Santa Clara (CA) - Intel shows off its tech horsepower, revealing more details about its upcoming 45 nm CPU generation: Penryn, a processor core that will replace the current 65 nm Core 2 processors, uses a new transistor technology to boost performance and reduce power consumption. The Penryn quad-core will carry a stunning 820 million transistors.
If you are trying to stay somewhat up to date with current computer technologies and run an Intel-based system, then you probably have updated recently to one of the firm's 65 nm offerings. But, at least for early adopters, this is old news, as the 45 nm generation is approaching quickly and we are not talking about some research product here anymore. 45 nm is a real technology.
One year ago, Intel announced the production of a 45 nm SRAM chip, a technology that is commonly used as a prototype platform for a new production process. Now, the company has produced the first 45 nm processors, code-named "Penryn" and claims that the chip is booting every popular operating system including the usual suspects Windows XP/Vista, MacOS and Linux.
The company told journalists on Friday that Penryn essentially will succeed the 65 nm Core 2 Duo (65 nm) Merom core, spinning off into mobile, desktop and volume server CPU derivates, just like its predecessor.
Penryn is not an entirely new processor, meaning that it will continue to use the architectural concept of Merom. While Intel said that Penryn will come with some architecture enhancements such as increased L2 cache size as well as Streaming SIMD4 (SSE4) extensions to speed up media applications, the other change relates to an increased clock speed and some thermal improvements.
410 million transistors and 6 MB L2 cache: The 45 nm Penryn processor core
On the surface, the 45 nm process allowed Intel to cut the space requirement of transistors roughly in half. A dual-core Penryn processor holds 410 million transistors, up from 293 million in Merom/Conroe/Woodcrest. Since Intel did not double the transistor count, it found additional space available on the (original Merom) die, which the company used to increase the L2 cache (on Penryn). While specifications were not revealed, sources told TG Daily that Penryn will integrate 6 MB L2 cache, up from 4 MB today.
A quad-core Penryn processor, which will be available in desktop and server flavors, will carry 820 million transistors, about half the transistor count of the current Itanium 2 processor with Montecito core. However, this is about 20 times the number of the transistors used in the first Pentium 4 processor with Willamette core, which was released as a 180 nm chip in 2000 (with 42 million transistors).
Intel will not counter AMD's approach to create a native, single-chip quad-core processor and continue to produce the quad-core Penryn in a dual-die, multi-chip package instead. Intel representatives told us that the company sees little reason to transition to a single-die solution at this time, as its current multi-chip approach with Kentsfield and Clovertown is successful and is achieving high yields. Using a multi-die solution, Intel is able to be more selective about the dies for quad-cores. Single-die quad-cores are not expected to arrive until the introduction of the Nehalem core, which will succeed Penryn with a completely new 45 nm architecture in the second half of 2008.
Read on the next page: High five for high-k: Penryn gets a new transistor technology
What makes Penryn especially interesting is its heavily modified 45 nm production process. Shrinking transistors have pushed the semiconductor industry deeper into the problem of leaking current. Still, Intel claims that it has found a way to scale transistors down to the 45 nm level, increase clock speed and reduce power consumption at the same time.
Typically, companies are squeezing out every bit of scalability out of existing processes to make the economics work. Apparently, technology and economics did not line up favorably anymore with 45 nm and Intel decided that it was time to ditch the old, 40-year-old general transistor design and use a new technology.
Compared to traditional transistors, which use a low resistance layer, a polysilicon gate electrode, a silicon-dioxide gate dielectric and the silicon substrate channel, the new transistor technology replaces the gate electrode polysilicon gate with a metal gate and the silicon-dioxide gate dielectric with a high-k dielectric. The term "high-k" refers to a "high" dielectric constant relative to the silicon dioxide.
Intel is very secretive about the materials it uses in this new high-k transistor. The company told journalists that the high-k gate is based on the element Hafnium, but the company declined to mention details about the two metals it uses for the metal gate. "There are hundreds of possible combinations," senior fellow Mark Bohr said, "and it is major achievement to have found this combination." Intel will use the new transistor as a competitive advantage and stated that it does not believe that other companies will be able to implement this technology until the 32 nm generation "or later".
Intels' new transistor. See our image gallery for details.
The new material choices have become necessary as they apparently allow Intel to better control current flow and reduce leakage. Especially the silicon-dioxide has become extremely thin over the past shrinks and evolved into a key concern of unnecessary and uncontrollable power consumption. The high-k gate, according to Intel, increases current flow when the transistor is in an "on" state, but decreases the leakage when the transistor is "off" by a factor of 10x.
Also, Intel claims that the metal gate and the high-k gate combination enabled the company to reduce the transistor switching power by 30%. Higher efficiency, of course, means that engineers can play with clock speed dial to yield either greater power consumption or more performance. We were told that the new material can bring up to 20% higher transistor switching (clock) speed or a 5x (!) reduction in source-to-drain leakage power when compared to 65 nm transistors running at the same clock speed.
Intel aims to maintain the current power envelopes (35 watt and lower on the notebook), so we can expect substantially higher clock speeds with Penryn. So, how much clock speed are we talking about? Intel declined to comment, but engineers familiar with the topic indicated that the data provided by Intel would mean more than 3.3 GHz for high-end desktop computers and more than 2.5 GHz on the notebook side. Keep in mind that Intel sources told us about a year ago that Conroe, the desktop version of Merom, was running far below its theoretical potential. 2.93 GHz is still a somewhat conservative clock speed as our sources indicated that Conroe CPUs were running at 4 GHz in Intel's labs.
Another question is the production process itself. New technologies require new processes that always bring risks, especially for a product that is positioned to cover the mobile, desktop and volume segments. Intel mentioned that the metal gate and high-k dielectric add "a few percent" to the production cost bottom line and can be integrated into existing manufacturing processes without the need for major modifications. However, the company mentioned that the high-k gate oxide requires an "atomic layer deposition technique" - which means that the high-k layer is built with one atomic layer at a time.
Production of 45 nm Penryn processors is scheduled to begin in the second half of 2007 in Intel's D1D factory in Hillsboro, Oregon. Within weeks, Fab 32 in Chandler, Arizona will go online, followed by Fab 28 in Israel.