Penny For Your Thoughts: Electrolytic Neural Nets, Part 5

There’s been some exciting development in solid-state and discrete neural networks based on solid state memristors. In 2023, a paper was published explaining a design based on silver nanowire memristor networks, and there’s a lot of useful overlap with my own experiments building an electrolytic neural network based on stripping and plating of copper on resistive surfaces. It may be possible soon to craft neural networks in a home lab.

Memristors

Resistors are a simple passive component of electronics, they resist the flow of power, and allow for controlling the amount of energy a circuit may consume, as well as divide voltage. Generally, ignoring heat losses, a resistor’s value doesn’t change. A memristor will change it’s resistance as power flows through it. Some become more resistive, slowly turning off the flow of water, others become less resistive — opening the floodgates as power flows. The technology is much younger than other types of components, with much of the research being done only in the last few decades.

Neural Networks

Neural Networks have been used in hardware and software for over 80 years, but have gained a lot of attention lately in the age of generative AI and Large Language Models like ChatGPT. A Neuron in the network aggegrates inputs from other neurons, and then — if some threshold is met — will fire and pass values on to the next layer. This usually is a sharp line, called a “selection function”, and not just a…