Intel’s 45nm Penryn: Cool Process but Ho-Hum Processor

Implications: Intel's turned its PR apparatus on high today, trumpeting the launch of its first 45nm processors, code-named Penryn. The company can certainly claim credit as the first CPU vendor to ship 45nm products, but the products it launched now are merely warmed over versions of the mature 65nm designs it introduced last year. The good news is that these chips work in the same desktop, notebook and server platforms as the models they replace. This simplifies the qualification process for OEM system suppliers and end users. The bad news is that the new process improves the new chips' performance by ten to fifteen percent on average -- hardly anything to get excited about. We don't expect these new chips to have much impact on the competitive positions of Intel and AMD, but the new chips are roughly 30 percent smaller than the ones they replace, and thus will indirectly increase Intel's production capacity by about a third. This explains why Intel recently trimmed its CAPEX forecast.

Analysis: Intel's move to 45nm production has far more potential than its initial products suggest. Some assume that the biennial migration to smaller and smaller process geometries (generally referred to as "Moore's Law") results from some act of congress, but in fact it is based on huge amounts of R&D by device and equipment manufacturers. For several decades, the challenges to Moore's Law lay primarily in lithography (i.e., the process of transferring the chip's features onto the silicon wafer), but more recently, scientists have had to introduce new materials into the process in order to make the smaller transistors operate at higher speeds.

Intel's new 45nm process addresses gate leakage, a particularly thorny problem. For over 40 years, manufacturers have built transistor gates using a combination of polysilicon and silicon dioxide. Each reduction in transistor dimensions was accompanied by a reduction in gate oxide thickness that improved transistor performance, but those gates had gotten down to about five atoms in thickness, and couldn't be made any thinner. Intel figured out how to use a hafnium-based compound instead of silicon dioxide and a metal gate structure in lieu of polysilicon. This new formulation (known as "high K Metal Gate") should suffice for several generations of CMOS technology and thus allows the industry to keep up with the Moore's law projections.

Although Intel is the first manufacturer to move to a high K Metal Gate technology, others will follow over the next few years. IBM has demonstrated a variation on Intel's scheme which it claims will reduce gate leakage by a factor of ten over Intel's approach, and which is easier to implement. Easy or not, IBM (and by association, AMD, its R&D partner) doesn't plan to include high K in its production processes until 2009 at the earliest.

Given that the Penryn processors coming off Intel's new lines are variations of the 65nm Merom (Core 2 Duo) processors Intel launched in 2006, Intel's competitive position vis-a-vis AMD won't change very much. With its Barcelona launch, AMD narrowed the performance gap Intel had opened up with Merom. Penryn has only a minor impact on that gap. Of course, if AMD cannot deliver on its promised 2.5GHz Barcelona chips, or next year's 45nm Shanghai chips, then Intel will extend its competitive advantage. Execution matters!

While Penryn may fall into the so-what category, Intel's next generation of 45nm CPUs, code-named Nehalem, are anything but boring. With Nehalem, Intel will finally move beyond the front-side bus architecture that has been a fundamental element of its designs for over two decades. It will also move to a "native" quad-core arrangement like the one used in AMD's Barcelona. Nehalem should offer a step-function improvement in CPU and system performance, and offer users a much bigger bang for their bucks.