Researchers at Lawrence Livermore National Laboratory (LLNL) argue that the human brain’s operation offers a useful model for the next generation of artificial intelligence (AI) hardware. The brain’s highly interconnected neural network delivers substantial computing capacity while consuming relatively little energy, and this efficiency has prompted growing interest in neuromorphic systems that mimic those biological principles.
Ionic computing: using ions instead of electrons
One of the most promising directions identified by the LLNL team is ionic computing, which processes information using ions rather than electrons. Ionic systems resemble the brain in several respects:
- storage and computation can occur at the same physical location, reducing energy-expensive data movement;
- they can operate at low voltages;
- they can use a variety of ions and molecules for information transmission, enabling diverse communication mechanisms.
Together, these features could enable far lower energy consumption compared with conventional semiconductor chips.
Why alternatives to conventional chips are needed
Most current AI systems run on traditional semiconductor devices that carry significant energy costs. Aleksandr Noy, an LLNL scientist and the lead author of the review, warned that modern AI development has become costly and energy intensive, and therefore new, low-energy solutions are required that can deliver comparable computational capabilities at a fraction of the energy cost.
Scientific and technological challenges
The field of ionic neuromorphic computing remains at an early stage, so the LLNL researchers surveyed the key scientific questions and technological gaps that must be resolved. Principal challenges include:
- developing new materials and architectures that efficiently support directed ion transport;
- improving the reliability and scalability of ionic devices for practical applications;
- ensuring interoperability between ionic systems and existing computing infrastructure.
Addressing these issues is crucial for transitioning the concept from laboratory demonstrations to usable technologies.
Where ionic neuromorphic systems could excel
The authors suggest that ionic neuromorphic computing need not directly compete with conventional chips across all domains. Instead, it may find success where extreme energy efficiency and biological or chemical compatibility are priorities, such as:
- brain–computer interfaces;
- in-sensor local data processing;
- environmental monitoring systems where low power operation is critical.
In these application areas, limits of traditional electronics are becoming increasingly apparent, and ionic approaches may offer relative advantages.
Publication and next steps
The LLNL team’s review was published in the journal Science in May of this year. The authors recommend intensified efforts in material research, device development, and architecture design to move ionic neuromorphic concepts from laboratory proof-of-concept stages toward practical systems.
Overall, the LLNL review highlights that the energy efficiency and operational principles of the human brain can provide valuable guidance for designing the next generation of AI hardware, particularly for use cases where energy and biocompatibility are decisive factors.
