In August I came across this article by Caltech researches using light and 3D printed layers to act as a neural network. I was amazed by this technology but wanted to build a more robust system that can't break from a simple tap. I also talked with Professor Siddhartan Govindasamy and he brought up the idea of using analog circuitry to replace digital circuitry. This lead me to the idea of making neural networks, a very computationally intensive digital process, analog.
What it does
My simple neural network shows the possibility of building analog neural networks at large scales. While my network can perceive the numbers of 0-3. Its 3 inputs can either be at a HIGH or LOW state. Based on the number of high inputs the corresponding output will be HIGH (if two inputs are HIGH the second output will be high).
How I built it
I first simulated everything using LTspice, then I used 20-year-old OP400AY op-amps that were laying around my house as the logic and mostly 11K resistors.
Challenges I ran into
The op-amps are very sensitive to any noise introduced such as a loose connection. This made me have to spend extra time debugging my circuit making sure everything is as secure as possible.
Accomplishments that I'm proud of
Ultimately it works!!
What I learned
I was able to expand my knowledge of analog circuits with this project. This was also the first time I used op-amps.
What's next for ANNe
I hope to write a program that takes a neural network and converts it into a dye that can then be manufactured and inserted into a bigger project.