acm-header
Sign In

Communications of the ACM

ACM TechNews

Dendrocentric AI Could Run on Watts, Not Megawatts


View as: Print Mobile App Share:
dendrite representation

A dendrite-based computational model could encode data in more than just base two.

Credit: Getty Images

A study by Stanford University's Kwabena Boahen proposes a method that would allow neural networks to run on watts drawn from a smartphone battery instead of megawatts of power in the cloud.

Rather than mimic synapses, Boahen's computational model emulates dendrites, where a neuron receives signals from other cells and which branch out to allow a single neuron to be connected with many others. The computational model is designed so that the dendrite responds only if signals are received from neurons in a precise sequence, which means each dendrite could encode data using higher base systems, based on the number of connections and the length of the signal sequences.

Boahen said a 1.5-micrometer-long ferroelectric field-effect transistor, comprised of a string of ferroelectric capacitors, with five gates could mimic a 15-micrometer-long stretch of dendrite with five synapses.

From IEEE Spectrum
View Full Article

 

Abstracts Copyright © 2022 SmithBucklin, Washington, DC, USA


 

No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account