Super small electronic variable resistor -- Exists?

[MeaningfromForm]

Hi I couldn't find of these exist anywhere.
Is there a component that is a variable resistor that adjusts based solely on receiving an up/down signal? And maintains its state on powerdown? I don't want this solution to involve any type of software, just a hardware solution. Looking for something that will work in the 8hz frequency less than 12v and 100mA
I couldn't find anything.

 

[davenn]

Hi
welcome to PF

Hi I couldn't find of these exist anywhere.
Is there a component that is a variable resistor that adjusts based solely on receiving an up/down signal? And maintains its state on powerdown? I don't want this solution to involve any type of software, just a hardware solution. Looking for something that will work in the 8hz frequency less than 12v and 100mA
I couldn't find anything.

Resistors don't do that without a lot of other surrounding components/circuitry
so what are you really trying to achieve ?cheers
Dave

 

[MeaningfromForm]

Hi
welcome to PF
Resistors don't do that without a lot of other surrounding components/circuitry
so what are you really trying to achieve ?cheers
Dave

Essentially I am trying to create a simple neural network using only electronics on a circuit board. This variable resistor I am talking about would act as a weight. And the next resistor in the line, whether it fires or not would represent the neuron

 

[MeaningfromForm]

Just googled analog neural network circuit, don't know why I didnt do that before. Yes that's what I want to create. From what I am reading technology seems to be limiting in terms of precision variable resistors or amplifiers that can maintain their state in the off position.

 

[davenn]

Essentially I am trying to create a simple neural network using only electronics on a circuit board. This variable resistor I am talking about would act as a weight. And the next resistor in the line, whether it fires or not would represent the neuron

I think you had better do some reading on resistors and what they are and other basic electronics

resistors don't do the things you are implyingDAve

 

[MeaningfromForm]

I thought I had... okay let me try to explain better what I want to do logically
I want an input to be adjusted either up or down then into fed into a gate which has an activation potential, so that over a threshold current it will pass through, and under it will not pass through.

Further more, the component which amplifies or reduces the signal needs to be adjusted by the circuit itself, so the circuit needs to have some sort of feedback loop to the "adjuster".

 

[Aaron Crowl]

There is only one thing that I can think of which will do what you want. Take a look at the memristor. Unfortunately it's cutting edge stuff that isn't available yet.

 

[MeaningfromForm]

Yea that's basically exactly what I am looking for. Unfortunate that its only experimental at the moment. I wonder if there is any suitable macroscopic equivalent, its response speed isn't super important, for example water with salt in it, the salt could be electrically removed to increase resistivity, obviously water and electronics is a terrible idea, but the first one that came to mind.

 

[anorlunda]

I know of no such device.

But you could use an external laser to ablate away material, making resistance higher. An external ion gun could deposit material, making resistance lower. If those external guns could be aimed, they could adjust a large array of resistors. It would take LOTS of engineering to make that work in real life.

If you always start with a virgin array, then only ablation is needed, skipping deposition.

 

[MeaningfromForm]

A

But you could use an external laser to ablate away material, making resistance higher. An external ion gun could deposit material, making resistance lower.

I like the idea, ofcourse like you said it would be difficult, too difficult of course, but these work around ideas are very welcome.

 

[Aaron Crowl]

Out of curiosity, why do you not want to create a digital design? You could even do this completely with software in a general purpose computer.

 

[MeaningfromForm]

Out of curiosity, why do you not want to create a digital design? You could even do this completely with software in a general purpose computer.

Because the computations performed by a processor to train a neural network are prohibitive. I believe its possible to design an analog neural network that can out perform a digital representation. Ofcourse if I can build it all, my first version won't out perform anything, but maybe after I build it, I can build it again, faster, with smaller components. Kind of like how vacuum tubes advanced.
Also I have an invention idea whoch requires a neural network that can be trained very fast, and very responsive.

 

[Aaron Crowl]

It sounds like you have an interesting problem. I'd like to help but I'm afraid I don't know anything about neural networks. I do know a thing or two about digital design. It is possible that something exists which you can buy off the shelf and use.

When someone says "computationally prohibitive" I start thinking about FPGA's. Have you considered using FPGA's? The Xilinx Zynq-7100 has about 2000 DSP blocks. With a realistic clock speed of 300MHz that means you could configure the chip to perform 600,000,000,000 multiply-add operations per second. Is that enough?

 

[MeaningfromForm]

Im sure that would work for what I want to do, but if I am relying on a classical processor I can never hope to compete with companies like google and their AI efforts.
In addition to the vast computational resources they have they also have mathmaticians working out complex algorithms to make those virtual nets learn.

I want to approach neural nets from a new perspective. I think an analog model can outperform a digital one. Consider this, a standar computers memory can only be a 1 or 0. If you think of the amplifiers setting as memory(which in terms of neural nets that's what it is) then the value can be 0, 1 or anything in between this is where the analog version makes its gains over the digital version which has to simulate these values with floating point variables.
The question is can I build it, and with how much complexity.

But as you suggested I could already create my invention using the already available technology, that is tempting and Ill have to give that some thought.
Thanls for your feed back.

 

[jim hardy]

Look into digital potentiometers.
Power off memory will be tough, though.

http://www.ti.com/lsds/ti/data-converters/digital-potentiometer-products.page#

EDIT Well well here's one with power-off memory
http://www.ti.com/product/tpl0102-100/description

 

[CraigHB]

I've used digital pots in a number of designs, real handy devices. Availability is fairly wide on them. The volatile type are probably more common, but there are non-volatile ones available that remember the last setting on power up. Some provide a one wire up/down interface, but an SPI or I2C interface is most common. You can find them in very small SMD packages.

BTW, Microchip has a wide range available. Maxim makes a 1024 tap pot which can be handy if you need higher resolution. Most are 256 tap and less commonly 512 tap. That indicates how many steps they have.

 

[anorlunda]

I want to approach neural nets from a new perspective. I think an analog model can outperform a digital one.

There's nothing wrong with that, although I suspect that it has been done before. However, you could implement the neural network with something analog, but do the learning with something digital, or do learning all software. The final neural weights, once learned, could be burned into an analog network.

You could even take an existing neural network (I think of Google's speech recognition network; the result of massive deep learning.) and have it burned into an analog network. Then see how much better the analog one performs. That would be a reasonable way to investigate the value of your ideas.

 

[Averagesupernova]

The closest I can think of that is analog and 'holds' would be a sample and hold circuit. It of course not a resistance but a voltage. Power down would also be a compromise here. If you want to get really crude a motor driving a potentiometer technically fits your requirement if you are liberal about defining super small.

 

[Tom.G]

Looks like you're a bit late.
https://www.google.com/?gws_rd=ssl#q=neural+network+chip+buy

 

[Baluncore]

Hi MeaningfromForm.
Maybe if you gave us the block diagram of the network you want to make, we might find a way to do it with much lower power and a lower voltage supply for the analogue circuits and pots. Some of us are good at mapping functionality to available analogue circuits.

 

[MeaningfromForm]

Nope it looks like what I wanted to create has already been invented so it's kind of pointless to continue, besides, I live in Edmonton Canada and I wouldn't be willing to share my idea with someone unless it was in person anyways...
Thank you everyone for the ideas, keywords to research and suggestions. Perhaps I will come back in the future and revisit this when I have a better understanding of the technical aspects.

 

[tech99]

I suppose a thermistor would retain its temperature for a short period and so exhibit a resistance which is determined by the last voltage applied.
Another possibility is an IGFET, where the gate-source capacitance would remember the last applied voltage.

 

[Baluncore]

I don't expect analogue to beat a digital processor emulating the analogue for the same silicon cost. We passed that point in the 1980s.

I think your standard cell could be designed like an analogue computer with a couple of integrators to remember the state.

Operational transconductance amplifiers can be driven by a voltage difference on the inputs and current pulses on the Ibc terminal, a capacitor on the output integrates charge which is the product of Ibc and the input voltage. One OTA with a capacitor gives you a difference amplifier, a multiplier and an integrator. Check out the LM3700 ( NE5517 ? ) which have two amplifiers and two buffers in one pack.

The LM3700 data sheet has plenty of OTA applications; http://www.ti.com/lit/ds/symlink/lm13700.pdf

 

[jim hardy]

just musing here from an old hardware guy

...\start daydream with boring anecdote
i once encountered in a TTL system a 3 level logic circuit that used "High" and "Low" TTL voltage levels but also detected and acted on a third level "In Between" ,
its purpose was to detect multiple slave responses to an address query from master.
Address_Acknowledge line driven high meant no response yet, middle meant single response (good) , low meant multiple responses(failure, program should try again)

That astonished me because i had never heard of anything but base two Boolean " High Low " for logic circuits.

I asked my old mentor and he said said "Sure, you can do math with any radix you want and it doesn't even have to be an integer. Mathematically it can be shown that a base e machine will require the least number of gates but i don't know of anybody who's developed a base e arithmetic book. "

Base e arithmetic? I can't even count in base e.

I soon figured out "well sure, i guess one could use any number of voltage levels for his logic,
and analog computing is just the limit as that number approaches infinity."
I bought a book on irrational numbers planning to learn how to count in an irrational number base maybe √2 but it was way too abstract for my flea brain.

That was almost forty years ago. Now i find lots of links on ternary (base 3) computing and even a few on base e. Try a search if you're curious .

As i said analog implies an infinite number of levels/values. There's a major practical drawback to analog circuits you have to keep them calibrated so that everybody attached to a particular signal agrees on its value . Some smaller number of logic levels might be easier to build and keep working..

Point being , i laud your investigating alternatives to binary computing and suggest there might be diminishing returns beyond some number of levels .
"e" being closer to 3 than 2 the point of diminishing returns might be around there . I find it a fascinating concept.

Seems to me you could build a cell out of something like a dollar Pic microcontroller
http://www.digikey.com/product-search/en/integrated-circuits-ics/embedded-microcontrollers/2556109?k=&pkeyword=&v=150&FV=fff40027,fff800cd&mnonly=0&newproducts=0&ColumnSort=0&page=1&quantity=0&ptm=0&fid=0&pageSize=25
controlling a digital potentiometer , a hybrid if you will,
and explore the behavior of multi-level computing machines.
It wouldn't get you to true analog though.\end daydream

sorry if I've bored you

old jim

 

[Aaron Crowl]

Nope it looks like what I wanted to create has already been invented so it's kind of pointless to continue, besides, I live in Edmonton Canada and I wouldn't be willing to share my idea with someone unless it was in person anyways...

You sound a little discouraged here. Don't be. From your earlier posts it sounded you had an idea for applying neural networks to some problem. That's a patentable idea. If it's something good then it can be monetized. You don't necessarily have to make the neural network hardware for this to be worthwhile. The application itself may be worth something. HP labs is working on making memristor nets on Silicon. If you have a nifty use for these when they come out then you are on to something.

 

[tech99]

Hi MeaningfromForm, I think everyone reading this thread is inspired by your ideas. Good luck with it.

 

[Averagesupernova]

Don't be discouraged because someone else beat you to it with some invention. When that happens think of it as your idea wasn't so horrible because someone else thought of it too. Just bad timing I guess, but it shows your thinking is on the right track.

 

[jim hardy]

You can patent an improvement to an existing one. But patent lawyers can be expensive. .

 

[Lish Lash]

The most flexible form of variable resistor is an optoisolator, usable in a variety of analog circuits that require a floating, non-polarized resistance element.

http://denethor.wlu.ca/pc300/optoisolators/analogoptoisolatorintroduction.pdf

 

[tech99]

The most flexible form of variable resistor is an optoisolator, usable in a variety of analog circuits that require a floating, non-polarized resistance element.

http://denethor.wlu.ca/pc300/optoisolators/analogoptoisolatorintroduction.pdf

But it does not have memory action.

 

[Lish Lash]

But it does not have memory action.

Indeed. That is one of the properties that makes an optoisolator very close to a pure resistance element. From an analog engineer's perspective, that is ideal.

 

[CWatters]

How about a FET with a capacitor on the gate. FET behaves like a voltage controlled resistor and the capacitor keeps it working while the power is off (for a short period at least).

 

[KenJackson]

I thought I had... okay let me try to explain better what I want to do logically
I want an input to be adjusted either up or down then into fed into a gate which has an activation potential, so that over a threshold current it will pass through, and under it will not pass through.

Further more, the component which amplifies or reduces the signal needs to be adjusted by the circuit itself, so the circuit needs to have some sort of feedback loop to the "adjuster".

This morning, after there was perhaps one reply to your original post, I started a post in which I was going to ask if you checked DigiKey, http://www.mouser.com/ , http://www.allelectronics.com/ or Fry's.

But Firefox crashed as it sometimes does on this PC, so I took the opportunity to upgrade packages. But I did something wrong and ended up with an incomplete upgrade which I've been fighting off and on all day. Finally, I got everything back to normal.

The other thing I'd say is what others almost mentioned. This sounds like an amplifier, which could be designed with almost any kind of transistor. Or you could us an opamp, which might be simpler.

 

[Alex163]

Base e arithmetic? I can't even count in base e.

I soon figured out "well sure, i guess one could use any number of voltage levels for his logic,
and analog computing is just the limit as that number approaches infinity."
I bought a book on irrational numbers planning to learn how to count in an irrational number base maybe √2 but it was way too abstract for my flea brain.

I don't want to appear to be pushing a product in my first post on this forum but given your interest, you may have already encountered this book - "Hacker's Delight" by Henry S Warren.
If you haven't, I would recommend it. There's a chapter on unusual bases that blew my rather closed mind when I first read it. There's a fun bit on base -1 +/- i, as well as a brief discussion on what base is "most efficient" from the point of view of cost. Standard binary is more costly than base e by a factor of 1.062 (According to the formula he used). This book might be useful for you, MeaningfromForm.

 

[anorlunda]

Nope it looks like what I wanted to create has already been invented so it's kind of pointless to continue, besides, I live in Edmonton Canada and I wouldn't be willing to share my idea with someone unless it was in person anyways...
Thank you everyone for the ideas, keywords to research and suggestions. Perhaps I will come back in the future and revisit this when I have a better understanding of the technical aspects.

It is one of the less peasant truths of the modern world. There are so many smart people on this planet, it is very difficult to have an idea that someone else didn't think of first. We who yearn to be original get frustrated by that.

 

[jim hardy]

I think he should build something anyway. We learn so much more from doing than from just reading about doing.

 

[Baluncore]

I believe its possible to design an analog neural network that can out perform a digital representation. Ofcourse if I can build it all, my first version won't out perform anything, but maybe after I build it, I can build it again, faster, with smaller components. Kind of like how vacuum tubes advanced.
Also I have an invention idea whoch requires a neural network that can be trained very fast, and very responsive.

The OP will not identify the conceptual model of the synapse or Neural Network that might use the component.

Nope it looks like what I wanted to create has already been invented so it's kind of pointless to continue, besides, I live in Edmonton Canada and I wouldn't be willing to share my idea with someone unless it was in person anyways...
Thank you everyone for the ideas, keywords to research and suggestions. Perhaps I will come back in the future and revisit this when I have a better understanding of the technical aspects.

The OP also admits that someone else has already invented the idea. But it is still kept a secret to PF.

To the small inventor, a patent is financial death. It costs real money to establish, then any challenge will destroy the small inventor with the incredible cost of the legal defence. The invention will then be available to all, without licence.

Publication of the idea is the best solution for all. That way no one can patent the idea, while others design the solution. But here the OP has chosen obscurity.

It is interesting to observe that this thread is a bit like a UFO sighting. The less information the OP provides the more interpretations are possible and the longer the suggestions and discussion will continue.

 

[davenn]

+1 what he said ^^^

and because of that I find these types of threads so pointless as everyone stumbles around in the dark
trying to outguess everyone else

 

[jim hardy]

To the small inventor, a patent is financial death. It costs real money to establish,

I'll vouch for that.
Though US patent office now has reduced rates for small entities .
Still it'll probably cost at least ten grand , much of that to a patent lawyer to translate your writeup into patent office jargon . It's a pretty strange dialect .

US7177383 is my $15K 'vanity press' misadventure .. Never again.

Get it working, submit an article to a hobby magazine, have fun.

old jim

 

[dlgoff]

US7177383 is my $15K 'vanity press' misadventure .. Never again.

At least you have a patent.

http://patft.uspto.gov/netacgi/nph-...50&s1=7177383.PN.&OS=PN/7177383&RS=PN/7177383

 

[CraigHB]

There's a huge problem with patents and consumer products. It makes them ineffective even if it's not a matter of money. A lot of the theft of IP occurs in China where US patents are not enforce-able. So you can spend all that time and money on a patent only for some Chinese outfit to copy your product and sell it for a tenth the cost. Granted the quality is almost always a lot better with an original product and that can count for a lot, but an inferior copy for a fraction of the cost deeply cuts into product sales for the original.

I've developed a couple consumer products myself and people often say to me, "you should get a patent on that." I don't bother to explain why it's a complete waste of time and money. The false paradigm in the US is that a patent actually provides some protection of IP to an individual. If an offending company does business in the US a patent is enforce-able, but still it's cost prohibitive. The only people that can benefit from the US patent system are the larger corporations that keep a team of lawyers on staff.

 

[anorlunda]

How to fix US Patent law sounds like a good thread for general discussions.

 

[Baluncore]

A lot of the theft of IP occurs in China where US patents are not enforce-able.

The problem of international asymmetry in patent law cuts both ways. For example, a US citizen could provide a signed statement, without proof or evidence to the US patent office that stated they were working on an invention before a foreign registered patent priority date. Preferred treatment to US citizens prevents some foreign patents in the USA.

If you patent your invention, the Chinese will have the information needed to copy it.

I did not patent my most profitable invention, but I hid the IP deep inside the product. Patents on other parts of the product expired in 1998. But the product then won UK trade awards in 2002 for “the most innovative product”. I manufacture ever increasing quantities of “the magic ingredient”.

There have been several failed attempts by others to clone the product, but without a patent they have no clue what a reliable solution entails, or how the hidden IP might overcome some quite surprising problems that I anticipated. After disappointing their customers in the first year, the clone manufacturers abandon the field to cut their losses.

My design has outlasted two eras of PCB manufacturing technology, so I am now completely replacing my PCB design software for the second time. I buy some electronic components from Britain and the USA, then manufacture in Australia. I also ship a finished electronics and interface kit back to British manufacturers for final assembly with steelwork. They export most of their finished product directly into Europe. China is not in the loop.

Originally, compliance with EU regulation was interesting. Now the thoughts of Brexit keep it interesting.

 

[anorlunda]

Happy 4000th post @Baluncore