The Future Of The Processor

If you add to that the need to

process multimedia content

(like video, speech, animation and music) the workload of processors will encompass not staid scientific and commercial stuff but computing intensive applications designed to shift tasks from the big desktops to small, handheld devices

"Technology will not limit how complex a processor we can make," says Dr Pradeep K Dubey, Research Staff Member at the IBM Research Solutions Center in New Delhi. "Cost will. But the idea is take this computing power and make it cheap enough for the masses." Dubey should know.A thoroughbred chip designer since the early eighties, he has worked for Intel and has been closely associated with the design and architecture of 80386, 80486 and Pentium family of processors.

But he believes we are in the midst of a fundamental change, which will radically alter future processor design.

Traditionally, communications and computers have not merged to the extent possible because of the different software models on which the two are based. The merger of the two will alter the workload of processors. Add to that the need to process multimedia content like video, speech, animation and music. The workload all of a sudden goes from staid scientific and commercial stuff to computing intensive applications. "Computers are going beyond performance issues with the merger of communication and multimedia. The trend is that a lot of the image processing and communications type work is now being done on general-purpose programmable platforms as opposed to some special single processor. All one has to do is just write the software. This change will encompass everything-even routers," says

Dubey.

The popularity of mobile computing adds another dimension to this. Big desktops perform these tasks today but the future will demand the same to be available on a cellular phone.

What do these changes mean for the hardware? For a multimedia enabled hand held, the processor would need to be small, energy efficient and powerful-all at the same time. But in the future if you could create an architecture that could scale from a telephone instrument to a server imagine the cost benefit and ease. That is the holy grail for a chip designer-a magic chip that does all. And it has been attempted. John Moussouris, co-founder of MIPS Technologies Inc., founded MicroUnity Systems Engineering Inc., 11 years ago, to do just this. MicroUnity was to develop a chip that promised hundred times the speed of an Intel Pentium at one-tenth cost.

The company had $130 million in backing from Microsoft, Tele-Communications, Time Warner and Motorola among others. But it still failed. MicroUnity was depending on a brand new architecture to be built around a new patented process, a difficult task to accomplish even for an established player let alone a start-up. However, Dubey believes that MicroUnity has not failed in the concept sense. "Better execution could still deliver," he says.

But even if a magic chip were developed, the limits of physics would catch up sooner or later. Its not an immediate problem, at least according to the Semiconductor Industry Association's National Technology Roadmap. The roadmap sets as a goal the continuation of current exponential increase in capacity and performance up to the year 2010. Effectively that means that current CMOS (complementary metal-oxide semiconductor) technology, which is used to manufacture microchips, will continue to deliver the goods. Perhaps even longer-if we factor in advances in technology along the way.

But to look beyond and hazard a guess is probably as futile an exercise as it would be inaccurate. Nascent and possibly disruptive technologies like quantum computing, DNA computing or optical computing seem like options today but then so did vacuum tubes half a century ago.

ABHAY SINGH





Cyber News Service