Mistakes in silicon chips to help boost computer power (BBC)

[eppur_se_muova]

By Mark Ward
Technology correspondent, BBC News

Silicon chips that are allowed to make mistakes could help ensure computers continue to get more powerful, say US researchers.
***
"To ensure correct operation you are purposefully running the chips at higher power than you need to," he said.

Error condition

That insistence on perfection also pushes up manufacturing costs because many chips have to be discarded if they fall short.

Professor Kumar said that it would become harder and harder for chip makers to ensure instructions are executed flawlessly as components shrink.

The tiny components in chips are already starting to give rise to errors. Instead of trying to eliminate this, he said, it should be embraced to produce so-called "stochastic processors" that are subject to random errors.
***
more: http://news.bbc.co.uk/2/hi/technology/10134655.stm

[AdHocSolver]

1. There is so much buggy software out there now, few people will notice anyway.

On a more serious level, Intel has reached the limits of its chip technology. New processor designs are needed to increase computer capabilities.

Intel doesn't want to lose its near monopoly in computer chips so it is trying to convince the public to accept buying crap from them in the future.

That is what this article is all about. It is b#llsh#t.

[Greyskye]

2. As someone who worked for over 20 years in computer chip design...

...you are full of it.

The design limits being pushed are those of lithography, not 'chip technology'. When you start designing structures on the nanometer level, you have to design them differently than you did when you were designing at a larger scale. This has been true for the last quarter century, as the design processes kept shrinking.

When I started doing memory chip layout, you could actually see the metal trace lines with the naked eye, and the entire silicon area was the size of my thumbnail. The last chip that I worked on would have just covered Roosevelt's ear on a U.S. dime.

Every year new lithography rules are produced for the next smaller design process. We continuously bought or developed new software to enable us to design and simulate the results before committing them to silicon. Every year the designs change to match the latest benefits and drawbacks that the new process rules gave us.

I'm sorry that you hate Intel so much, but this isn't a conspiracy on their part. :eyes:

[AdHocSolver]

3. Intel could lose their monopoly on patents in processor technology.

Since their manufacturing methodology appears to have reached its limit of capability, what better way to hold on to market share than to convince buyers that, even though the quality of their product will have to be compromised in order to produce a competitive product that uses newer technologies, buyers should stick with Intel.

As someone who programmed computers for over 20 years, including PC's, I am well aware of the difficulty in producing reliable software for stable processors. So Intel comes along and wants the public to accept processors from them that would add erratic behavior to the mix of buggy software.

Then there is the problem of computers used in embedded systems. From all the articles I have read, there is considerable reason to believe that the "surge" problem exhibited by several Toyota models is in reality a computer problem, not a floor mat or sticky throttle problem.

Intel trying to convince people to accept an extra degree of erratic behavior in computer systems, when it is difficult enough to maintain reliability with current technology, is totally irresponsible.

By the way, the ogre in the computer field is not Intel, it is Microsoft.

[Greyskye]

4. Did you actually read the linked article?

Did you comprehend what I wrote in regards to shrinking lithography?

It's not a software problem, as you suggest. It has to do with physics on at the nanometer and sub-nanometer scales. In the last few years, we had to institute completely new physical design guidelines known as DFM (Design For Manufacturability) rules. Some examples of the purpose of some of these rules: what happens when you etch out around the poly and metal layers in order to extend a straight minimum width metal line without the trench wall collapsing? Rules to address how to array out contacts and vias without causing planer issues. How to lay out the diffusion around the poly gates without causing electromigration issues and unwanted high resistance areas. How to keep separate metal lines which come next to each other in a 'T' formation to keep from shorting together.

What the article the OP is referring to, is the flaky world of physics when you get down to the sizes we are approaching. It has nothing to do with Intel's way of designing microprocessors, this has everything to do with a career I spent almost 25 years in: Semiconductor Physical Design. I don't claim to be an expert in too many things, but this is one of them.

[AdHocSolver]

10. The only purpose of a processor is to execute software correctly.

Most software is full of bugs already. A major requirement of good software development is to eliminate as many bugs as possible while, at the same time, writing programs that fail "gently" so as to prevent catastrophic failure.

To achieve this end, the processor must operate in a consistent manner in all cases. Any erratic behavior by the processor, even only one percent, is totally unacceptable.

This quote from the article demonstrates that this guy doesn't know what he is talking about when he describes his ideas on system reliability:

*******
To cope with this, Professor Kumar and colleagues are researching ways to make applications more tolerant of mistakes. The "robustification" of software, as he calls it, involves re-writing it so an error simply causes the execution of instructions to take longer. In another approach, the more robust software logs a user's actions. As the software is used, this log can be consulted to spot when something unexpected occurs.
*******

Writing good software is already the most difficult and time consuming effort in developing any system. There aren't enough skilled programmers on this planet who could make his "vision" a reality. It is hard enough to develop good software for reliable processors, let alone write software for processors that are erratic even for one percent of the time.

In the end, it is a software problem. Trying to write software that will take into account erratic processor operation is a fool's errand. Those guys are selling snake oil. They are idiots.

[Greyskye]

11. Back up and re-read the article.

:banghead:

If you truly believe that this stochastic approach is unworkable, then you are advocating giving up the push to keep on track with Moore's Law. This is not something limited to Intel. This is a physical limit caused by what goes on when etching and doing lithography on a sub-nanometer scale in silicon. Every single semi-conductor maker on the planet is working under the same physical laws, and they all are in the same boat. Intel is in the news simply since they are usually one to two generations ahead of anyone else out there in the production world, and this is one of the solutions that they are evidently looking into.

The chip 'errors' that they are talking about in the linked article are practically random variables at the scale we're talking about. One of the ways you reduce these, is by using the DFM rules I mentioned earlier. Intel is looking into ways to further reduce these errors, in this case using stochastic programming.

Do you own a cell phone? Chances are good that I was on the team that developed the flash memory chip that your cell phone uses. And you know what? We built redundancy into the flash memory array, so that when one, or two, or 10 rows or columns in that memory array fail due to process issues in the fab, that failure is detected, those rows or columns are blocked off, and part of the redundant memory array structure kicks in instead. And you don't lose data. Which is what you care about in the end; having an inexpensive reliable product.

That is the sort of thing they're talking about, except on a larger, smarter scale. And if it can't be done, either through this method, or some alternative, microprocessors are going to stop getting smaller, faster, and cheaper at the rate that we've become accustomed to over the last 20 years. Sure, they'll keep getting faster and more powerful by hook or by crook; but if you still want them cheap, we've got to think outside the box. This is one of those out-of-the-box shots. If this doesn't work, then they will either find another way around this particular issue, or we're going to have to abandon silicon for something else, or someone will come up with an entirely new computing paradigm that doesn't have the failure rates associated with sub-nanometer technology.

They aren't idiots, as you call them. I think that this is an interesting approach that has a shot at success. But what the hell do I know; I've only given presentations at international semiconductor conventions. The last one I presented at was the International Cadence Usersgroup convention in Santa Clara. Do you have any credentials in this field?

[AdHocSolver]

12. If Moore's law no longer applies using this technology, then it is time to develop a new technology.

Moore's law is not a law of physics. It is an observation/prediction by Intel co-founder Gordon Moore in a 1965 paper that the number of transistors that can be placed inexpensively on an integrated circuit doubles about every two years.

As commentary in Wikipedia points out:

**********
His prediction has proved to be uncannily accurate, in part because the law is now used in the semiconductor industry to guide long-term planning and to set targets for research and development.<11> This fact would support an alternative view that the "law" unfolds as a self-fulfilling prophecy, where the goal set by the prediction charts the course for realized capability.
**********

In other words, Moore's law charts a business model, and that predictive model has run its course. However, Intel has not come up with any new technology in which to transition, so it is trying to convince the public to continue using its technology based on an acceptance of reduced reliability from its products.

The thought of a piece of computer controlled medical equipment relying on a processor with an already built in error rate of even one percent does not make for optimism.

The same goes for riding in an airplane whose navigation and communication equipment would rely on microprocessors with a minimum allowed error rate of even one percent.

Building redundancy into memory chips so that bad circuitry can be locked out is not the same as allowing faulty circuitry to remain in a processor that executes code.

As for credentials, I worked for over twenty years as an electronics technician both in repairing equipment to the component level, and as a hardware developer prototyping and debugging hardware for an electronics manufacturer. This is in addition to developing software for embedded systems.

[Greyskye]

13. You aren't actually bothering to read a single thing I'm saying, are you?

:banghead: :banghead:

Give me a break.

Lecturing me on Moore's Law? That's actually insanely funny. Congrats on being able to use Wikipedia, however. :eyes: Not only did I not make the claim that it's an actual physical law, we have drifted way away from the point here, and have been for a long time.

Here is where we started:

On a more serious level, Intel has reached the limits of its chip technology. New processor designs are needed to increase computer capabilities.

Intel doesn't want to lose its near monopoly in computer chips so it is trying to convince the public to accept buying crap from them in the future.

That is what this article is all about. It is b#llsh#t.

You made a claim that I showed to be patent nonsense. You continue to show that you don't have a clue about what this article is talking about, and your credentials, while quite nice, are irrelevant to this conversation and have absolutely nothing to do with semiconductor physics at the sub-nanometer scale. You have yet to address a single point that I've made, instead setting up yet another straw-man for me to tilt at.

You keep talking about 'processor design', as if by using a different architecture, this problem will somehow vanish. That's the difficulty in this conversation; you don't seem to understand that we are not talking about chip architecture. This is something completely different. And until we are speaking the same language, we're simply going to be going around in circles.

Have a nice day. :eyes:

[FiveGoodMen]

5. So that eventually they'll be just as smart and flawed as we are?

I thought we built computers to excel at things that we find difficult.

[eppur_se_muova]

9. They'll develop more "personality" if they're allowed to be flawed. nt

:evilgrin:

[FiveGoodMen]

6. dupe

[Greyskye]

7. Link?

[FiveGoodMen]

8. Sorry. My post #6 was a duplicate of my post #5