On Wednesday (May 4), Intel announced another milestone in semiconductor technology as they demonstrated a production-ready 3-D transistor technology for 22 nm called Tri-Gate.
With chips for servers, desktops and laptops running Windows, Intel drove home the point both that their 22-nm Tri-Gate devices are viable and that they can be tailored to suit the power versus performance tradeoffs for a range of applications. The 22-nm devices are shrinks of the Sandy Bridge microarchitecture known as Ivy Bridge in the "tick" side of the product introduction cycle that Intel has delivered six times going back to 65 nm.
The Tri-Gate announcement ended the speculation about the direction Intel would take for 22 nm. Mark Bohr's analyst briefings prior to the announcement explained why Tri-Gate is the right choice for 22-nm. Microprocessors are sampling now with delivery to OEMs slated for the end of this year. At least one Wall Street analyst may have this confused as this quote appeared yesterday:
"This architecture will separate the men from the boys. No one else has a tri-gate transistor in volume production."
Volume production at the D1D fab on Oregon will ramp up next year.
The Tri-Gate transistor is known outside Intel as a FinFet because the silicon channel is akin to a fin jutting up from the semiconductor substrate. MugFET is another term for this type of transistor, a shortened form of "multi-gate." A little less specific than Intel's name, but I was really pulling for MugFET to get wider adoption. I guess I have no future in marketing.
Intel will not yet disclose additional details on either the structure or specific performance numbers and did not give a timeframe for any such announcement, but the materials supplied to the media provided several relative performance figures.
The first 3-D transistors (if you exclude recessed channel DRAM array transistors) appearing at 22-nm will have a big impact on the industry. As Mark Bohr pointed out, 22-nm Tri-Gate transistor performance exceeds the capabilities of present and future SOI technology.
Comparing other technology options for 22-nm, the best technology platform needs to provide depleted channel devices. Compared to their 32-nm bulk technology, Intel's fully depleted Tri-Gate transistor provides 18 percent faster devices at operating voltages suitable for desktop and server chips.
But in a clear sign of Intel's desire to break into the ultraportable space occupied by tablets and smartphones, they highlighted the 37 percent speed improvement at lower operating voltages suited to devices aggressively marketed not only on time between battery charges but also capability to perform intense computing functions.
The 22 nm release is still early, even by Intel's schedule. It could mean 10 nm, there will be no silicon channel, not even fully depleted. Otherwise, it would be preferable to start to offer finfet at 14 nm.
It is true that Tri-Gate will be giving better performance compared to existing 32nm Technology getting manufactured upon, but still Intel will have to come out with some better processor variant that will be able to compete ARM IPs. As the market is getting developed over ARM these days within a year or two the scenario will get changed completely and it will become hard to find Intel processors in mobile devices.
Here is very cool youtube video
Link for more info:
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.