The human brain is the ultimate supercomputer. It uses a highly-branched and interconnected network of neurons and synapses to achieve massive computational power with extreme efficiency. In the age of AI, the brain, a paradigm of efficient neuromorphic computing, is providing inspiration for scientists.
Ionic computing - which uses ions to compute instead of the electrons in typical devices - could provide a path forward for neuromorphic technology that rivals the brain's efficiency. But the field is only a few years old, and many challenges remain before it moves beyond proof of principle and toward real-world deployment.
To bring neuromorphic ionic computing research to the next level, researchers at Lawrence Livermore National Laboratory (LLNL), along with colleagues at other national laboratories and academic institutions, outlined the key scientific questions and knowledge gaps that must be addressed. They hope the work, published in Science, will guide the developing research agenda for the field.
Current neuromorphic computing platforms are based on conventional electronics that contain complementary metal- oxide semiconductor, or CMOS, devices. These chips are at the heart of modern AI, but they are nowhere near as energy efficient as the human brain.
"Modern AI is very costly and very power-hungry, and has entered an unsustainable development trajectory," said LLNL scientist Aleksandr Noy, who led this work. "We need a low power solution that can deliver some of the same computational capabilities at a fraction of the energy cost."
While other neuromorphic approaches are also under active investigation, ionic systems offer a unique set of capabilities because of their biocompatibility and similarly to biological systems.
The brain has a few key features that ionic computing could emulate. It stores and processes information in the same place, eliminating the costly shuffle of information between memory and CPUs. Its massive network allows the brain to adapt and cancel out noise. The brain operates at a low voltage and frequency, directly benefiting its power efficiency, and it can use multiple different ions and small molecules to transport information rather than just a single carrier like an electron.
Making those features a reality will require new materials with enhanced ionic properties and new architectures that facilitate the ion movement like, for example, those based on nanofluidic phenomena. These devices will also need the ability to interface with existing computing technology.
As neuromorphic ionic computing progresses, the authors suggested that it should not try to compete with CMOS-based technology. Instead, it should focus on applications where energy efficiency and chemical or biological compatibility are crucial, like brain-computer interfaces, in-sensor computing and environmental monitoring. These are domains where conventional electronics face fundamental limitations that ionic systems are better positioned to overcome.
This publication was a multi-institutional effort that brought in researchers from around the U.S. to contribute unique perspectives and expertise. That collaboration, Noy emphasized, must continue.
"Large teams can provide the critical mass of skills and knowledge that is required to accelerate the progress in this field," he said.
