A memristor is a passive component like a resistor, capacitor, or inductor. It’s not widely known, as it was first postulated in the early 1970s. The concept can be boiled down to a resistor that changes resistance based on current, and has a “memory” of that resistance. When you push current one way, the resistance will rise up. You can lower it by pushing current in the opposite direction. Once it reaches 0 (or some threshold), it will then rise again. Alternatively, a memristor can become more conductive with current, returning to its resistive state when the current is reversed. This function is very similar to how the connections between neurons function, and thus memristors have become a focal point for creating neural networks on chips. For more information read https://www.americanscientist.org/article/the-memristor
In your brain, or in a computer AI, a neuron has a certain value to determine if it is “on” or “off”. In your brain, this value is stored but it’s the strength or weakness of that neuron’s connections to other neurons that matter in the big picture. Those connections function much like a memristor, once the neuron reaches some threshold of inputs, it “activates” and sends signals to other neurons. These signals are resisted/modified by the connections between neurons, the equivalent of a memristor. If you can create a reliable memristor, you can create an “neuron on chip”.
I believe I may have a functional design for a memristor that lowers resistance with current, and would like others to attempt my experiment (and improve on it), with the goal of ultimately creating a simple device, much like a resistor, that can be used in DIY/maker circuits. I have not attempted the TiO2 neuron discussed in the american scientist article, but may try to build one next. I’m going to call mine an “electrolytic memristor” — as it’s based on a NaSO4 / Deep Eutectic electrolyte, and aluminum’s passivation properties. It also appears to have an interesting “diode”-like behavior when operating.
Below is a diagram of what I constructed. It isn’t an ideal case, but shows the general layout of the parts involved.
We start with a length of resistive carbon. In my case, I drew a dark “resistor” with a pencil on a strip of tyvek (a plastic mesh material). I then cut that strip out, and laid it on a piece of kaptan tape. One each end I added a piece of aluminum foil as a current collector, and made sure to add more kaptan to insulate the collectors from the electrolyte. I then made an electrolyte with a Choline Chloride and Urea deep eutectic solvent, a little water, and a small amount of Sodium Sulfate (NaSO4), and some fine alumina (Al2O3) powder used in lapidary. I added another piece of tyvek as a separator and folded the cell over. Ideally, it looks like the diagram above.
My theory of operation involves pH and the passivation of Aluminum. Sodium Sulfate is pH 7, neutral. In a neutral solution, Aluminum forms a passivating layer of oxide that (mostly) prevents conduction of electricity. so, with no voltage applied, the resistor is at its “base resistance”.
When a voltage is applied (let’s say something like 2v), the U shaped resistor forms a bit of a hybrid resistor/capacitor/battery, with a gradient from 2v to 0v across the length. In the U shape, Sodium (Na+) ions will migrate to the negative terminal, and Sulfate (SO4-) ions will migrate to the positive terminal. This pH gradient happens over the length, but in our U shape, is going to be mainly near the top of the U. We begin to form Sulfuric acid at the positive terminal and Sodium Hydroxide at the negative. The Alumina (and any aluminum) near the positive will dissolve and cause the Aluminum ions (Al+) to migrate to the negative terminal, where they plate out on the resistor as Aluminum metal (Al0). This effectively lowers the resistance of the negative side of the U, as the electricity will conduct through the Al in preference to the resistor. In effect, it is an aluminum-based battery. There are a lot of side reactions here, but my hypothesis is that when power stops flowing, we now have a layer of aluminum on the resistor, all the way to the half way point. This would effectively halve the resistance of the entire cell at the extreme. Now, as the cell sits, the Na+ and SO4 will rejoin each other, and become neutral again. Our aluminum will then form a passivation layer and be protected from the electrolyte, holding this new resistance.
Apply current, and the cell reaches half resistance. Now, reverse the current, and we strip that aluminum and transfer it to the + side. If we stop part way, we reach a point where we have aluminum on both sides, but at 50% This doesn’t happen all at once, likely due to side reactions (AlSO4 in solution, Na[AlOH] as well) and appears to give us the ability to tune our resistance between 50% and the 100% as aluminum moves into/out of our electrolyte.
Here’s an interesting side effect. When we charge the cell, we end up with a diode-like behavior. One direction in my home cell will read 4M Ohm, the other 0 Ohm. This may be due to the battery-like nature of the cell, or some semiconductor interaction with the passivation layer and the pH gradient. it’s an interesting property, regardless… a programmable diode. This diode property (and the resulting resistance) seem stable as long as the voltage is underneath our 1.66v aluminum plating threshold.
One idea is to use this technique to make a memristor “diode” memory, since I can be tuned + or -, and isn’t affected by magnetic fields. So even if it isn’t a very precise memristor, it can hold some state, presumably forever once the passivation layer has been created, and if we stay under the Aluminum plating voltage.
Another interesting side effect of this method is the ability to “tap” or add additional output leads to the memristor. This effects a series of memrestors in one device, allowing voltage sensing at the tap to determine state of the memristor without pushing current through it. This might be an interesting way to monitor the state, or even create some other interactions. If the diode effect works like I believe I’ve seen in operation, this might be a way to create an electrolytic transistor that might be programmable NPN vs PNP.
I assume smarter people than myself will be able to figure out the math on this.
IS THIS A BATTERY? This is a fair question. It is.. sorta. A really really bad one. It’s shorted by itself (that U bend), so it should naturally tend to 0v over time while maintaining that Al/Al2O3 matrix. It does exhibit some capacitance, due to the two plates — and again the short. It’s kind of a weird hybrid of a capacitor, battery, and resistor that I believe end up handling the math required to make it a memristor. So it will never be a “TRUE” memristor for some people, and I’m ok with that. The idea was something very easy to build that exhibits memristor like behavior. Honestly, I’m more interested in the “programmable diode” aspect as a kind of 1-bit (or “fuzzy” bit if you want to measure between 0Ohms and 15MOhms) memory.
EDIT: someone pointed out some other memristors, and I’m particularly interested in this brass/copper/sulfide one ( Details @ http://sparkbangbuzz.com/memristor/memristor.htm and https://www.eevblog.com/forum/projects/making-a-memristor/ ) I’m mainly adding to this article for completeness as well as to study what he did. I’ve considered the Copper/CuO pourbaix for a similar process for the electrolytic memristor, which might be much more sensitive.
EDIT: Thought I’d add some build pics. Next build I will make sure the negative goes out the other side, so the whole mess can be rolled up into a “can” design once built. Note the center tap — which will result in two cells in series. I’m not sure it’s needed, but my hope was to use it to read voltage while the thing charges.
I’ll edit later and add some charts/diagrams when I get some time to build an arduino tester/logger.