65nm to 45nm: process technology explained

The biggest headline about the Core 2 Extreme QX9650 is that it is Intel's first desktop processor to use a 45nm production process. The company's previous generation of desktop CPUs were produced using a 65nm process. But why is this change so important, and what does it mean for the processor business?

Die shrink another day

CMOS manufacturing advances trace the history of CPUs as much as the designs of the processors themselves. Each shrink in the size of the microscopic transistors, which make up the CPU, means more can be fitted in the same space, with a number of implications.

At the most basic level, you couldn't even make today's processor designs with the process technologies of just a few years ago - they would be unfeasibly massive. The 386 had just 275,000 transistors. Intel's Core 2 Extreme QX9650 has around 800 million - nearly 3,000 times as many. Using the 386's 1µm production process, the QX9650 would be about a foot square!

Power requirements are another issue. Smaller transistors consume fewer watts to cycle, which again means you can practically have more of them than with a larger process technology.

If you made up enough transistors for a QX9650 with 386s, they would consume around 3000W - yet an entire Core 2 Extreme QX9650 PC, including other components, only requires a little over 200W under full load.

Why smaller is better

Lower power consumption has another handy side effect. If your transistors draw fewer watts, they won't get so hot. So you can run them at a higher frequency without burning them out - or overloading the motherboard power supply circuitry they draw from.

There are other factors to consider, but each new process technology almost always means a higher ceiling on clock frequencies.

The last, but far from least benefit of smaller transistors comes when you keep the basic CPU design the same. In this case, the processor itself becomes smaller - known as a 'die shrink'. Since the fabrication system uses a standard-sized semiconductor wafer - currently 300mm is the largest - you can fit more onto each one.

The wafer production cost itself is the same, so each processor becomes cheaper to make. For example, a 45nm processor takes up half the area of a 65nm one with the same design. So moving to 45nm halves the manufacturing cost - although you also have to factor in the price of developing the new process and building the factory capable of performing it. This can be very expensive indeed.

The 45nm advantage

So it would seem smaller is always better for semiconductors, making you wonder why such miniaturisations don't occur more rapidly. However, there are always difficulties which must be overcome to enable each reduction in transistor size. These include parasitic capacitance, where parts of the miniature integrated circuit keep their charge when they shouldn't, current leakage and latchup.

The latter two have been a particular problem with recent process reductions, as the gaps between the tiny wires are so small it's becoming increasingly difficult to prevent current flowing where it isn't meant to.

AMD and IBM have been using Silicon on Insulator (SOI) technology to combat this and enable their moves down to 65nm.

The 45nm challenge

With Intel's switch from 65nm to 45nm, however, the company continues to use the older bulk CMOS technology, but with the addition of High-K dielectrics and metal gate technologies.

Traditionally, silicon dioxide has been used as the dielectric in the tiny transistors, but is prone to leakage at the manufacturing scales now used. Alternative materials with a high dielectric constant (High-K) prevent this.

In contrast, metal gates take the parts of the processor intended to be conductive in the opposite direction. Previously, less conductive polysilicon has been used for circuitry, because it makes manufacturing easier. Metal, in contrast, has almost zero electrical resistance.

These two technologies have allowed Intel to take the current lead in process technology. This gives it a competitive advantage on power consumption, processor clock speeds and manufacturing economy. It's not all about how well you design your chip architecture in the processor business.