When will we stop?

Today I read an article in TOI that Intel is planning shelving its Tejas project. It was attributed due to the underperformance of its high-end processors as they produced more heat and their performance deteriorated. So now they are going for multi-processor on single silicon wafer architecture, if I may say, than try to squeeze in as much single processor processing power on a single silicon wafer.

I was just wondering when this quest to fill in as much as one can on the tiny silicon wafer would end. Nobody knows, not me at least! People speculate about it.

The semi-conductor journey began in late nineteenth century when Ferdinand Braun made the first cat's whisker diode for radio receivers. The devices made were discrete: that is, they performed only one function and to get any useful thing done from then they had to be wired together to make a functional unit. The story remained the same until in 1959, Jean Hoerni and Robert Noyce(One of the founders of Intel) developed a new process called planar technology at Fairchild Semiconductors which enabled them to diffuse vaious layes onto the silicon surface to make transistors, essentially a network of interconnected discrete components.

It itches me to write here that Fairchild was _the company to have a great impact on silicon industry; which gave us the now famous companies like Intel. Actually there were 8 brilliant people in Shockley Semiconductor Laboratory who left the parent company; they said they were feeling stifled and there ideas did not find any space in the company; though there were reasons other than this. These 8 people were called The Traitorous Eight!

Back on our track. Started only in 1959, by the end of 1960 nearly 90% of all the semiconductor manufactured was Integrated Circuits Since then the semiconductor has never looked back, We had Large Scale Integration then Medium Scale Integration and now Very Large Scale
Integration. There’s no dearth of real estate. There’s plenty to be used.

But now we have probably hit the wall. The process of squeezing as much power into the silicon real estate as much as we could has now seen one a slow down. All the Integrated ICs have transistors which are more or less like toggling device; on and off. For operating any ICs these transistors have to be switched on and off, at a very high rate, for 2GHz processor it toggling almost every 2 nanosecond! This heats up the processor/IC and degrades its performance. With the scale going smaller and smaller the scenario worsens.

For the industry to follow the 5 commandments we moved on to the vertical plane . So instead of expanding in the horizontal plane we expand in the vertical plane. So it would seem that we would continue to grow on the vertical plane as we grew on the horizontal plane and some years down the lane we would face the same problem. Well no for the simple reason that all the discrete entities in either of the plane have to be connected with wires, as small as 90 nanometers wide! So this poses problems as to how to route these wires from one plane to another. It may sound easy but believe me it isn’t so.

So the obvious is to move to something, which does not need wires for transmission, does not generate as much heat! Plasma and light would be the most rewarding. In fact there are nano-transistors , which use plasma as conductor for currents. The light has also been used. They have put small lasers inside the ICs, which emit directed light to the receiver. This reduces the heat generation and also the power consumed. But my question remains the same, when are we going to hit the wall with these optics in ICs.

How long will we go on reducing the size to opto-couplers, Laser diodes and photo-receivers? Will we go on reducing the size until there’s only a photon to represent the optics and only an electron to represent the transistor? Wouldn’t the photon then disturb the electron’s motion to create havoc in our transistor? Even if it doesn’t then aren’t we converging into the realms of Quantum mechanics. Will we move on to neurons taking place of transistors or may be processors/ICs to do all the computation at the speed of light! Will there be something more fascinating than the processor on semiconductor real state, something more fascinating than the neurons, something that anticipates and does in femto-seconds?

Or will there be processors that transport all the computation and problems in the past and programs it to deliver it to the future then, which is now or a few second of past, so that as soon as we given the problem we get the solution or as soon as we think of the problem the solution is before us.

Am I dreaming?