For better computing, look at the brain

The human brain is the most efficient data system we know of, partly due to its ability to perform computations in the same locations where memory is stored.

Artificial computing, on the other hand, suffers from a data transfer bottleneck, especially as big data and artificial intelligence continue to grow. As the amount of data in the world approaches 200 zettabytes — over 10¹¹ terabytes — by the end of 2025, conventional computation struggles to keep up not because of hardware limitations, but because of the speed at which data moves between processing and memory storage. To address this shortcoming, Paolo Fantini makes a case for adapting a bioinspired memory storage paradigm.

“Shuttling around all this data carries with it a significant power and performance price when compared with the human brain,” Fantini said.

Key to the brain’s ability to conduct in-unit computations is the way in which connections between neurons change to strengthen or weaken information storage. This process, called spike-timing-dependent plasticity (STDP), updates the strength, referred to as weight, between synapses depending on the relative timing of the brain’s response to stimuli.

“An electronic analogue of a biological synapse needs the ability to exhibit STDP in order to mimic the synapse functionality,” Fantini said. “It has been shown that many emerging memory devices can reproduce the synapse plasticity, fitting the biological behavior of synapses.”

According to Fantini, the ability of these emerging memory devices to incorporate the synapses’ plasticity will enable new computing paradigms, creating better performance and overcoming the computational bottleneck.

Source: “Memory technology enabling future computing systems,” by Paolo Fantini, APL Machine Learning (2025). The article can be accessed at https://doi.org/10.1063/5.0253063 .

This paper is part of the Neuromorphic Technologies for Novel Hardware AI Collection, learn more here .