<aside>

“It is astounding how much effective computation gets done in the 20 watts in our brain. And that is really what we set out to try to figure out when we started the whole neuromorphic thing. We wanted to understand that phenomenon: how can it possibly be?” — Carver Mead (2022)

</aside>

The von Neumann architecture powering most of our computing devices separates the functions of memory and processing that is very unlike the brain. Similarly, the Artificial Neural Networks powering modern LLM based AI systems are only loosely modeled on the behavior of actual neurons. They are grounded in a discovery made in 1943.

Neuromorphic computing is an approach to designing computer hardware and software that takes direct inspiration from the structure and function of the human brain than traditional computing approaches.

Neuromorphic computers are “non-von Neumann computers whose structure and function are inspired by brains and that are composed of neurons and synapses.”

Opportunities for neuromorphic computing algorithms and applications (Schuman et al. 2022)

These systems make use of energy-efficient approaches modeled after biological neural networks that hold the promise of “consuming far less energy than current AI and behaving much more like the intelligence that’s trying to design it.