Physics / mechanism
Organoids are three-dimensional aggregates of cultured cells that self-organise into tissue-like structures. The best-characterised application in the sources is the brain organoid: a spheroid of human neurons and supporting cells used both as a model of neural development and, more speculatively, as a computational substrate. Growth is driven by the cells’ own developmental programmes rather than by external patterning, which means the resulting structures are heterogeneous between batches; the authors of the Brainoware work describe organoid generation as “uncontrolled, heterogeneous” 2023 Nature Electronics Brainoware.
The dominant physical constraint is mass transport. Without vasculature, oxygen and nutrients reach only about 300 to 500 µm into the tissue by diffusion, producing a necrotic core and capping both achievable size and maturity 2023 Frontiers Organoid Intelligence Roadmap. Current organoids sit below roughly 100,000 cells, around one three-millionth of a human brain, against a stated target of some 10 million neurons for useful computation 2023 Frontiers Organoid Intelligence Roadmap. Vascularisation is identified as the critical blocker.
Input/output is the second constraint. Coupling to a 3D spheroid is typically achieved with high-density multi-electrode arrays that contact only the surface, so interior activity is neither readable nor addressable at useful bandwidth 2023 Frontiers Organoid Intelligence Roadmap. In the reservoir-computing configuration, the organoid is treated as a fixed nonlinear dynamical system on an HD multi-electrode array: stimulation patterns are injected, the resulting spatiotemporal activity is read out, and only a linear output layer is trained 2023 Nature Electronics Brainoware. Maintenance overhead is non-trivial, described by the authors as a “24/7 task” 2023 Nature Electronics Brainoware.
The energetics argument underpinning interest in organoid compute compares a brain at roughly 20 W with a supercomputer at roughly 21 MW for comparable throughput, an implied efficiency gap of about 10⁶, though the roadmap authors flag the comparison as apples-to-oranges 2023 Frontiers Organoid Intelligence Roadmap. Several authors of that roadmap hold equity in organoid companies including Cortical Labs, AxoSim and TISMOO 2023 Frontiers Organoid Intelligence Roadmap.
Competitive landscape
Organoids as a computing substrate compete against conventional silicon on tasks where silicon baselines are trivially strong. The peer-reviewed ceiling is about 78% accuracy on speaker classification from a pool of eight on the Japanese-vowel dataset, plus better-than-untrained-ANN prediction of a Hénon map 2023 Nature Electronics Brainoware. Named senior neuroscientists dispute the whole approach rather than its current performance: Tony Zador calls the read-out-the-dish method “misguided” and a “scientific dead-end”, and describes useful organoid computation as “completely beyond what we could even conceive of right now” 2025 Statnews Biocomputing Backlash. Madeline Lancaster, an organoid pioneer, calls organoid-intelligence claims “very much science fiction” 2025 Statnews Biocomputing Backlash.
Adjacent to the compute thesis, organoids sit within a broader toolchain for observing and engineering 3D tissue. Volumetric imaging methods aim at diffraction-limited reflection tomography of thick samples under monochromatic illumination, while DNA-based spatial encoding proposes to read out molecular state, ancestry and physical neighbourhood inside intact specimens without a microscope. Both address the same underlying problem as the organoid I/O bottleneck: extracting structured information from the interior of a three-dimensional living sample.
Evidence base
- Organoids used to date contain under 100,000 cells, roughly 1/3,000,000 of a human brain, against a target of about 10 million neurons; diffusion limits viability to 300 to 500 µm depth, making vascularisation the critical blocker 2023 Frontiers Organoid Intelligence Roadmap (2023-02-28).
- A human brain organoid used as a reservoir on an HD multi-electrode array achieved about 78% accuracy classifying speakers from a pool of eight, and predicted a Hénon map more accurately than an untrained ANN baseline 2023 Nature Electronics Brainoware (2023-12-11).
- The Brainoware authors characterise the work as proof-of-concept, with heterogeneous organoid generation and 24/7 maintenance; commentators place general biocompute decades away 2023 Nature Electronics Brainoware (2023-12-11).
- Sergiu Pasca warns that “overly expansive claims can confuse the public and policymakers” and has organised an Asilomar-style ethics meeting; Madeline Lancaster warns blanket regulation could “prevent all work” 2025 Statnews Biocomputing Backlash (2025-11-17).
- Programmable spatial coherence tomography demonstrates diffraction-limited volumetric reflection imaging under monochromatic illumination, jointly retrieving aberrations, illumination profiles and sample motion without guide stars (2026-05-08).
- An NSF CAREER award of $688,712 funds volumetric DNA microscopy to capture cell ancestry, molecular state and physical neighbourhood together in a single intact specimen (2026-07-20).
Frontier (open questions)
- Can a vascularised or perfused organoid sustain viable tissue beyond the 300 to 500 µm diffusion limit, and what cell count does this reach relative to the stated 10 million neuron target 2023 Frontiers Organoid Intelligence Roadmap?
- Can any published organoid system exceed the 78% eight-class benchmark, or match a silicon baseline on any task at equal or lower total energy including culture maintenance 2023 Nature Electronics Brainoware?
- Does an interior-addressing I/O method (3D electrode penetration, optical, or DNA-encoded readout) achieve bandwidth beyond surface-only multi-electrode contact?
- Do batch-to-batch reproducibility metrics exist for organoid generation that would let a training protocol transfer between independently grown organoids 2023 Nature Electronics Brainoware?
Synthesised 2026-08-31 from 7 KB sources by the resynth pipeline; citations are KB source slugs.