match the power of a human brain.