IBM plans to spend $3 billion in the next five years on research and development for computer chips, both to stretch the limits of conventional semiconductors and to hasten the commercialization of exotic new designs.
The chip program, announced Wednesday, will be among the largest research initiatives at IBM, which spends about $6 billion a year on research and development.
“We’re investing to really push the frontiers” of chip technology, said John E. Kelly, senior vice president and director of research at IBM.
The effort, Kelley said, will have two main goals. The first is to wring further improvements from silicon chip technology by shrinking the tiny circuits from today’s 22 nanometers down to 7 nanometers, a few atoms wide. The second is to accelerate progress on novel and promising, if unproved, approaches – designs that employ quantum physics, carbon nanotubes and chips inspired by the brain, called neuromorphic chips.
To step up its chip research, Kelly said, IBM will be hiring more scientists and investing in industry partnerships and academic collaborations.
The company’s research announcement is also an emphatic statement about IBM’s unfolding strategy in the chip business. It will invest heavily in research, pursue streams of licensing revenue, and yet leave the manufacturing to others.
For several months, IBM has been seeking to sell its chip-manufacturing operations, said a person who has been briefed on the talks. The most likely buyer is GlobalFoundries, a large contract chip manufacturer, the person said, for a price of probably less than $2 billion.
Selling its semiconductor-manufacturing operations would mesh with IBM’s well-established course of shedding hardware businesses with lower profit margins. Over the years, the company has retreated from printers, personal computers and disk drives, as well as its pending sale of smaller server computers. Designing chips and licensing technology, without manufacturing, can be lucrative. Qualcomm, a maker of chips for mobile devices, is the showcase example.
IBM’s semiconductor business, according to A.M. Sacconaghi, an analyst at Bernstein Research, generates about $2.5 billion in annual revenue, including the chips used in IBM computers and sales to other companies. Last year, Sacconaghi estimated, the business had an operating loss of roughly $500 million. But the chip unit, he said, is responsible for contributing licensing revenue from its intellectual property of about $400 million a year, which, if included, would bring the operating loss down to $100 million.
IBM’s research agenda is a recognition that the end of the silicon era of computing, using complementary metal-oxide semiconductor technology, or CMOS, is finally on the horizon. As the circuit widths get ever-smaller, heat dissipation increases and chips eventually become unstable. By the end of the decade, Kelly said, the conventional technology will be hitting “not just speed bumps, but a wall.”
Richard Doherty, an analyst at Envisioneering, a technology research firm, said, “IBM is not giving up on silicon, but it is saying it’s time to place an array of bets, and to move beyond silicon.”