IISc's Brain Co-Processor Moonshot Targets Stroke Recovery

The Indian Institute of Science (IISc) has launched what it calls a "moonshot" project to build brain co-processors, a programme, reported in early March 2026 and funded by the Pratiksha Trust of Kris and Sudha Gopalakrishnan, that aims to develop implantable and non-invasive systems able to decode brain activity, process it with artificial intelligence, and re-encode the result as stimulation or neurofeedback. The first clinical target is cognitive rehabilitation for stroke survivors, restoring functions such as reaching and grasping. Crucially, IISc intends to build the implant hardware, the neuromorphic computing systems and the AI software stack within India, and to create India-specific neural datasets, including stereo-EEG and electrocorticography recordings.

What a "co-processor" actually means

The phrase is deliberate. A read-only brain-computer interface listens to neural signals and decodes intent, letting a paralysed user move a cursor or a robotic arm. An open-loop deep-brain stimulator does the opposite: it writes a fixed pattern of electrical pulses into the brain regardless of what the brain is doing, which is how Parkinson's is treated today. A co-processor does both, and in a continuous loop: it reads neural activity, computes on it, and writes back stimulation that responds to what it just read.

For a stroke patient, that closed loop is the point. A stroke kills tissue and severs pathways; rehabilitation is the slow process of the brain rewiring around the damage. A co-processor that can sense a patient's intention to move, and deliver precisely timed stimulation that reinforces the attempt, could in principle accelerate that rewiring or bridge a broken pathway directly. The adaptiveness, responding to the brain moment by moment rather than blasting a fixed signal, is what separates it from the stimulation devices already in clinics.

The three hard problems

The ambition is enormous because it requires solving three hard problems at once, in a single device, at low power. The first is decoding: extracting a reliable signal of intent from neural data that is noisy, individual and non-stationary, drifting over time even within one person. The second is computation: processing that signal fast enough that the stimulation arrives while it is still relevant, which for a closed loop means latencies measured in milliseconds. The third is stimulation: writing back into the brain precisely enough to help rather than harm, in a system that has to be safe to implant or wear for long periods.

Each of these is unsolved at the level of robustness that clinical use demands, and doing all three together, in hardware small and efficient enough to be practical, is the mountain the project has chosen to climb. It is genuinely moonshot-scale, which is presumably why it is framed as one.

Why build the whole stack in India

The decision to build hardware, compute, AI and datasets domestically is the strategic core of the effort. Most of the world's neural-interface datasets are drawn from Western populations and Western clinical settings, which limits their relevance and raises real questions of data governance when Indian patients' brain recordings are involved. Creating India-specific stereo-EEG and electrocorticography datasets is foundational infrastructure, in the same way a country builds its own reference genome rather than borrowing one that does not represent its people.

The compute layer connects to other strands of Indian deep tech. IISc's own Centre for Nano Science and Engineering has produced indigenous neuromorphic hardware, brain-inspired chips built on molecular memristors that pack thousands of conductance states into a single device, exactly the kind of ultra-efficient, brain-like processing a low-power wearable neural system would want. Building the silicon, the algorithms and the data under one roof is an attempt to own the entire vertical rather than assemble it from imported parts.

The clinical pull, and the caveats

There is also a clinical pull already visible in the country. IIT Kanpur's brain-computer-interface robotic-hand exoskeleton for stroke rehabilitation, developed over years by Prof. Ashish Dutta with DST, ICMR and UK-India funding, reported encouraging recovery in early pilots, evidence that closed-loop, brain-driven rehabilitation is not science fiction and that there is Indian clinical demand to pull the technology forward.

The caveats are large and worth stating plainly. A moonshot is, by definition, years from the clinic. Invasive brain-interface work in India is at an early stage, and the headline-grabbing invasive milestones so far, implanted interfaces restoring speech or movement, have come from the United States and China. Specific principal investigators and the funding quantum were not all disclosed at launch, so this is a statement of intent backed by philanthropy more than a demonstrated device. But the intent is the significant thing: rather than wait to import closed-loop neurotechnology once others have proven it, India is choosing to try to build a sovereign version, hardware, algorithms, data and all, with a clinical problem of genuine national scale as the target.