Intel Plans Platform for 2015

Looking to the platforms of 2015, where the goal is for computing to be “less prescriptive and more predictive” Intel’s Justin Rattner, senior fellow and director of the corporate technology group offered a preview of some of the technologies now being explored.

“Microprocessors

take four to five years to develop, and if you don’t hit that design window you miss the cycle,” he said in his keynote address at the Intel Developers Forum this week covering the Santa Clara, Calif.-based company's research and development efforts. “That means you have to begin 10 years ahead. You literally have to imagine the future.”

Demonstrating Intel’s latest universal communicator research platform which now includes the company’s next generation PC processor for notebooks, code-named Yonah, Rattner showed how the device could now offer streaming video and could serve as a collaboration platform for voice over IP calls.

And Steve Sullivan, head of research and development at Industrial Light and Magic discussed how new generations of computing could help his company deliver computer generated images for the movie industry.

Those and other future applications are driving the continued progress for chips.  But the question remains whether or not new applications will require a brand new architecture.

“We’ve given that a lot of thought and agreed that answer is less of a technical answer and more of a biological answer because platforms and their ecosystems are very much like living organisms,” Rattner said.  In nature, while some radical mutations occur, the more successful species are those that undergo more modest changes in response to their environment, evolving over time, he added.

“The key to all of this is to never stop evolving,” Rattner said.  “If we stop we die. Evolution is a fundamental part of our existence.”

And the same is true for the Intel platform, which Rattner said has been evolving since the start. 

“We’ve created entire new species from that platform,” he said.  From the single processor Intel created servers, notebooks and entertainment devices, and Rattner said the company will continue with those evolutionary steps.

With that in mind, Rattner offered a glimpse at some of the “ingredient technologies” to get the platform to the next decade.  Some of those that Intel is investigating include autonomics or self-maintaining systems, low power, sensors, and machine learning.

Virtualization is another big one, according to Rattner, who said the technology will go beyond just what was demonstrated on processors at the Intel Developers Forum this week.  Future virtualization initiatives will focus on the larger problem – the entire system, allowing virtualization of storage and displays and other peripheral applications.

Parallel computing, a big theme at IDF this week as the company put its full weight behind future multi-core platforms, will continue to evolve as well.  Intel is looking beyond multi-core to many-core which Rattner said is 10s of cores or maybe even 100s of cores per die that will each be supporting multiple threads.

To achieve this companies must solve the overwhelming challenge of programming machines that contain thousands of processors.

Another daunting issue that will arrive as many-core computing arrives is the memory bandwidth challenge.  The raw processing power of all those cores push memory bandwidth requirements way up, Rattner said.

“We started to look at packaging alternatives as a way to address this,” he said.  One of the avenues of research is 3D stacking.  In wafer stacking Intel is investigating the creation of a die stack made of many die. On top of the die is a DRAM device, Rattner said.  This research is currently taking two wafers and bonding them together, making connections across the entire wafer. By stacking it would be possible to increase the number of pins from a few hundred to a million or even 10 million pins.

“That’s the kind of increase in memory to process connectivity that will be needed,” Rattner said.

An alternative calls for stacking of the individual die, he said, that offers the advantage of not having to match individual die sizes.

Today in the lab Intel is stacking as high as eight layers, however, only the base die is a full thickness die. The rest are thin die, Rattner said.  The answer to whether or not it is practical to stack more than eight as far as the manufacturing process is concerned is still a few years off, but Rattner said it is certainly technically possible.

Intel is also working on the bottleneck of chip-to-chip signaling, the impetus behind the company’s recent research into and announcement of the creation of the first continuous wave silicon laser. All of these investigations are part of Intel’s work to create the platform for 2015.

“We need to move in an evolutionary fashion,” Rattner said. “We are going to be taking the series of small steps that will get us to that destination over he next 5 to 10 years.”