Learn Your AI Modeling ABCs ….. Neural Network Behavioral Models with LTSPICE …….. Mark Stansberry ….The StatisticsMatrix
Contact: mark@statisticsmatrix.com
Professor Carver Mead was one of the first well-known scientists and engineers that calculated, given the right design, materials and manufacturing technology, that analog circuit design had the potential to produce artificial intelligence circuitry that has orders of magnitude better performance than digital artificial intelligence circuitry.
Models of neural networks suggest that information is learned or memorized through repeated pulsing of synaptic or neural cells. Current pulses, or charge, can be used to change the conductance of neural cell connections, allowing them to transfer information between neural cells and store that information. Additionally, continued pulsing of neural cells creates action potentials. As charge builds up on a neural node, that node’s charge and the other nodes’ charges add up to trigger an action, like moving left or right, or even falling asleep.
The Integrate-and-Fire (trigger) model is well-known in the neural circuit community. The neural behavioral model illustrated below, (which could also be considered a memristor behavioral model) performs two functions. One, it sums the accumulated charge of current pulses and two, changes the conductance of a resistor. In other words, it models a charge controlled voltage source and a charge controlled conductance.
In an actual application, thousands of these models would be used to simulate a working neural network. The charge controlled conductors perform two functions in a neural network. The first purpose is to change the interconnect conductance between neural cells, and the second purpose is to store learned analog information. Scores of, if not thousands of the charge controlled voltage sources feed scores of if not thousands of action potential neurons. When the sum of the charge controlled voltage sources reaches the learned trigger level, different action neurons will fire. That learned level is set with the charge controlled conductance that is stored within the action potential neurons (the memristor’s conductance value).
For more information about this and other AI electronic circuit behavioral models, contact me via LinkedIn. My contact information is on my LinkedIn profile.
#memristor #EDA #neurons #ai #SOC #ASIC #IPCore #analog #LTSPICE #SPICE #circuitdesign #behavioral #pulse #health #science #famouspeople #conductance #resistance #charge #current #innovation #model #mixedsignal #digital #cost #power #speed #efficiency #semiconductor #manufacturing #design #integratedcircuit #integrator #calculus #carver #resistance #conductance #synthesis #synapse #cell #macro #block #simulation #memory
