facebook rss twitter

Intel divulges details on 22nm chips. Pulls in production for ultra-low-power models

by Tarinder Sandhu on 22 September 2009, 20:06

Tags: Intel (NASDAQ:INTC)

Quick Link: HEXUS.net/qat4e

Add to My Vault: x

Following on from Paul Otellini's keynote speech, where he commented upon Intel's upcoming 22nm manufacturing process, senior fellow Mark Bohr provided further details on the underlying technology and implementation that will power Intel's chips in 2011, and beyond.

Bohr commented that Intel has built a 22nm third-generation high-k metal-gate SRAM test circuit - a pragmatic proof of technology - based on a 364-million-bit array. Packing in 2.9bn transistors at a twice-thick density compared to 32nm, Intel is building the test chips at the D1D fabrication facility in Oregon, which also serves as a production factory for recently-certified 32nm Westmere chips that will start rolling off the lines in Q4 2009. The next iteration of process Intel's tick-tock architecture, Sandybridge, based on 32nm, is due to into production in late 2010, followed by the 22nm parts in H2 2011.

The race between Intel's manufacturing and GLOBALFOUBDRIES appears to be hotting up. Both companies are slated to bring 22nm products to market within nine months of each other, according to the latest roadmaps, and AMD may well debut it with the much-vaunted Bulldozer core.

Bohr further mentioned that no major lithography changes were needed from the shift to 22nm: they'll be using the 193nm immersion process.

You may expect that Intel has one process for all its chips. Rather, they're split into two processes. The first is for traditional CPUs - server, client, mobile - dubbed P1270 and based on 12-inch wafers. The second process is for system-on-a-chip (SoC) parts - P1271 - and it's optimised for low-power usage and aimed toward the low-power segment currently characterised by the nascent Atom chips that power the majority of netbooks.

It's no secret that Intel wants to introduce architectures that reduce idle and load power-draw to a level that's competitive with industry leader ARM. The reason for doing so lies with exploiting the multi-billion-unit mobile handset and embedded market that is dominated by designs from the UK company.

To that effect, Intel is pulling in the schedule that dictates when the two processes are released. 45nm SoC chips followed on a year after their higher-power counterparts. Bohr commented that the 32nm SoC's release - Medfield core, most likely - will be reined in to six months behind the client and mobile parts, and 22nm SoCs will follow just three months after their counterparts.

It's telling that Intel is putting so much focus on SoC. There's an inextricable trend towards low-power computing. What remains to be seen is whether the archaic Intel Architecture (IA) is best-suited to efficient processing at the sub-1W level.

HEXUS IDF 2009 coverage

click for more IDF 2009 coverage




HEXUS Forums :: 0 Comments

Login with Forum Account

Don't have an account? Register today!
Log in to be the first to comment!