Highlights of Intel's Justin Rattner at CHM: Oct 8, 2013

The event was in the form of an interview/conversation led by CHM CEO/moderator John Hollar.  Mr. Rattner spent a lot of time discussing his early life at Hollywood High School and how he got involved in electronics and as a EE student at Cornell University.  It was less about Intel's early history.  However, here are a few quick takes on Intel from Mr. Rattner:

On Gordon Moore:  "His ability to pick the technology that was going to win in the end. Gordon managed all the twists and turns of semiconductor technology- from P channel to N channel to CMOS."  In the mid 1970s, he allocated 7% of Intel's R&D budget on CAD, recognizing that LSIs would be too complicated to design using a pocket calculator.

On Bob Noyce:  "Very approachable; couldn't spend more than 5 minutes with him without getting excited about whatever he was thinking about."

On High Performance Computing:  "The early supercomputers were experimental parallel machines.  We barely knew how to program them.  There was a constant sense of discovery of what was going on. We were immersed in a research environment, even though our original mission was to provide tools for researchers."  Rattner founded the computer server labs at Intel.

"We first had to convince people with money that flashing up a bunch of microprocessors was THE path to high performance computing."  Rattner was the only one who spoke of the "attack of the killer micros," at a conference in upstate New York. "Tide turned in 1991 with the big Delta machines, which beat the Cray XMT supercomputer in a performance test.  "A microprocessor machine never lost its title as fastest supercomputer, with possible exception of a Japanese built machine," he said.

"We had to tackle all manner of fundamental arithmetic, algorithms, scheduling, communications, etc.  There wasn't much off-the-shelf knowledge available at that time. So Intel put together a first rate team of numerical analysts, computational scientists and algorithmic experts that worked closely with customers like Argonne National Labs."

Intel had a 5 to 7 year view and could practically tell you what the performance would be."  In the early 1990s, the big parallel machines eclipsed vector machines in performance.  However, there was criticism that the parallel machines were too hard to program and the algorithms themselves weren't that efficient (compared to those running on vector machines).  It was difficult to get Federal government funding for supercomputer research, he said.

On Moore's Law and how it's going to play out:  If it is independent of the underlying technology, it will last for many decades in the future. If it pertains to a specific technology, it has already ended, e.g. Intel's Silicon Gate semiconductor process which was replaced by several others over the last few decades.  Intel is now looking at 'non-charge effect" silicon devices and others that deal with quantum effects.  CMOS is pretty good in efficiently pushing charge around, but Intel is looking at alternative quantum effect technologies for the future. Rattner thinks that the opportunity is there for many new device architectures with alternative quantum effects.  Will we still call it Moore's Law in 2020, when transistors don't operate on the same fundamental physics as they do today?  Molecular and nano-tube transistors are examples of future material science structures that will be manufactured in many different ways.  Silicon will still play a role as a substrate, but other materials may be overlaid on top of it.  You will probably see new semiconductor technologies on memory devices, such as flash memory replacements, he said.

Intel Labs has been working on Digital Radio for quite some time.  Rattner said Analog Radio's don't scale, i.e. Moore's Law doesn't apply. Digital Radio returns to the fundamental mathematics of communications (i.e. information theory).  Intel is focused on Digital Radio - both reception and transmission- as a "computational problem" that's built using high end microprocessors.  Sadly, Rattner didn't give a progress report on Intel's research in this area, which is still not a commercial technology or product (see Reference below from 2009).

Photography via image processors is another interesting research area.  "Bio" (biology?) is also on Rattner's list of future digital research projects.  Image processors in smart phones are getting more and more powerful with each successive generation, he said.

Unlike conventional video that's transmitted over 3G/4G wireless networks that operate "open loop," video aware wireless networks would operate  "closed loop," which will provide a better user experience.   The basic concept is for the receiver to monitor and measure image quality in real time and send back status information over the communications channel to the transmitter.  Such that at each stage of the channel, the system can determine the proper video resolution with a higher or lower frame rate.

Intel's work with Stephen Hawkings for a new User Interface to enable him to communicate better.  Hawking is paralyzed due to a degenerative disease called amyotrophic lateral sclerosis (ALS). He uses small muscle twitches in his face to select words on a custom computer system so he can communicate. Sadly, his condition has progressed to the point where he can only manage roughly one word per minute. After meeting with Hawking himself, Intel’s Rattner is spearheading a project to improve Hawking’s computer system, and allow for an increase in words per minute.

http://www.extremetech.com/extreme/146269-intel-working-on-a-new-system-to-boost-stephen-hawkings-typing-speed-by-10x

 Note:  CHM CEO/Moderator John Hollar emailed this author, stating  that Intel's work with Hawking was probably the most interesting part of this conversation he had with Rattner.

References:

Event Video at http://www.youtube.com/watch?v=C8HbjTACgp0&feature=c4-overview&list=UUHDr4RtxwA1KqKGwxgdK4Vg   

Could it really be true: All Digital Radios from Intel?

http://4g360.com/profiles/blogs/could-it-really-be-true-all