..:: “Prescott” Processor Overview ::..
Before we delve into exactly what modifications Intel has made to the Pentium 4’s NetBurst microarchitecture, let’s take a look at the official specifications for the processor, along with taking a good look at the “Prescott” die. “Prescott,” Intel’s first microprocessor on their new 90nm manufacturing process features an incredible 125 million transistors packed into a die measuring 112mm2. In contrast to the “Northwood” core, “Prescott” contains roughly 70 million more transistors packed onto a die size roughly 11.8% smaller than that of “Northwood.” With the “Prescott” die, Intel has managed to more than double the amount of transistors utilized, and yet has also been able to shrink down the die size by roughly 12%. The bulk of these added transistors are due to the doubling of the L2 cache to 1MB. Intel has also doubled the L1 cache to 16KB. Needless to say, with all of these extra transistors, it seems that heat could be quite an important issue with “Prescott.” Speaking of which, let’s cover the main issue dealing with the “Prescott” core that has been contested for quite some time now, the official TDP rating.
For several months, rumors had it that Intel was having troubles with the move to the new 90nm manufacturing technique, and were also delaying launch to address some heat output issues that the “Prescott” core was having. Initial estimations by various news sources placed the TDP for the processor at a rating above the 100W mark. The official specifications as listed by Intel for both the 3.20GHz and 3.40GHz “Prescott” processors is 103W. The 3.00GHz and 2.80GHz versions fall under a TDP of 89W. The official VCore for “Prescott” is 1.25V – 1.40V, although our testing sample was rated for 1.36V. The Icc max rating for the 3.40GHz and 3.20GHz version is 91A, while the 3.00GHz and 2.80GHz versions are rated for 78A each. All of these ratings are quite a bit higher than even the 3.20GHz “Northwood” core is rated at.
Various independent testing from other media outlets has shown that the actual TDP for these processors might not be quite as high as these ratings given by Intel, although we must take into account that these ratings are typically a worst case scenario rating. From our testing, we have found that the increase in TDP is readily obvious in comparison to the “Northwood” core. We have experienced load and idle temperatures that are roughly 8-12% higher than those which we achieved on an identically cooled “Northwood” system, running at the same core frequency. As is stated by Intel, current Pentium 4 solutions will be able to support the “Prescott” core, although temperatures will indeed be higher than some would hope, so for those looking to overclock, high-end cooling is going to be a must-have if you choose to purchase a “Prescott” Pentium 4.
..:: “Prescott” Manufacturing Improvements ::..
Intel has also been able to make several advancements in manufacturing with their 90nm technology. With their 90nm technology, Intel now manufactures “Prescott” chips with seven layers of copper interconnects, whereas with the “older” 130nm technology they had six layers of copper interconnects. Intel has been able to keep the number of metal layers down to only seven with “Prescott,” quite a feat considering the transistor count, and the fact that this is still two layers less than that of AMD’s current solutions. The lower number of metal layers helps to reduce costs that are incurred during the manufacturing process, and also avoids any additional complexity to the manufacturing process. Intel has also made advancements when it comes to the lithography that they utilize when manufacturing these chips. This should come as no surprise as the drop to a 90nm manufacturing process requires better and more accurate lithography. The 90nm process uses 192nm lithography, while the 130nm process was limited to 248nm lithography.
Intel has also moved to a new low permittivity interlayer dielectric, Low K CDO, or Carbon-Doped Oxide material. In previous versions of their 130nm processor technology, Intel was utilizing a Silicon based interlayer dielectric. Typically, Carbon based Low K polymers yield a dielectric constant in the area of 2.1 to 2.6, with and increased or decreased value depending on whether it is fluorinated or nonfluorinated, whereas the older Silicon based polymer Intel used with their 130nm process had a dielectric constant of roughly 3.4 to 3.6. This new Low K CDO allows for faster electron flow for lessened signal delay times, reduced cross talk between interconnect layers, and less power loss due to the decrease in the dielectric constant.
The last main development that Intel has made with their new 90nm process technology deals with the idea of Strained Silicon. Strained Silicon allows for a rise in drive current for both NMOS and PMOS transistors which is of course, the technology that is utilized by Intel for their microprocessors. With Strained Silicon, Intel uses a very thin layer of single-crystal Silicon which features a stressed or strained lattice. Due to the state of the lattice on a molecular level, there is a lower resistance to electron flow, so the electrons are able to move faster through the lattice than they would with normal Silicon. This allows for an increase of the transistor switching speeds, increasing the top end speed of the chip as a whole. Intel claims that they have seen increase of up to 10% for NMOS transistors and 25% for PMOS transistors by using this Strained Silicon manufacturing technique.