Maximum Frequency of a Modern Microprocessor
(What are its factors and what physics has to do with it)

Over the past nine years the development of the process technology has not only become more complicated, but also more efficient – in an exponential manner. The first processor released on Intel’s P5 architecture was a breakthrough in process technology… manufactured on a .25 micron process (250 nanometers), this process allowed the processor to increase in clock speed from 100 to an astounding 233 MHz (megahertz). This was the original Pentium. As time went on, Pentium Pros (P6), Twos, and Threes were released, each having a more complicated core (to decode instructions and increase processor speed by eliminating slow aspects of the processor and bottlenecks) and being built on a more efficient process. In this essay, I will explain the effects of electrical resistance on the speed and heat dissipation of modern microprocessors, specifically the AMD Athlon XP Processor and the Intel Pentium 4 Processor.

Clock Speed: (How is it determined and how physics is involved)

There are two main factors that determine how physically fast a processor can run: process technology and its pipeline. A pipeline is made up of the steps that a processor takes to decode and basically process information. Typical pipelines are between four and seven stages long; modern “speed-demon” -type processors such as the P4 and Athlon have longer (20 in the case of the Pentium 4, and 10 for the Athlon). There are tradeoffs that engineers have to make with regards to their pipelines, simply because there are pros and cons to short/long pipelines. Long pipelines allow for substantial increases in the speed of a processor (MHz-wise), because as a processor goes through the steps in the pipeline, the slowest step is the bottleneck. With a longer pipeline, slow, complicated steps can be split up into several smaller steps which can be performed more quickly and simpler. On the other hand, if there is a mistake in the processor’s clock cycle, it must flush the information from the entire pipeline – the longer the pipeline is, the more time is lost in a full flush. Longer pipelines are also much more complicated to design and produce in the company’s Fabrication Plants. (When I say more complicated to produce, I mean that the company will have problems getting high yields out of the processor core… meaning that they wont get as many processors built that can withstand high speeds as they need). They also require a larger die size, which means more money per chip to produce. The Athlon holds down a ten stage pipeline, while the P4 has a 20 stage pipeline. Therefore, the P4 is less efficient per clock cycle then the Athlon XP, which explains why at the same clock speed the AMD chip will outperform the P4 by a large margin.

The other factor which helps to determine how fast a CPU will be able to scale in clock speed is the process technology on which it is produced. Intel began its ascendance to the now profitable kingship of micro processing (by revenue Goats, not performance), in about 1980 when they released the 8 MHz 8088. Two and a half years later, it released its 80286, and then in increments ranging from 5 years to six months released ever faster processors. (By comparison, the Pentium Pro, 2, and 3 are all 686, and the Pentium 4 is a 786 processor). Up until almost 2 years ago, all processors used aluminum interconnects in their processors. There was not a problem of the electrical resistance until processors started pushing GHz (gigahertz) clock speeds. The Intel Pentium 3 Coppermine processor was the first consumer processor to use copper interconnects, and this allowed it to scale almost 400 MHz faster then its aluminum predecessor. One can use simple mathematics to determine the percentage that it reduced friction between the transistors in the processor. For Example:b

Original P3 max speed: 733 MHz
P3 Coppermine max speed: 1133 MHz
1133/733 = 1.54
Therefore, the friction was reduced by over 54% by using copper in place of aluminum in the processor.

AMD as well moved to copper in its Athlon processor, and was able to achieve similar increases in clock speed.

Another aspect of process technology is the size of the interconnects and transistors. This is measured in either microns or nanometers. Modern processors are produced on a .13 micron (130 nanometer) process. Simply, the smaller the process, the lower the resistance between transistors and therefore the lower the heat dissipation of the processor is. It lowers the resistance for one reason: the distance between the transistors. The shorter the distance, the more quickly the electricity will reach its destination and more quickly a clock cycle will be completed. As well, the smaller the process is, the smaller the interconnects actually are, so the resistance is further decreased. One of the most limiting factors to a CPU’s speed is the heat created by core. If a processor does not have proper cooling at its rated speed, it will burn up in a matter of seconds (usually in under 6). As the process technology improves and shrinks, the amount of energy required to run the processor (and therefore the amount of energy dissipated when the processor is running) decreases greatly. For example, the 2 GHz Pentium 4 on a .18 micron (180nm) process requires over 75 watts of energy to run. As a contrast, the 2 GHz P4 on a .13 micron process requires just under 55 watts. It’s hard to imagine that so much heat could be generated by something that is just 131 mm2 in surface area. Another technique that is used to reduce friction is called SOI technology.

SOI (or Silicon on Insulator) technology is a process which changes the way that a basic transistor or “Gate” works. To understand how this works one must first realize that pure silicon does not conduct electricity. The impurities added to the silicon are what allow it to conduct in a processor. So I don’t take 15 pages explaining SOI I am just going to make it very simple. A transistor works like a power switch. When it is “on” there is energy flowing through it, and when it is off there is not. The amount of voltage sent to the transistor determines whether it is on or off (low voltage = gate closes, high voltage = gate opens; closed:off::open:on). Because there is impure silicon throughout the chip, the electricity can move through the transistor many different ways, of which a straight path is just one. SOI Technology puts a thin layer of pure (non-conducting) silicon between the Gate and the impure silicon that makes up the rest of the chip. Therefore, less voltage is required to switch a gate from off to on, and there is less resistance because the electricity is taking a much more direct path from one end to the other of the gate. Unfortunately SOI chips are very difficult to mass-produce and therefore are just starting to be adopted with AMD’s 8th Generation Hammer architecture. As an aside, IBM was the first to develop a chip using SOI, and is going to implement it in its upcoming Power PC 970 processor. The SOI process will allow chips to use less power (therefore dissipating less heat and removing a requirement for exotic cooling at exotic] clock speeds) and increase clock speed greatly (therefore charging vast amounts of money for very little silicon). Just one more way of using physics to make money :).

All of this sums up one thing: process technology is key in the design and production of processors, and almost every consideration related to process technology is affected by physics. The resistance between transistors is due to friction and therefore creates heat. If there is too much heat, the processor doesn’t work. These concerns make up most of the thinking and devising that engineers have to do when they are creating a new architecture or optimizing an older one. Physics is in everything.

One more thing; assuming that the move from .18 to .13 micron technology will decrease the resistance by another (about) 50%, we can assume that the P4 will reach almost 3.2 GHz, and that the Athlon XP will reach nearly 2.6 GHz. So there you have it… no coefficients of friction involved, but I was still able to use math to do it. Its quite amusing to me that the roadmaps for the introduction of the .13 micron chips from AMD and Intel correspond to the 3.2 max for the p4 and the 2.6 max for the Athlon roughly.

Update: Wow was I wrong about that Athlon XP speed...