I can't remember who said or wrote it, but back when optimism reigned regarding AI, some academic asserted, "Man, like the ant, is simple." Wow, have we ever gotten a cold slap of reality since then! AI has been the fusion power and flying car of computer science, all rolled into one.
Adaptive morphogenetic systems are going to be the answer, I feel, whether using neural nets, fuzzy logic, or classifier systems, whether adapted by simulated annealing, genetic algorithms, or ... ? I once used a GA to help the local high school AD schedule district track meets. It was a simple app of a potentially powerful technique. I've been less successful with it in solving the Beale cipher, factoring products of large primes, or winning the Netflix Prize. So I have to keep working for a living!
After reflection, we have made several changes to our website based on comments made on this forum, so I do thank you for the comments.
We have ordered our first set of PCBs. They should be here next week. The wirewraps were only prototypes.
I really urge you to read the article in Robot magazine. Tom Atwood, the author (and editor) did a fantastic job interviewing my father-in-law and distilling the information down to very understandable language.
As far as how far the technology will go, I have heard Dr. Gouge mention that he projects it to be on the order of small mammals.
You should know that Dr. Gouge is the technical expert. I have only recently been working with him. I built the website with my wife's help and maintain it and work with him, but he is the one with the finger on the pulse of machine intelligence technologies. I hope to learn all I can from him.
mallred said...
After reflection, we have made several changes to our website based on comments made on this forum, so I do thank you for the comments.
We have ordered our first set of PCBs. They should be here next week. The wirewraps were only prototypes.
I really urge you to read the article in Robot magazine. Tom Atwood, the author (and editor) did a fantastic job interviewing my father-in-law and distilling the information down to very understandable language.
As far as how far the technology will go, I have heard Dr. Gouge mention that he projects it to be on the order of small mammals.
You should know that Dr. Gouge is the technical expert. I have only recently been working with him. I built the website with my wife's help and maintain it and work with him, but he is the one with the finger on the pulse of machine intelligence technologies. I hope to learn all I can from him.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔ Largos - a nano operating system for the Propeller www.mikronauts.com - a new blog about microcontrollers
Phil Pilgrim (PhiPi) said...
I can't remember who said or wrote it, but back when optimism reigned regarding AI, some academic asserted, "Man, like the ant, is simple." Wow, have we ever gotten a cold slap of reality since then! AI has been the fusion power and flying car of computer science, all rolled into one....
I know AI can be impressive but I'd be willing to bet this whole bit-banging paradigm is completely off the mark. Although I'm sure there are levels of brain functioning that operate very similarly to AI, there's a lot more to how we (our minds) work with information than any algorithm can imitate. Take acupuncture, for example. I have no fear of needles whatsoever and I had always regarded acupuncture as completely bogus/placebo type effect. All that changed, however, when a personal friend of mine, who is also a "normal" physician, stuck me here and there to give me a demo. Whoa. One switch to turn on this, one switch to turn on that. Afterwards, I was a true believer. Problem is, acupuncture doesn't seem related to the standard wiring of our nervous system. It's got it's own lines of communication of some sort. But I don't think anybody really knows upon what physical principle it's based. So, if the paradigm of wires doesn't hold up, it's hard to imagine that more and more wires will lead to intelligence.
I wouldn't agree with this in the article:
"The boards are extremely sensitive. You must be careful with them.
Do not subject them to harsh environments, either physically or electrically, such as excessive heat, high
humidity, condensed moisture on the board, or static electricity."
Although it may just be there as CYA language, and general good practice, I have found that the protoboard, propeller, and most of parallax's stuff to be very tolerant of harsh environments and hazards(like children who have grabby little fingers and walk on carpet;^))
I think it would be better to point out good general practice than to say that "the boards we are modifying" are sensitive. I would guess that the wire wrapped ones are pretty sensitive, but I don't know!
Actually, the older PropellerProto boards are more sensitive to higher temperature soldering than the new ones.
Anything over 600F was sure to twitch my nose. Don't know what the limit is for the new boards, but they are more robust.
I bought the magazine and read the article. I have more questions after reading the article. I cannot tell if it is snake oil or the breakthrough of the century. He is down on conventional neural nets. "Neural nets are nothing but statistical models whereby you alread know all the possible connections, and all you do is give a weight to a connection and make it stronger by increassing the weighting." But the neural nets I have seen are highly connected. After training many connections have a very low weight, essentially not conncted. "Until we get away from this business of predetermining how they are connected and instead come up with learing algorithms that allow making mistakes and then correcting them, we have not done it. There is no learning withou being able to make mistakes." I am not sure I see the difference between this and training a conventional neural net. Learning in his system is supervised. He uses error correcting codes (ECC) to combine all this abstract information in logic streams as opposed to data streams. "The ECC codes are, in a sense, analogous to the logic packets in the human brain." I don't know what a logic stream or logic packet is. What I could use are a few examples.
From the machineinteltech.com web site the series of articles will be: Memory expansion, servo/motor control system, vision control system (binocular), electrosensory control system, mechanical construction of the robot platform, then two articles on training the AI. A little over a year to complete the articles.
What Dr. Gouge is hinting at may bear some similarity to a Learning Classifier System, which was first publicized by John Holland (the father of the genetic algorithm) around 1975. Here are a couple references:
Looking good. I see some decoupling caps. I'll probably open a hornets nest here, but I've found with LS you need caps for all the chips, but with HC or HCT you can get away with caps for maybe one in three chips. All that is really needed is a current path that is fairly short to the nearest chip.
I'm not sure what package you are using but if it is Eagle there are some nifty ways to make the supply lines fatter, but still keep the data lines thin. It probably isn't necessary though, and I've built boards with thin supply lines.
I'm looking forward to seeing the board when it is populated.
Post Edited (Dr_Acula (James Moxham)) : 6/5/2009 2:49:20 AM GMT
The AI Winter lasted from 1970 to 1995. This was the time when all the grandiose plans failed. But, the Phoenix arose with Mark Tillman and others who realized you can't program from the top-down, instead programming from the bottom-up. This is called subsumptive programming. Now that it has had limited success, the researchers are groping for the Next Big Thing. Perhaps Dr. Gouge is feeling his way toward that path.
Some of the keenest work in emergent complex systems is being done at the Santa Fe Institute in New Mexico, where John Holland is a fellow. James Huselton is referring to Mark Tilden, whose B.E.A.M. robots exhibit remarkably complex behavior using simple components. Subsumption architecture arose out of research by Rodney Brooks at MIT in 1986. It's described in the book, Mobile Robots, by Jones, Flynn, and Seiger.
Good points there PhiPi. The first time I read about subsumption architecture was Rodney Brooks' early papers. As an aside, he is from Adelaide like myself and did his early work at the university down the road, before moving to the US. www.absoluteastronomy.com/topics/Rodney_Brooks
The premise is to build things from the bottom up, eg build a walking robot modelled on an Ant, or even just an individual limb, before you take things further. I always wished I had access to all the cool stuff they have at the MIT lab!
Post Edited (Dr_Acula (James Moxham)) : 6/5/2009 3:08:46 AM GMT
Its seems that I'm in and out of the forum lately...actually I'm been away more than not. So I missed the bulk of this topic the first time around... It really did take almost an hour to read the whole thing.
I personally love a bit of drama every once in a while...so it was all quite entertaining.
I haven't read the Robot Magazine article yet...but I'm intrigued enough now that I probably will.
Personally, I think many of the claims being made are too crazy to be true. I'd be happy if I were wrong...but I don't think I am.
I love the Propeller and have used it in many project over the past 2-3 years, but I seriously question the premise that the Propeller is the "perfect" platform for a revolutionary new type of "machine intelligence". Again I haven't read the article yet, but I've done alot of acedemic research in Cognitive Science and all sorts of traditional and non-traditional (NN, GA, FL..) computational intellegence.
Its been mentioned before, but this is just not a cost effective solution. For the same money you can buy a whole lot more processing power. If there were any substance to the "innovative approach" being implemented on the propellers. Wouldn't it work even better on a "simulated" massively parrallel architecture using conventional PC and networking technology.
If you really need some power go buy a shelf full of PS3's...the Cell Processor in them is an insanely powerful 8 core processor with tons of ramand bandwidth... Putting that processor in the PS3 was a masterfull plan by Sony, IBM, and Panasonic (or maybe it was Toshiba I can't remember)... but it was brilliant mass market the processor you want to make the next super computer out of [noparse]:)[/noparse]
The point is...there isn't any shortage of parallel processing power. At least not if you're seriously looking at the Propeller for your base platform.
Any revolutionary algorithm that runs on some kind of Propeller + RAM board would run much better on other platforms for a fraction of the cost.
Of course if you're the one selling the product then you might not want a solution that sells "at a fraction of the cost..."
Naturally I'm skeptical of such bold claims, but I'll be the first to admit I was wrong when I am.
A shelf full of ps3 would be powerful but not very mobile.
My prop galore board which has 8 props and the same bus he is talking about using is only 7x1. In an 8 inch cube you could get 256 of these boards. That is 2048 cogs running 20 mips each. 41 gips.
All beit at a cost of $51k
The sony cell runs at 10.2Gips so 4 of them would be needed to match the 8x8x8 cube of props.
And the AMD Phenom II X4 940 Black Edition runs at 42.8 GIPs
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
propmod_us and propmod_1x1 are now in stock. Only $30. PCB available for $5
Need to upload large images or movies for use in the forum. you can do so at uploader.propmodule.com for free.
The point is that if you really have a new computational intelligence model that is going the change the world as we know it... and all you need to make it a reality is some parallel processing...then the propeller is definitely NOT the platform that you're going to be spending your time and money developing this new system on.
Especially given the extreme measures its obviously taking just getting this super secret, patented, ground breaking, new cognitive system to "run" on the propeller.
As for not being "mobile", I'm all for the importance of embodied intelligence...but realistically the next generation of machine intelligence is not very likely to come from embedded processor technology... Its much more likely the first successful implementation will be some kind of "tethered" robot body and sensors connected to a processor that can actually handle the torrent of "real world" information processing.
The whole premise that a large stack of props is even remotely on the same computational playing field as even a primative neurological systemis ludicrous at best.
A stack of props isn't even going to be computationally competitive with a high school computer lab running a beowulf cluster, let alone a state-of-the-art graduate school lab, which any serious researcher could easily get their hands on if they have a truly revolutionary idea.
Dr_Acula (James Moxham) said...
Looking good. I see some decoupling caps. I'll probably open a hornets nest here, but I've found with LS you need caps for all the chips, but with HC or HCT you can get away with caps for maybe one in three chips. All that is really needed is a current path that is fairly short to the nearest chip.
There are what appear to be a couple of decoupling capacitors, but they are a long way from the chips and probably won't be very effective. I always use one per logic chip, they are cheap enough. A view of the underside of the board would settle the matter.
Leon
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Amateur radio callsign: G1HSM
Suzuki SV1000S motorcycle
mallred said...
FYI, this is a four-layer board. The middle two layers are power and ground, so the entire board has access to it.
Speaking as an ignorant newbie, I would like to know if having power and ground planes eliminates the need for decoupling caps? Also interesting to me is how close to the pins some of the traces run - some running right between pins. That certainly makes it easier to route traces, but for some reason that seems risky to me. For one thing my own (lack of) soldering skills would probably short that out in no time, but I am also curious about the potential for interference, or crosstalk, between the trace and a pin. I'm only asking this because I'm still learning how to design my own PCBs, not to cast aspersions upon the Gouge revolution.
4-layer designs are far superior in noise immunity, parasitic capacitances, stray inductance, and crosstalk than two-sided designs.
Properly designed, that is.
Decoupling capacitors are mandatory for multi-layer PCBs.
Crosstalk between pins and tracks is negligible, but parallel tracks, especially when one of them is carrying a clock signal, often give crosstalk problems.
Leon
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Amateur radio callsign: G1HSM
Suzuki SV1000S motorcycle
It is about time for me to jump into this forum and clear up some confusion. For the moment, I will just say that I have known Jim for over 30 years. He taught me a lot of what I know about computers and electronics.
Let me clear up some confusion.
Mark Allred is not Dr. Jim. He is helping Jim from a business perspective rather than a technical one.
The Robot Magazine issue with the interview of Dr. Jim is on the newsstands now. It is the Jul/Aug issue with the tall black robot on the cover. On the Current Issue page at botmag.com, the link under the Machine Intelligence article is to a document with instructions to connect the Expanded RAM (ERAM) board to a Propeller Proto Board, not the article in the magazine. I don't know if you can read the magazine article online. Check with Robot Magazine to see if they provide online access to the articles in the current issue if you have a paid subscription.
The photos in the construction article (same article linked to earlier in this forum) show the wire-wrap prototype of the ERAM board. I have seen photos of the production ERAM board. It is a professionally manufactured 4 layer glass epoxy PCB. The parts locations are screen-printed on the board. The parts layout is the same as on the wire-wrap board.
The last few paragraphs of the magazine article give some veiled hints at Dr. Jim's approach to Machine Intelligence. Jim thinks outside the box, way outside the box. His approach is totally different from what everyone else is doing. Based on the "success" of current approaches to artificial intelligence, in my opinion, the results they get are, well, artificial. To make any real progress on recreating human intelligence requires a very different approach. That is the way Jim is attacking the problem.
Well, apparently, I missed some of the other more recent posts while writing my first post. It looks like we have a lively discussion going on about recent work on Artificial Intelligence. I think I have some reading to do.
Let me share a little about the difference between the Propeller and the Cell Processor. A recent episode (41, I think) of the Futures in Biotech Podcast, discussed the Folding@Home project and the significant increase in computational power it gets from running on the PS3 over a standard computer. If I understand it correctly, that project depends heavily on floating point computations. If you need floating point computations, you definitely don't want a Propeller Cog. On the other hand, if your approach is very different and focused on simplicity rather than complex computations, the Propeller is a good development platform. Trying to split a complex program into multiple threads to run on multiple high-end cores is a very difficult proposition. Big bucks are spent on the development of compilers that generate code to run on systems with multiple processors, each with multiple complex cores. It you are trying to do a lot of simple loosely related tasks on separate processors, everything gets much simpler and the Propeller gets you into the ballpark.
russells said...
...·Trying to split a complex program into multiple threads to run on multiple high-end cores is a very difficult proposition. Big bucks are spent on the development of compilers that generate code to run on systems with multiple processors, each with multiple complex cores. ...
Symmetric Multiprocessing (SMP) on Linux is quite easy to use. I helped design a system with 96 MIPS CPU cores (one 1.5"x18"x18" unit for connecting·in a 16 unit backplane) running 800MHz each for network threat detection, security protocols, network virtualization, and management ... 16 cores per chip (6 chips) per Linux instance sharing 2GB fully coherent DDRII memory per chip communicating with other chips over 10GB/s interface links. The task was not AI, but certainly complicated. SMP is a wonderful thing for big distributed systems ... the right tool for the job as they say.
Comments
Adaptive morphogenetic systems are going to be the answer, I feel, whether using neural nets, fuzzy logic, or classifier systems, whether adapted by simulated annealing, genetic algorithms, or ... ? I once used a GA to help the local high school AD schedule district track meets. It was a simple app of a potentially powerful technique. I've been less successful with it in solving the Beale cipher, factoring products of large primes, or winning the Netflix Prize. So I have to keep working for a living!
-Phil
We have ordered our first set of PCBs. They should be here next week. The wirewraps were only prototypes.
I really urge you to read the article in Robot magazine. Tom Atwood, the author (and editor) did a fantastic job interviewing my father-in-law and distilling the information down to very understandable language.
As far as how far the technology will go, I have heard Dr. Gouge mention that he projects it to be on the order of small mammals.
You should know that Dr. Gouge is the technical expert. I have only recently been working with him. I built the website with my wife's help and maintain it and work with him, but he is the one with the finger on the pulse of machine intelligence technologies. I hope to learn all I can from him.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Largos - a nano operating system for the Propeller
www.mikronauts.com - a new blog about microcontrollers
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
propmod_us are now in stock. propmod_1x1 arrive on 26th. Only $30
Need to upload large images or movies for use in the forum. you can do so at uploader.propmodule.com for free.
I know AI can be impressive but I'd be willing to bet this whole bit-banging paradigm is completely off the mark. Although I'm sure there are levels of brain functioning that operate very similarly to AI, there's a lot more to how we (our minds) work with information than any algorithm can imitate. Take acupuncture, for example. I have no fear of needles whatsoever and I had always regarded acupuncture as completely bogus/placebo type effect. All that changed, however, when a personal friend of mine, who is also a "normal" physician, stuck me here and there to give me a demo. Whoa. One switch to turn on this, one switch to turn on that. Afterwards, I was a true believer. Problem is, acupuncture doesn't seem related to the standard wiring of our nervous system. It's got it's own lines of communication of some sort. But I don't think anybody really knows upon what physical principle it's based. So, if the paradigm of wires doesn't hold up, it's hard to imagine that more and more wires will lead to intelligence.
"The boards are extremely sensitive. You must be careful with them.
Do not subject them to harsh environments, either physically or electrically, such as excessive heat, high
humidity, condensed moisture on the board, or static electricity."
Although it may just be there as CYA language, and general good practice, I have found that the protoboard, propeller, and most of parallax's stuff to be very tolerant of harsh environments and hazards(like children who have grabby little fingers and walk on carpet;^))
I think it would be better to point out good general practice than to say that "the boards we are modifying" are sensitive. I would guess that the wire wrapped ones are pretty sensitive, but I don't know!
Just my 10 bits worth.
I too would take a little exception to the fragility of the Protoboard.
I've found the board to be very robust dispite all kinds of mishandling.
Looking forward to seeing the next article.
Good community contribution! Glad to see it.
OBC
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
New to the Propeller?
Visit the: The Propeller Pages @ Warranty Void.
Anything over 600F was sure to twitch my nose. Don't know what the limit is for the new boards, but they are more robust.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
--Steve
Propalyzer: Propeller PC Logic Analyzer
http://forums.parallax.com/showthread.php?p=788230
From the machineinteltech.com web site the series of articles will be: Memory expansion, servo/motor control system, vision control system (binocular), electrosensory control system, mechanical construction of the robot platform, then two articles on training the AI. A little over a year to complete the articles.
John Abshier
····en.wikipedia.org/wiki/Learning_classifier_system
····subsimple.com/classifier.asp
-Phil
Post Edited (mallred) : 6/4/2009 7:01:17 PM GMT
Leon
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Amateur radio callsign: G1HSM
Suzuki SV1000S motorcycle
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
JMH
I'm not sure what package you are using but if it is Eagle there are some nifty ways to make the supply lines fatter, but still keep the data lines thin. It probably isn't necessary though, and I've built boards with thin supply lines.
I'm looking forward to seeing the board when it is populated.
Post Edited (Dr_Acula (James Moxham)) : 6/5/2009 2:49:20 AM GMT
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
JMH
-Phil
The premise is to build things from the bottom up, eg build a walking robot modelled on an Ant, or even just an individual limb, before you take things further. I always wished I had access to all the cool stuff they have at the MIT lab!
Post Edited (Dr_Acula (James Moxham)) : 6/5/2009 3:08:46 AM GMT
I personally love a bit of drama every once in a while...so it was all quite entertaining.
I haven't read the Robot Magazine article yet...but I'm intrigued enough now that I probably will.
Personally, I think many of the claims being made are too crazy to be true. I'd be happy if I were wrong...but I don't think I am.
I love the Propeller and have used it in many project over the past 2-3 years, but I seriously question the premise that the Propeller is the "perfect" platform for a revolutionary new type of "machine intelligence". Again I haven't read the article yet, but I've done alot of acedemic research in Cognitive Science and all sorts of traditional and non-traditional (NN, GA, FL..) computational intellegence.
Its been mentioned before, but this is just not a cost effective solution. For the same money you can buy a whole lot more processing power. If there were any substance to the "innovative approach" being implemented on the propellers. Wouldn't it work even better on a "simulated" massively parrallel architecture using conventional PC and networking technology.
If you really need some power go buy a shelf full of PS3's...the Cell Processor in them is an insanely powerful 8 core processor with tons of ramand bandwidth... Putting that processor in the PS3 was a masterfull plan by Sony, IBM, and Panasonic (or maybe it was Toshiba I can't remember)... but it was brilliant mass market the processor you want to make the next super computer out of [noparse]:)[/noparse]
The point is...there isn't any shortage of parallel processing power. At least not if you're seriously looking at the Propeller for your base platform.
Any revolutionary algorithm that runs on some kind of Propeller + RAM board would run much better on other platforms for a fraction of the cost.
Of course if you're the one selling the product then you might not want a solution that sells "at a fraction of the cost..."
Naturally I'm skeptical of such bold claims, but I'll be the first to admit I was wrong when I am.
My prop galore board which has 8 props and the same bus he is talking about using is only 7x1. In an 8 inch cube you could get 256 of these boards. That is 2048 cogs running 20 mips each. 41 gips.
All beit at a cost of $51k
The sony cell runs at 10.2Gips so 4 of them would be needed to match the 8x8x8 cube of props.
And the AMD Phenom II X4 940 Black Edition runs at 42.8 GIPs
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
propmod_us and propmod_1x1 are now in stock. Only $30. PCB available for $5
Need to upload large images or movies for use in the forum. you can do so at uploader.propmodule.com for free.
Post Edited (mctrivia) : 6/5/2009 4:54:19 AM GMT
news.bbc.co.uk/2/hi/technology/7443557.stm
The point is that if you really have a new computational intelligence model that is going the change the world as we know it... and all you need to make it a reality is some parallel processing...then the propeller is definitely NOT the platform that you're going to be spending your time and money developing this new system on.
Especially given the extreme measures its obviously taking just getting this super secret, patented, ground breaking, new cognitive system to "run" on the propeller.
As for not being "mobile", I'm all for the importance of embodied intelligence...but realistically the next generation of machine intelligence is not very likely to come from embedded processor technology... Its much more likely the first successful implementation will be some kind of "tethered" robot body and sensors connected to a processor that can actually handle the torrent of "real world" information processing.
The whole premise that a large stack of props is even remotely on the same computational playing field as even a primative neurological systemis ludicrous at best.
A stack of props isn't even going to be computationally competitive with a high school computer lab running a beowulf cluster, let alone a state-of-the-art graduate school lab, which any serious researcher could easily get their hands on if they have a truly revolutionary idea.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
propmod_us and propmod_1x1 are now in stock. Only $30. PCB available for $5
Need to upload large images or movies for use in the forum. you can do so at uploader.propmodule.com for free.
There are what appear to be a couple of decoupling capacitors, but they are a long way from the chips and probably won't be very effective. I always use one per logic chip, they are cheap enough. A view of the underside of the board would settle the matter.
Leon
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Amateur radio callsign: G1HSM
Suzuki SV1000S motorcycle
Post Edited (Leon) : 6/5/2009 9:08:04 AM GMT
Speaking as an ignorant newbie, I would like to know if having power and ground planes eliminates the need for decoupling caps? Also interesting to me is how close to the pins some of the traces run - some running right between pins. That certainly makes it easier to route traces, but for some reason that seems risky to me. For one thing my own (lack of) soldering skills would probably short that out in no time, but I am also curious about the potential for interference, or crosstalk, between the trace and a pin. I'm only asking this because I'm still learning how to design my own PCBs, not to cast aspersions upon the Gouge revolution.
thanks,
Mark
4-layer designs are far superior in noise immunity, parasitic capacitances, stray inductance, and crosstalk than two-sided designs.
Properly designed, that is.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
JMH
Crosstalk between pins and tracks is negligible, but parallel tracks, especially when one of them is carrying a clock signal, often give crosstalk problems.
Leon
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Amateur radio callsign: G1HSM
Suzuki SV1000S motorcycle
Let me clear up some confusion.
Mark Allred is not Dr. Jim. He is helping Jim from a business perspective rather than a technical one.
The Robot Magazine issue with the interview of Dr. Jim is on the newsstands now. It is the Jul/Aug issue with the tall black robot on the cover. On the Current Issue page at botmag.com, the link under the Machine Intelligence article is to a document with instructions to connect the Expanded RAM (ERAM) board to a Propeller Proto Board, not the article in the magazine. I don't know if you can read the magazine article online. Check with Robot Magazine to see if they provide online access to the articles in the current issue if you have a paid subscription.
The photos in the construction article (same article linked to earlier in this forum) show the wire-wrap prototype of the ERAM board. I have seen photos of the production ERAM board. It is a professionally manufactured 4 layer glass epoxy PCB. The parts locations are screen-printed on the board. The parts layout is the same as on the wire-wrap board.
The last few paragraphs of the magazine article give some veiled hints at Dr. Jim's approach to Machine Intelligence. Jim thinks outside the box, way outside the box. His approach is totally different from what everyone else is doing. Based on the "success" of current approaches to artificial intelligence, in my opinion, the results they get are, well, artificial. To make any real progress on recreating human intelligence requires a very different approach. That is the way Jim is attacking the problem.
Sorry for the long post.
Russell
Let me share a little about the difference between the Propeller and the Cell Processor. A recent episode (41, I think) of the Futures in Biotech Podcast, discussed the Folding@Home project and the significant increase in computational power it gets from running on the PS3 over a standard computer. If I understand it correctly, that project depends heavily on floating point computations. If you need floating point computations, you definitely don't want a Propeller Cog. On the other hand, if your approach is very different and focused on simplicity rather than complex computations, the Propeller is a good development platform. Trying to split a complex program into multiple threads to run on multiple high-end cores is a very difficult proposition. Big bucks are spent on the development of compilers that generate code to run on systems with multiple processors, each with multiple complex cores. It you are trying to do a lot of simple loosely related tasks on separate processors, everything gets much simpler and the Propeller gets you into the ballpark.
Russell
Welcome to the forum, and thanks for sharing.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
--Steve
Propalyzer: Propeller PC Logic Analyzer
http://forums.parallax.com/showthread.php?p=788230
Post Edited (jazzed) : 6/7/2009 6:56:51 PM GMT