Intel's Crystal Ball Sees 'Humanized' Computing By 2015Intel's Crystal Ball Sees 'Humanized' Computing By 2015

Intel presented a long-range look into the future at its Developers Forum Thursday, predicting that by 2015, people will be working with "humanized" computing frameworks that operate intuitively on voice and vision commands, and that new silicon strategies could bring about single chips with hundreds of processing cores and devices that incorporate multiple types of functions.

"We need a world where technology becomes more natural," Justin Rattner, Intel senior fellow and director of the corporate technology group, told forum attendees. "Instead of the clunky interfaces of today, we need computing interactions to be more like a conversation."

The advances could provide Star Trek-like real-time translation of foreign languages in videoconferencing applications, voice-command file searching, and interactive movies in which the viewer can affect the outcome of the plot.

"Computing needs to become less prescriptive, and more predictive," Rattner said.

Enabling the technology of the future will require accelerating of some of the ongoing themes that are emerging now at Intel, he said. Dual-core systems that are hitting the market this year will evolve into four-core systems in the near future and eventually into processors with hundreds of cores and thousands of individual threads.

Making those "superprocessors" work more effectively and automatically within various computing platforms will evolve naturally as Intel continues to expand its "embedded IT" effort for improved hardware virtualization, self-configuration, and self-management, he said.

Steve Sullivan, a director of R&D at 3-D film animation company Industrial Light & Magic, said the move to highly parallel processing platforms could help reduce the tremendous server demands his company and others like it experience today.

IL&M, which is currently completing Star Wars Episode III: Revenge Of The Sith, uses farms of a 1,000 or more servers to complete graphics rendering, which requires massive computing power. Reducing the number of servers needed for tasks like that would require substantial increases in the real-time processing capabilities of processors.

To make more-powerful multicore processors will require not only increased parallelism inside the processor core, but also parallel programming capabilities, Rattner said.

Hardware virtualization, which is being introduced in some of Intel's newest processors, will need to expand to address not only the processor, but also storage, network, and display subsystems, he said.

As processors begin to incorporate dozens or more processing cores or elements within single silicon dies, semiconductor companies like Intel will begin to run into constraints getting information loaded and unloaded on to the devices because there are a limited number of interconnect wires connecting the die to the motherboards and between chips.

Two possible answers are 3-D stacking of separate die elements and wafer stacking, Rattner said. In wafer stacking, two individual silicon wafers would be bonded on top of each other, allowing direct attachment of thousands of pins. With die stacking, multiple silicon dies, including such elements as processors, different types of memory, and analog circuitry, would be placed on top of each other, six or seven layers deep, increasing connectivity four- or fivefold, Rattner says.

Both types of stacking would have trade-offs in manufacturability, yield, and cost, and both could potentially be used in manufacturing flows simultaneously to meet specific requirements, he said.

Meanwhile, Intel in February disclosed that its researchers had built a continuously shinning silicon laser that could eventually be used to replace copper wire with light connections. Rattner said the silicon laser could be used in chip-to-chip interconnections, backplanes, and data-center configurations.