45nm Transition

Intel will be rapidly ramping up the transition from the 65 nm process technology to 45 nm. A remarkable process since they have so many fabs worldwide. At the moment, 45 nm processors are already being fabricated at the D1D (Oregon) and Fab 32 (Arizona) fabs.

Soon, Intel will finish convert Fab 28 in Israel, followed by Fab 11X in New Mexico. These two fabs are the start of a rapid ramp up to 45 nm. Intel intends to reach the crossover point between shipments of 65 nm processors and 45 nm processors to occur in Q3'08.

High-K Metal Gate Transistors

As you all know by now, the 45 nm process technology is not merely about shrinking the transistors. The fabrication of the transistor gate insulators have not changed since the silicon transistor was invented, until now.

Intel will be replacing the silicon dioxide with hafnium as gate insulator in the new 45 nm processors. The use of this new high-k dielectric hafnium insulators will allow Intel to reduce power leakage and increase the gate switching speed (for higher clock speeds).

The new 45 nm processors will also be the first Intel processors to transition from low lead to 100% lead-free. Intel also intends to eliminate halogen from the packaging process by the end of 2008.

The Penryn Family

Pat then showed us what Intel has in store for us in the coming weeks. Starting with the November 12 launch, Intel will be steadily delivering members of the Penryn family. For more details on the specifications and pricing, you should take a look at our Intel 45nm Core 2 Desktop Processor Pre-Launch Update.

If you have already read it, you will note that Intel did not list what processors they will be launching in Q3'08. Although he did not come right out and say it, Pat G hinted that the new desktop processors to be announced for Q3'08 will support a higher FSB speed of 1600 MHz.

On the topic of the rather limited performance advantage we saw in the Intel 45nm Core 2 Desktop Processor Pre-Launch Update, he told us that the 45 nm server processors like Harpertown and Wolfdale-DP should see a much larger speed boost, due to the improved write-combining. He didn't go into the details, but he was probably referring to the MOVNTDQA instruction in the SSE4 which allows for up to 7.5X higher memory throughput for streaming loads.