Twente researchers build evolutionary circuits

Sep

22

2015

Twente researchers have designed a circuit that can be configured as sixteen different logic circuits. The circuit consists of a disorder network of gold nanoparticles, and can evolve to perform Boolean functions.

The article on the circuit is Monday in the journal Nature Nanotechnology published. Wilfred van der Wiel, a physicist and professor of nanotechnology at the University Twente explains where the idea came from to create such a circuit. “Now we make logic circuits with a lot of transistors and chips we make according to the Von Neumann architecture was with blueprints for certain functions There are already a number of people have said,. That works great, but we throw in the process of design not much potential power away? That’s the price you pay to think in that permanent blocks. Pretty predictable, less powerful. ”

The latter is, according to Van der Wiel pity. “Why we do not inspiration in the brain? It is very different, not linear, but work very quickly because so many operations in parallel, something which does not happen in a conventional processor. It always goes sequentially through a clock . The individual components of the brain have been slow, but by the parallel operation the brain is very powerful. ”

From this, the question arose whether it would not be possible to do with “dead material ‘with which this non-linear process to mimic. That did the researchers in this case with gold nanospheres. At a temperature below 5 degrees Kelvin or -268ºC which behave themselves as a so-called single-electron transistors. By a single electron transistor can only move one electron at a time when the correct tension on it.

The latter is important, Van der says Wiel: “If you’re an extra electron on a piece wants to put metal, in this case only 20 nanometers, then sit electrons as it were in each other’s and you will notice a lot of each other disposal . Most of the time, this system therefore no current flows through, a phenomenon that Coulomb blockade is mentioned. This blockage can be overcome by putting a little bit of tension in the nanosphere, leaving it flow to pass through. Then it says that basal transistortje in the on position. ”

“When you zoom out,” Van der Wiel, “continues you will see a very strong non-linear electrical component. If you are already bringing particles together in a disorderly network, you still love individual islands by a molecular shell only one nanometer thick around the golden balls Electrons can however still the one to the other hopping We as a network of all those non-linear switches When we thought… if we do that, we have a sufficient complex system to build functional circuitry? ” And indeed it turned out to be.

The network of nano gold balls is about 200 nanometers in size and the balls individually have a diameter of 20nm. All of which are capacitively coupled to each other, whereby a lot of cross talk occurs. The latter is normally eliminated as much as possible, but the Twente researchers there did with this design not to worry about because use is made ​​of a ‘genetic algorithm’.

Van der Wiel: “Because we use a genetic search algorithm, we can all kinds of physics there, use this way we throw anything overboard That means that without design also make a design flaw, so we are also to some extent insensitive.. defects. If there are two balls that create a short circuit, then you will be growing around it. As long as you have enough knobs to turn to we can find logic. ”

This should ultimately result from a working circuit. It works as follows: The network is connected to nine electrodes. Two are inputs and an output. The remaining electrodes are used as electrodes configuration. “The inputs we put simple pulse trains, zero or one, and we’ll see what comes out at the output. The inputs are fixed. Is there at the output does not measure the desired behavior, you go to the buttons of the six configuration tensions rotating, then you look or the output looks more like it. ” says Van der Wiel. With the adjustment of the voltage through the six electrodes, the potential landscape changes of the network of nanoparticles. “Which particle what we do not know exactly. Because there are so many options, we use a genetic algorithm. The best sets of configuration voltages’m going to cross broad and so on. All the tricks boxes of natural evolution can be applied in this genetic evolution. ”

The desired outcome in the paper by researchers at the University of Twente’s MESA + – and CTIT institutes was a network that acted like a logical function such as AND, NAND, OR, NOR, XOR or XNOR. To get here was so used artificial evolution, something which even prevent mutations or, as Van der Wiel decision, “sometimes you got a lucky shot where you come to a much more appropriate solution, you will not hang in a local optimum. ”

darwin on a chip

Artist impression of the layout of the circuit. The gold beads are all about 20nm in size. Furthermore, concerns two of the electrodes for the input voltage, and there is output stream. The remaining six electrodes provide for the control of the circuit. If you miss an electrode there is another at the bottom.

Samsung announces 950 Pro SSD to
Americans spend more money on first streaming music than CDs

Read more about

Science

Latest IT jobs

Support employee / System

Serac Applications BV, Utrecht

Head of IT Development

Central Bureau of Statistics, The Hague

Software developer

Net Display Systems, Eindhoven

More jobs
Related content

17-09
News – Researchers change direction magnetic field of light without heat
30-04
News – Researchers find way to make ultrathin transistors

Moderation faq • Change view
Comments (22)
22 22 16 6th 1
2
Banaanx
September 22, 2015 09:56
Cool! During my bachelor assignment at the University of Twente, I have a sort of proof of concept on a large scale this worked. Good to see that now in miniature too will succeed. Personally, I worked with a large bucket of tennis balls instead of gold particles. We put a circuit in (a lot more complex than the intended NAND gate, but oh well) and each ball was fitted with six aluminum electrodes. Next, electrodes 16 on the inside of the tray, to link to the PC and look for it!

The great advantage of this technique is also not acquire more power. We are pushing the limits, but of lithography and should therefore 2D to 3D processor framing. And not a few layers of processors: the same resolution in all directions.

That can be done in roughly two ways, both of which rely on self-assembly. Option one is perfect structures grow in a fixed design, similar to what we do in 2D. However: any impurity in your structure a potential means not working entirely.

This research focuses on the other option: you aim your processor raw material in a container and program your processor afterwards. You need more particles for a given logic function, but which can be very small. Precision is also no longer an issue. The next step is perhaps for all those found to pack logic gates and throw together in a larger container, and to seek a more complex circuit.
Mac Hack
September 22, 2015 08:41

Van der Wiel: “Because we use a genetic search algorithm, we can all kinds of physics there, use this way we throw anything overboard That means that without design also make a design flaw, so we are also to some extent insensitive.. defects. If there are two balls that create a short circuit, then you will be growing around it. As long as you have enough knobs to turn to we can find logic. ”

I think this is to some extent a little “wishful thinking”. Without design, you can not make design mistakes, but if you need enough buttons that you also need to run well, then you are to some extent also working on designs. But then the behavior of the algorithm, for example, to evolved around short-circuits. If you make mistakes in there you also get defects. Because there are so many combinations that things can go wrong, well that one can hardly imagine it all in advance. And you can also think awkward buttons up front and update it. And if you need to change the configuration of the algorithm which in turn affect other combinations. The only thing I see happening is that they are the problem where defects occur because of design flaws shift to the buttons and their values ​​thereof.
Jael_Jablabla
Macshack • September 22, 2015 09:46

[…]

I think this is to some extent a little “wishful thinking”. Without design, you can not make design mistakes, but if you need enough buttons that you also need to run well, then you are to some extent also working on designs.

I think he means more so that you two draft-free “devices”, which both have the same output, but different internal structure. Think of a brain. Two people can ride both, but that part of the brains that is responsible for that is certainly not identical.
polthemol
Macshack • September 22, 2015 10:24
design means that you’re focusing circuits make in the context of their problem. That’s not what they do here. Your argument is a bit semantic 😉
Mac Hack
polthemol • September 22, 2015 10:32
I disagree with you. I think you make him symantisch. By means of the application of the buttons and the determination of the values ​​thereof, you have to have a specific goal in mind. To achieve this goal, you must design the right buttons with the correct values. Of course, these buttons making it possible to make combinations which in advance are not designed / could have been. My experience with genetic algorithms is that coming up with the right buttons and their values ​​as much a part of the design is like making a deterministic model. You verschuifd only design where one can definitely get an advantage. But it is as much a design like designing a deterministic model.
keko12
September 22, 2015 08:53
Partially overlapping, but also additional information:
Neural networking nano gold
gwie
September 22, 2015 07:49
Fat! If I understand correctly you should therefore try very much until the correct / desired configuration has been found, but the circuit operates then be faster than a conventional circuit and it is possible to change the circuit at any other circuit.

I do wonder if it is really possible to make this all required circuitry and whether the circuit will not change if the same configuration extended period of time is maintained.
Splitinfinitive
gwie • September 22, 2015 08:25
Maybe I’ll have a stupid question considering I am not totally at home in this matter. But if I read it right then this is possibly a proof of concept for a completely different CPU design. But if I by parallel hatching then I immediately think of a GPU. This can also work very well in parallel.

But if you want the same as above can not you try as much as possible small cores on a CPU slices a la gpu? That example, each core 100 transistors big late but fills up to the level of today (I think we were already in the 6 billion transistors?) Then you can also do very good work in parallel?
gwie
Splitinfinitive • September 22, 2015 08:30
You need to look at FPGAs and how they work. The design of the research can be compared to a big fpga whose architecture is not fixed. Through try and ‘evolution’ is the right architecture to obtain the desired circuit.
Rob Coops
gwie • September 22, 2015 09:13
Not only extended time maintain the same circuit but I also wonder whether it is really efficient. After 20nm * x globules to an OR / NOR / xor etc to make circuit sounds rather grand. Perhaps soon but for number of circuits per square mm seems to be a standard solution yet slightly more efficient use of the available space.

Also, I do think that all the fiddle to obtain a desired circuit is fun on a small scale but sometimes trying to make a complex circuit which serves a specific purpose. The number of inputs is high, the quantity of combinations as well, and there also is the fact that the movement on next state of the input weather influence will want to have on the result of the previous input, which appears in combination with many output channels, and the whole what difficult to make. I suspect that other than a nice study that does not really get much further because it can not be made reliable and widely used.
Even if you manage to create a highly efficient high-speed circuit that does exactly what you want, you can still do it 10,000x? Or is it rather difficult to gold beads exactly the same position with respect to gain from each other? How long does it take a complex circuit to make it work exactly as you want, and if the control voltage is so important can you keep the same during the life of the circuit? To Not to mention the small problem of temperature (still quite tricky to 5 kelvin)

My interpretation is so nice idea but very difficult to put this into useful outside the lab.
tijnvw
Rob Coops • September 22, 2015 10:33
I think you’re quite right in itself, were it not that the first computers of the “current sort also so incredible were big and clumsy, you can also speak in this context of” only useful in the lab. ” As an example: the father of a good friend recently told me that he used to work on the mathematics faculty of the university, and they there at some point had their first (scientific) computer. In winter, when the street lamps were burning and he went out the door around 17 am, he could see that the computer turned off because the street lights then went even burn a little brighter. One computer (with probably less computing power that current mobile phone in your pocket) that draws enough power to influence the urban network: something can only be suitable for laboratiumgebruik. Nevertheless, the grandparents of today’s mobile phones ensures that you now have a very fancy computer in your pocket.

Moral of this story: maybe this invention as a prototype still totally unsuited to compete with the current state-of-the-art computers; may be it contributes to a kind of technology we all use over X years 🙂
antubus
September 22, 2015 09:03
Very cool, does anyone know the title of the paper?
Edit: found the title:
Evolution of a design less nanoparticle network into reconfigurable Boolean logic
link to Nature: http://www.nature.com/nna…/full/nnano.2015.207.html

[Response changed by antubus on September 22, 2015 09:07]
Author
letatcest
antubus • September 22, 2015 09:15
Fyi, the Nature link is the first link in the article he 😉
antubus
letatcest • September 22, 2015 10:49
Okay, that was not so smart of me | 🙁
Author
letatcest
antubus • September 22, 2015 10:55
Ah dude, does not matter 😉 I must also say that what we ambush the lot 😉
Xymox
September 22, 2015 09:13
And what if the “turning the knobs” is now part of the output of the system?
Pure non-linear process which sends itself. You would be a “natural selection” to choose the system so that it evolved to the optimum in the “environment”.
Mac Hack
Xymox • September 22, 2015 10:42

Pure non-linear process which sends itself.

Send, where to? Send implicated a goal. A goal is formulated by the designer.

You would be a “natural selection” to choose the system so that it evolved to the optimum in the “environment”.

Exactly! Someone has to make a choice about the fitness of a particular outcome to determine if you continue to use this generation to get crosses. Therefore it has nothing to do with natural selection. The selection is done there by the designer because the designer has a specific goal that he wants to go.
Stoney3K

[karma king]
Science

Macshack • September 22, 2015 12:15

[…]
Exactly! Someone has to make a choice about the fitness of a particular outcome to determine if you continue to use this generation to get crosses. Therefore it has nothing to do with natural selection. The selection is done there by the designer because the designer has a specific goal that he wants to go.

Natural selection works only if there are enough ‘threats’ that help a faulty copy to soap as it does not meet your needs, and if the system can reconfigure itself (“breed”) to a new generation.

The idea of ​​natural selection is not that haphazard and random mount configurations are tried until one is who comes through the filter. 😉
Jael_Jablabla
Xymox • September 22, 2015 09:49

And what if the “turning the knobs” is now part of the output of the system?
Pure non-linear process which sends itself. You would be a “natural selection” to choose the system so that it evolved to the optimum in the “environment”.

Control! 😉
terracide
September 22, 2015 08:33
The only thing it lacks is a type of learning function that allows the optimal circuitry being held. Circuitry is now loose and every time random is if the previous circuit more optimal than the current one. But the beginning is there, of something revolutionary, a parallel brain 🙂
Jael_Jablabla
September 22, 2015 09:57
I still do not understand the basis of this article.
Switches are built evolve and thus they can form complex structures? What I do not understand why gold nanospheres are necessary, why the tempteratuur 5 Kelvin should be, and why uberhaupt is thought in “switches”. What’s the problem here is trying to solve: to build an evolutionary switch or build something that uses it? There are more things I do not understand in this article.
BW83
September 22, 2015 11:17
This is clearly beyond my nonscientific cap 😉

Can someone explain to me what you’re here to mean in practice, what kind of benefit you Achieve it? Or can not translate this into practice where John Doe in everyday life to do with it?
Reaction ban

Dear Visiter,

Tweakers limits the functionality of the site for some countries to reduce spam, fraud and abuse.

Your IP address (117 217 123 117) has been reduced to country code IN and the country falls under this restriction. Your IP address is thus excluded from posting comments on Tweakers. For more information, also this explanation.

If you log in, we can determine whether this indeed applies to your account.

In many cases, we can make personal exceptions, send an email before to frontpage@tweakers.net. Mentioned in that mail your user name and IP address.

Samsung announces 950 Pro SSD to
Americans spend more money on first streaming music than CDs

Viewing:-212

In: Technology & Gadgets Asked By: [17584 Red Star Level]

Answer this Question

You must be Logged In to post an Answer.

Not a member yet? Sign Up Now »