TI envisions phone becoming central repository for apps, media and info

SanDisk senior vice president of memory design Khandker Quader doesn’t see memory constraints as an issue in the future. Like any other silicon process, Flash memory is subject to Moore’s Law, meaning the capacity will double while cost remains constant every one and a half years. The current Nand-based flash technologies will probably reach their apex in 2012 with 64 GB of capacity on a single chip, Quader said, but other technologies are lining up to replace it. One promising line of research is 3D read-write technology, a three-dimensional matrix storage technology akin to building new levels on a one-story chip.

“You’re creating arrays of memory, and you’re stacking them,” Quader said. “You can have very high capacity in a very small chip area, giving you high capacity memory.” How high? Quader said by merely extending Moore’s law linearly to 2025, memory capacity on a single chip would be slightly more than 2 terabytes. To put that in perspective, that’s 128 times the memory capacity of the iPhone 3G.

CENTRALIZED VS. DISTRIBUTED

Localizing that much information on a personal device seems to run counter to the trends in both the Internet and the wireless industry. Future wireless networks will not only have far greater capacity, they’ll have new topologies and reach new pinnacles in spectral re-use, giving individual users far more access to dedicated bandwidth. Devices won’t just be connected to the network and the Internet; they’ll share information with a multitude of other devices, sensors and even ordinary household objects. (See the previous I-feature stories Shrinking Cells and Narrowing beams and Frolicking in the Social Grass.) The trend in mobile and data communications is toward distributed information and computing, not centralized.

Goren isn’t discounting these future distributed models, but he also thinks those models fail to account for one of the basic principles of wireless communication, the scarcity of spectrum. Wireless networks will get faster, pipes will get fatter and network topologies will become denser, Goren said, but applications’ hunger for bandwidth will scale along with them. And while processing power and storage capacity will grow extraordinarily in 15 years, the amount of identifiable new spectrum won’t increase at the same pace.

“We have to keep in mind the most expensive resource in the network is spectrum,” Goren said. “Compute power is going up exponentially, and it is still much cheaper than spectrum. Yes, a fast network is important. A wide pipe is important. But downloading a 1080p movie—even downloading the compressed video--is still going to fill up the network.”