University of Missouri researchers are developing a new way of building computers that takes inspiration from the human brain to solve growing energy problems in artificial intelligence. Energy use by artificial intelligence data centers is expected to double by the end of the decade, raising concerns about sustainability and cost.
The solution involves neuromorphic computing, an approach that reimagines computer hardware to work more like biological neural networks in the brain rather than traditional chips. The brain performs complex tasks using only about 20 watts of power, roughly the same as a small light bulb, while current computers are far less efficient.
Rethinking the computer chip
For decades, computers have used transistors. In ordinary chips, however, processing and memory storage occur in separate locations. Data must constantly move back and forth between these areas, which slows performance and consumes large amounts of energy.
The brain handles information differently. Connections between brain cells (synapses) perform both processing and memory storage at the same place. This design allows efficient learning and adaptation with minimal power. The Missouri researchers are creating organic transistors made from carbon-based materials that copy this synaptic behavior. These devices can store and process information simultaneously in one location.
Tests of several similar-looking organic materials showed that performance depends heavily on the interface, the thin boundary where the semiconductor layer meets an insulating layer inside the device. Small differences in material structure at this boundary led to large differences in how well the transistors worked.
The findings offer clear principles for designing better neuromorphic hardware. Such systems could enable artificial intelligence that learns more efficiently, uses far less electricity, and excels at tasks like pattern recognition and decision-making. Although brain-inspired computing remains in early development, these advances narrow the gap between biological efficiency and machine performance.
This study is published in ACS Applied Electronic Materials.