A (0:02)

A few weeks ago, a Patreon member sent me a paper titled Axon Like Active Signal Transmission by a team at Texas A and M, Stanford and Sandia National Laboratories. The paper discusses how the team recently transmitted a signal in an experiment. Big whoop, right? But the way they transmitted the signal is interesting because it mimics how neurons do it. Self amplification without additional devices. The result also involves a theory named the Edge of Chaos. Now, who can ignore that? In this brief video, I want to check out how they sent a signal through a wire. If an electric signal is sent through a wire, that wire, unless it is a superconducting one, will cause that signal to lose energy. This loss is because electrons traveling through the wire collide with the atoms inside the wire. The interconnects of modern integrated circuits are mostly made from copper, which all things considered, is a good conductive material. I did a video about how ICs made that transition many decades earlier. A fun one. But more than just wire material, shape and length also matter. The signal in a wire will decay in an amount that is directly proportional to the wire's length or inversely proportional to its width. This resistance, combined with another thing called wire capacitance, causes RC delay, a time delay in the signal's arrival at its destination. This affects performance, especially as chip features shrink and more transistors go onto the ic. So chip designers break up the wire lengths and insert what are called repeaters in between them to restrengthen the signal and drive it faster. Kind of like the IC version of Kettleman City on the 365 mile drive between Los Angeles and San Francisco. Leading edge integrated circuits might have thousands of these repeaters all over the chip. Inserting repeaters has trade offs. There is still some time delay. And since the repeater is actively powered, some additional power is necessary too. But chip designers deal with it because inserting repeaters is something that has been done for over a century. People inserted repeaters into the old big wired telecommunications networks of yor too. The worry is that down the line, interconnects become the dominant determinant of a chip's performance and power profile. Either designers suffer ever worsening propagation delays as the copper interconnects shrink, which ultimately affects the chip's speed and in some cases, reliability through time mismatch failures. Or the alternative is to have repeaters everywhere, which jacks up the chip's energy use and generates far, far too much heat. So with that in mind, let's take a look at our biological neurons. The nervous system's equivalent of a long distance cable is the axon. Certain neurons have these axons for transmitting electrical impulses to the dendrites of distant neurons. These axons, much like the rest of your body, are made from human flesh, which is not particularly known for its electrical conductivity and is certainly nothing like copper. Resistance is such that electrical signals can't be transmitted more than a millimeter, but before they lose their information. So how does a signal travel from the brain to a faraway toe without experiencing decays like that found in silicon? And these are not short distances. The longest human axon is maybe a meter long, and the axons of some whales can get to be 30 meters long. Moreover, the brain and its nervous system are famously energy efficient. How does the axon do all this long distance propagation without needing data center like amounts of energy? The path to the answer is that the axon has its own repeaters built in, creating a self amplification effect throughout its whole length. Normally, the axon's inside is negatively charged as compared to its outside, with an excess of potassium ions. Outside the axon, we have an excess of sodium and chloride ions, the former of which is positively charged but can't get in because of ion channels in the axon's lipid protein skin. Then the neuron decides to send a signal or spike. This usually happens when its input currents breach a critical threshold. When that happens, the ion channels in the lipid proteins in the axon's skin open, letting in positively charged sodium ions from the outside. This causes what is called a depolarization. This positively charged spike does not last long. Soon after the proteins close and the sodium ions return to the outside, the axon's inside returns to its old negatively charged state. This is repolarization. If that was all there was, then this wave of positive ions would not travel very far, certainly not 30 meters of whale. So we must repeatedly refresh the positively charged wave as it travels towards its receiving neuron. This is the job of our little biological repeaters, the so called nodes of Ranvier. Most of the axon's length is wrapped in an insulating fatty lipid sheath called myelin. The myelin blocks the axon's ion channels from the outside. But every so often we have these gaps about a micrometer long in the myelin. These are the nodes of Ranvier, and they have clusters of ion channels that allow an influx of new ions to regenerate the spike at the next segment of the axon. Thusly, the nodes help push the signal down the axon segments towards its final destination. This effect was first discovered by the British scientists Alan Hodgkin and Andrew Huxley that had been studying the axons of the giant squid, the longest known in Mother Nature. And for this they won the Nobel Prize in Medicine in 1963. For over a decade, R. Stanley Williams of Texas A and M wondered if there was any way to replicate this effect in a non biological way with an eye towards rejiggering how computer chips are designed and made. Williams had been part of the Hewlett Packard team that discovered the first implementation of the memristor back in 2008, and I covered this in a prior video. Since then he's been working on discovering some of the memristor's other properties and capabilities, as well as other ways to utilize human neural inspired properties to solve complex problems. To Williams and his team, the ideal way to replicate the axon's self amplifying behavior would not be to literally copy its physical structure. That's a meat experiment. But rather they should copy its theoretical behavior. The key thing about how these axons propagate signals is that each segment draws power or resources from its own local area, as opposed to relying on the energy of a central authority to drive the signal. Yet how do the proteins in the axon segment know when or how to help propagate their part of the spike if they're not being told to do so from a central authority? This requires exploring something known as the Edge of Chaos. The Edge of Chaos name first appeared in a chapter written by Norman packer in a 1988 book called Dynamic Patterns in Complex Systems. The name is very 1980s action movie and the idea has been around for longer than that. Roughly, very roughly speaking, it says that systems balancing between chaos and orders can find some sort of semi stable state. And in doing so, they can be useful for communicating important information and even running calculations. This kind of makes intuitive sense to me. In a previous video I said that the human brain turns its own inherent chaos into a feature. Modern VLSI digital chips instead depend on highly coordinated signals sent around the chip. If the wrong data arise at the wrong time, the wrong, then the chip slows down, or as mentioned earlier, might even crash. But the brain's 86 billion neurons do not and cannot have that coordination. The neurons fire spikes independently, based only on their own past experiences as well as the spikes they receive from their neighbor neurons. You might expect such a system to end up being total insanity and chaos. Yet the brain can somehow organize These independently firing neurons to create enough orderliness to perform calculations and do tasks. I must admit this concept is not easy to digest. Being on the edge of chaos grants flexibility. For example, think of a high performance jet fighter. These planes are intentionally designed to be so aerodynamically unstable that they have to be partially computer controlled. But that same instability also lets the planes be exceptionally agile and responsive to its pilot's commands. Flipping, diving and doing all that Top Gun Maverick stuff. This is an edge of chaos state. How do you determine whether a system is at the edge of chaos? Look at how it responds to external stimulus or perturbations. Think of two systems, one too stable and the other too chaotic. If a system is too orderly, then an external signal basically vanishes into said system without propagating. This makes the system too stable, rendering it predictable and useless. Like as if ChatGPT just replied with the same answer to every prompt you gave it. Or if you pushed a floating beach ball into the water and it instantly pops back up. But the system that is too chaotic is too easily overwhelmed by every external signal, rendering it totally useless for problem solving. Like a chatgpt that replies in insane ramblings, you still need to be somewhat coherent. Edge of chaos systems like our brain can somehow occupy this middle ground. Chaotic enough to amplify useful signals, create original thought and do useful things, yet also orderly enough not to go totally insane. The neuron's axon is said to operate at an edge of chaos state, and that helps contribute to its self amplification effect. The spike entering the axon is the external signal entering the system, and it is just a very small signal, a tiny flood of ions. Yet despite the axon having so many individual cells acting on their own accord, they are all somehow able to respond to this very tiny spike to amplify and propagate it. In 2012, Professor Leon Chua, most well known for first proposing the memristor, but also somewhat known because his daughter Amy Chua, popularized tiger mom parenting, published a paper making this connection. It is quite math heavy I have to say, and rather difficult to read. I will basically beg to bypass explaining it and just say that it presents a pathway for Williams and his team to replicate the effect in non biological materials. The team chose to run their experiment with a material called lanthanum cobalt oxide or lanthanum cobaltide. They chose this material because of an exotic behavior it has. It is internally unstable because it becomes abruptly much more electrically conductive as it heats up. So the interesting thing is that when an electric signal Passes through something, that something's resistance generates heat and weakens the signal. But in the case of the lanthanum cobalite, that heat also causes the material to become less resistant, creating interesting feedback effects. This gives you this rather unstable electrical behavior that the team sought to take advantage of for their experiment. They first deposited a layer of the lanthanum cobalite onto a substrate of lanthanum aluminum oxide. Then, on top of that, the team deposited two lines made from a resistive material, platinum. These platinum lines were made to be as long as an average interconnect line, about 200 to 400 micrometers. Just in case you were wondering, platinum is about six to seven times more resistive than copper and is not generally used in semiconductors. Through the wires, the team applied a small external direct current to the lanthanum copilot sheet. This caused the sheet to achieve edge of chaos conditions capable of amplifying signals without the need for more devices. The team applied an alternating current voltage signal at one end of one of the wires. Normally then you would find the signal comes out the other end with some decay. But this time, when they measured the other end, they found the same signal to have been amplified by as much as 70%. The transmission line absorbed the energy applied to the lanthanum cobalite sheet and used it to amplify the signal and keep it from decaying. Just like how a segment of the axon absorbs ions from its environment to help propagate a spike. This was because of the edge of chaos conditions achieved in the sheet, which made it more amenable to amplifying small signals, again without the need for custom built repeaters to be inserted into the line. Impressive. So what's next? The immediate next step would be to find out what other materials can also produce these results. And then seeing if the wider semiconductor industry might be interested in integrating it into their recipes. The semiconductor industry is starting to see the end of the copper interconnected, so a lot of new ideas are on the table. Is it possible down the line that future IC interconnects be made to self amplify? Well, probably not immediately. Right now, the best candidate after copper looks to be ruthenium. It will require some substantial changes in existing semiconductor recipes, but so far as I can tell, it's the most logical evolutionary next step. Self amplifying interconnects might be useful in some specific use cases. For example, might be interesting to see how they can help the global interconnects carrying power across these increasingly larger chips. At the same time. This is clearly a revolutionary type methodology, and I don't think it's realistic to ask intel or TSMC to tear down the way they produce chips for some power gains. So a great deal of work remains ahead to find something that can be logically inserted into the process. So new materials. But the idea of getting these interconnects to self amplify and removing many of these repeaters is very tempting. And I know that someone somewhere in the industry would be interested in making it happen. All right everyone, that's it for tonight. Thanks for watching. Subscribe to the channel, Sign up for the newsletter, and I'll see you guys next time.