REGISTER | LOGIN
Breaking News
News & Analysis

AI Tapped to Improve Design

Electronics center explores machine learning
2/3/2017 00:00 AM EST
14 comments
NO RATINGS
1 saves
More Related Links
View Comments: Newest First | Oldest First | Threaded View
Page 1 / 2   >   >>
Intellijoule
User Rank
Author
Re: A more disconcerting implication
Intellijoule   2/11/2017 8:40:42 PM
NO RATINGS
Note this: To synthesize a good secure kernel start with GEMSOS security kernel from Aesec.com.  Their Gemini Computer kernel is in the NSA and DOD machines for the many years. This might have changed, The GEMSOS operating system with its security kernel has never been cracked by any hackers known, worldwide. GEMSOS can be ported with simplifications to run on the new Intel Joule. No major firm has yet adopted the NSA Common Criteria recent rules. Dr. Schell (MIT and former NSA Deputy director) tells us that methods for starting a secure kernel design in the public domain and on Richard Stillman's GNU.org site. GEMSOS is done and running on DOD machines today but no ARM devices are permitted. Intel and AMD dominate the x86 platforms that make secure kernels truly possible by allowing hardwired kernels with ring and segmentation with gating between the rings and segments allowed by the kernel. UNIX comes to mind.

Intellijoule
User Rank
Author
Re: machine learning
Intellijoule   2/11/2017 8:23:54 PM
NO RATINGS
Our firm, Intellijoule.com, started a project of synthesizing secure kernels for IoT processors using an LSTM RNN, and convolutional RNN.  Developing a good kernel that is NSA Common Criteria Compliant can result in many thousands of lines of code in C or python. The cost of developing a secure kernel is terribly high. We are interested in anyone that would like to partner on this effort. NSA knows that ARM-based IoT controllers cannot be truly secure.  Noo ARM processor in NSA or DOD systems. 

 

 

alex_m1
User Rank
Author
Re: ...
alex_m1   2/8/2017 5:21:23 PM
NO RATINGS
In the long term, sure AI would out-compete us all. But in the short term, a growing design producitivity means greatly reduced chip NRE and more new designs so it may balance out.

realjjj
User Rank
CEO
Re: ...
realjjj   2/6/2017 7:42:22 PM
NO RATINGS
Even if costs would be high, scale solves that as the tool would have a broad customer base.

Wish the effort was open source (assuming it isn't), but maybe that day arrives at some point too.

resistion
User Rank
Author
Re: ...
resistion   2/6/2017 7:01:41 PM
NO RATINGS
On the other hand, AI costs might be underestimated, depending on the sophistication. For example, machine training would probably have to be as good as or better than training an entire team of engineers. Then there are the usual nuisances. For example, I just experienced problems logging onto the server. It could be crippling if everything depended on that.

realjjj
User Rank
CEO
Re: You are not get automated out of jobs
realjjj   2/6/2017 8:55:04 AM
NO RATINGS
"I think you might be guilty of confusing productivity with creativity"

That's not very nice or true.

"Paul Franson, a professor at NCSU, said that one student has already used machine learning to reduce iterations in chip routing from 20 to four."

Is that a substantial increase in labour productivity, is it significant? After all, the entire point of creating the tool is for it to be significant. How much and how fast would productivity increase in IC design is debatable and a bit early to pinpoint.

To fully replace the human, that's another matter and your point relates to that.

resistion
User Rank
Author
Re: You are not get automated out of jobs
resistion   2/6/2017 8:36:30 AM
NO RATINGS
I would guess a general division of labor could be machines follow rules and humans deal with exceptions.

Ron Neale
User Rank
Author
Re: You are not get automated out of jobs
Ron Neale   2/6/2017 6:43:08 AM
NO RATINGS
T the three Rs and Dennis ) I think you might be guilty of confusing productivity with creativity, successful engineering and engineers use a combination of learning and creativity. The role of AI will be significant when the mechanism of creativity is understood and implemented in hardware. Keeping track of a complex set of design rules and the constraints of simulations is a relatively simple role for modern computer systems and hardly represents AI.
To move the discussion onto my favourite subject emerging memory devices I happen to believe some of the reliability defects and randomness observed in the characteristics of some emerging memory technologies might hold the key to creative AI and creative machines. You will find a brief outline of my view in PhysicsWorld Nov 2016, pp 16 and a comment Physicsworld Jan 2017.pp26.
(physicsworld.com)

realjjj
User Rank
CEO
Re: You are not get automated out of jobs
realjjj   2/6/2017 5:56:10 AM
NO RATINGS
Any tool that greatly increases labor productivity (output per work hours), takes jobs away from people. The tractor hasn't fully replaced humans, but it had quite an impact on agriculture.

Machines used to only compete with our muscles, computing changes that and machines start to also compete with our brains. It's been slow so far but machine learning greatly accelerates the shift.

If today you need X engineers working for Y hours to complete a certain task and one such tool would allow for the task to be completed in XY/5, then 80% of those engineers lose their jobs - to simplify it a bit.

When increases in productivity are mild, society can keep up with it. When the increases are exponential and in both areas, muscle and brain power, there is no escape. AI enables supply to exceed demand, work hours wise. That's the crucial difference between today and the industrial revolution, we can't do other jobs anymore and human work will tend towards becoming obsolete.

In this particular case, it's fine medium term as costs are too high and, even if some lose their jobs, there is a substantial social and economic benefit. Long term it's the same as in any other area, fine if politicians do their jobs, it's a simple problem of supply and demand balance with supply being easy to control.

rick merritt
User Rank
Author
You are not get automated out of jobs
rick merritt   2/6/2017 1:35:20 AM
NO RATINGS
No one at the panel foresees AI desinging chips for engineers, just helping them with the thorniest tasks, like tracking the growing number of design rules/constraints and giving faster feedbaqck on optimal routing paths, etc.

 

Page 1 / 2   >   >>
Most Recent Comments

Datasheets.com Parts Search

185 million searchable parts
(please enter a part number or hit search to begin)
Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed