Whole-cortex and whole-brain recording
The aspiration behind this track is whole-cortex and ultimately whole-brain recording: to observe the distributed activity that any real cognition or movement spreads across many areas at once. Today we live far from it. High-resolution electrophysiology covers patches; whole-brain coverage is available only at coarse resolution (imaging modalities) or sparse depth. Bridging that gap is the field's defining structural challenge.
The credible path is not one giant array but many distributed ones. Distributed modular implants — networks of small, independently placed nodes, potentially untethered motes — trade a single monolithic device for a swarm that can tile large territories without one impossible surgery. This inverts the design problem from 'build a bigger probe' to 'coordinate many small ones'.
The co-registration problem
A swarm of arrays raises a problem a single array never has: co-registration across arrays. To read distributed activity as one signal, you must know how each node's coordinate frame relates to the others, and hold that alignment stable as each node independently drifts. Stitching many partial, drifting views into one coherent picture of the brain is an open estimation problem, not a solved one.
Fundamental bounds and the read/write asymmetry
Beneath the engineering lie fundamental bandwidth bounds. Channel capacity is finite: for a given signal-to-noise ratio and usable bandwidth, there is a ceiling on bits no amount of hardware exceeds. These information-theoretic limits mean 'unlimited bandwidth' is not a coherent goal — the honest goal is approaching the bound the biology and physics allow.
A Shannon-flavoured ceiling: capacity grows only logarithmically in SNR and linearly in usable bandwidth B. Doubling channels does not double this — improving signal quality and usable bandwidth per channel does far more than piling on redundant electrodes.
There is also a stark read/write asymmetry: we can read from far more sites than we can meaningfully write to, because patterned stimulation that the brain interprets as natural is much harder than recording, and stimulation perturbs the very tissue it targets — the observer/perturbation limit. High-bandwidth output to the brain lags high-bandwidth input by a wide margin, and closing that gap is a distinct, harder frontier.
An honest reading
Separating results from aspirations: demonstrated today are thousands of channels chronically in animals and on the order of hundreds in humans, with clear clinical utility. Emerging are 10^4–10^5-site CMOS arrays, distributed motes, and on-implant compression that make larger scales conceivable. Open are truly chronic million-channel recording, whole-brain coverage at cellular resolution, and — just as unsolved — the theory of what to do with all that data once you have it. Progress is real; the timelines in popular coverage usually are not.