How Small Can You Go? IBM Heads for 22nm


Source: Reuters

Semiconductor technology today is in the 65- to 45-nanometer (nm) range, with Intel targeting 32nm for next year. IBM is going one step further, announcing plans for its move to 22nm.

Both companies have similar road maps. Intel intends further die-size shrinks, from 45nm to 32nm, 22nm, 14 or 15nm and eventually 10nm. It just hasn’t said when exactly it will get there, beyond 32nm in 2009, nor has it said how. Building a chip at the atomic level isn’t like assembling a bike.

IBM (NYSE: IBM) today said it would partner with Mentor Graphics to use next-generation computational lithography (CL) software for creating and manufacturing the 22nm semiconductors, which will ship in either late 2011 or early 2012. While that’s a way off, IBM has done some product runs and has some samples in-house for experimentation and evaluation.

Current lithography methods simply can’t make the layers of a chip at 22nm due to the limitations on the process. In other words, they can’t make a chip that small. Kevin Warren, director of design and technology integration for IBM Semiconductor, said the company will use the same equipment to make 22nm parts as it will to make 32nm parts, which begin production in 2009.

“There really is not another significant improvement in lens or numerical aperture on the horizon,” he told “Other aspects of the equation that govern scale have reached their theoretical limits. The industry has acknowledged this now.”

So the solution is to use existing lithographic tools with massively parallel computation to make smaller products. Computational scaling is an attempt to model and optimize the entire process flow for each new design. Like Intel (NASDAQ: INTC), IBM is moving to high-k metal gate design, which will be used in 32nm and 22nm processors.

For end users, this means Moore’s Law stays on track, and they continue to see commensurate density scaling they are used to seeing. “We will be integrating more stuff on chip; the chips will be closer to the memory for faster performance. The parts should be cheaper because we’re taking less silicon area to do the work,” Warren said.

He stressed that this won’t just benefit IBM System 595 servers running Power 6 processors but the smallest of devices, too. “It’s very important for things like cell phones in addition to servers,” Warren said. “It all depends on that same gambit of being able to add more and reduce cost.”

IBM fully intends to go to 15nm using this technique, at the very least. Whether it can get to 10nm is another story.

News Around the Web