Last week Intel described details from their forthcoming 45 nm manufacturing process and some details of the first 45 nm processors, codenamed Penryn. In this article we will explain what is new on Intel’s 45 nm process and Penryn CPUs.
The main problem in shifting to a smaller manufacturing technology – i.e., using smaller transistors – is leakage current. While on old CPUs using bigger manufacturing technology leakage isn’t such a big issue, when we are talking about CPUs with very small transistors, leakage can represent not only a big waste of power but also overheating. On the other hand, smaller transistors translate into a faster switching speed – i.e., higher performance.
Transistors inside the CPU are traditionally built using a polysilicon gate electrode and a silicon oxide gate dielectric, a material referred as a “low-K” material, meaning a relatively high leakage current.
For years the Holy Graal in the CPU industry has been the development of a high-K dielectric material to be used on the transistor gate. This material would present a far lower leakage current compared to a low-K material like silicon oxide.
What Intel has announced last week is that they developed such material (Hafnium-based, a chemical material on the same column of Zirconium and Titanium on the periodic table) and is using it on their 45 nm manufacturing process. Also, the gate electrode has been changed from polysilicon to metal (Intel didn’t say which material is used). The combination of a metallic gate electrode and a high-K dielectric material produces a higher current when the transistor is “on” and a lower current when the transistor is “off” – translating into a lower leakage current.
What is interesting is that this technology has been used for one year now – Intel’s 45-nm static RAM chips use this technology, but Intel didn’t disclose this when their 45-nm SRAM chips were released.
The main advantages brought by the high-K + metal gate transistors used on Intel’s new 45 nm manufacturing process compared to the current 65 nm manufacturing process are:
- Approximately 2x improvement in transistor density. This means that Intel can fit more transistors in the same area. Thus they can build smaller chips or keep the current chip size but putting more transistors in it.
- Approximately 30% reduction in transistor switching power.
- Over 20% reduction in transistor switching time (i.e., transistors are at least 20% faster, which can be translated in faster chips) or over 5x reduction in leakage current between the transistor source and drain.
- Over 10x leakage reduction on the transistor gate.
Intel also has announced that they will keep their roadmap and will be announcing their 32 nm manufacturing process in 2009 and their 22 nm manufacturing process in 2011.
The first CPU generation using the new 45-nm manufacturing process is called Penryn.
Penryn isn’t the codename of a specific processor, but the codename of the 45 nm core that will be used by mobile, desktop and server CPUs.
Intel didn’t disclosure a lot about these new CPUs. All they said was that they have prototypes right now running several different operating systems, there will be dual-core and quad-core versions (with 410 million transistors and 820 million transistors, respectively), they will use a new SSE instruction set, called SSE4 (that will bring 47 new SSE instructions to the CPU), larger caches and new microarchitecture features.
Read our article Penryn Core New Features for a complete list of new features that will be brought by Penryn core.