BIOBRIEFPAPER
The 'brain inspired hardware' label hides a clean split: chips that imitate neurons, and lab grown neural tissue used as a processor.
The public label "neuromorphic computing" covers two parallel research programs that share a vocabulary but face different bottlenecks. A new 2026 Advanced Science review maps the split.
On one side sit silicon chips engineered to mimic how neurons spike and connect: subcellular device analogs, spiking neural networks, and brain-computer interfaces. On the other sit biological neural networks, including 2D and 3D lab-grown tissue, recruited as a computational unit. The review, indexed as PMID 42591065, treats both as "neuromorphic" while flagging that they are not the same engineering problem.
The silicon track chases lower-power, always-on sensing and faster on-device learning. The biohybrid track chases direct read and write between living tissue and machines, and inherits the field's hardest open questions: reproducibility, scaling, and the bioethics of using organoid-class tissue as a processor.
The two tracks share enough substrate (spiking neural networks, bioinspired learning, brain-computer interfaces) to remain one research community. The review surfaces the bifurcation and the milestones each side is now working toward.