Research analysis

August 4, 2026

When the reservoir does not do the remembering

A memoryless quantum dot, in simulation, solves a memory-hungry forecasting task once its past inputs are fed in parallel rather than remembered internally. For organoid intelligence it questions whether the tissue's own memory is the ingredient worth paying for.

9 min read · Read → August 4, 2026

When spiking features fit brain data better

Swapping the target features in an fMRI decoder from a standard network to a spiking one improves how well human visual cortex can be read, with the decoder held fixed. For organoid intelligence it is a readout-design hint wrapped around a hype-correction.

8 min read · Read → August 4, 2026

When the key to a network is a moment in time

A copyright method trains a spiking network to work only when an authorised token appears in a specific time slice, and to fail for everyone else. For organoid intelligence it is a distant analogy for owning a living computer, and a warning about where that analogy breaks.

9 min read · Read → August 3, 2026

Delay plasticity as a missing organoid substrate

A mathematical model makes axonal conduction delays a plastic variable that activity-dependent myelination tunes to lock spiking networks into commensurate timing. It names a slow computational layer that cortical organoids, small and largely unmyelinated, mostly fail to build.

9 min read · Read → August 3, 2026

Extrinsic memory and the wetware premise

A simulated quantum dot with no intrinsic memory solves memory-hungry benchmarks once past inputs are spread across spatial channels instead of stored in the system's own dynamics. That space-for-time trade puts a hard question to the organoid-as-reservoir story.

9 min read · Read → August 3, 2026

Realigning a decoder as living tissue drifts

A pretrained spiking decoder keeps working across a month of primate reaching by matching its internal membrane-potential distributions to each new session, updating under nine percent of its weights. For organoid computing, the lesson is that most of the adaptation has to sit in silicon, not in the tissue.

9 min read · Read → August 1, 2026

Moire graphene synapses and substrate uniqueness

A twisted double bilayer graphene device shows analog potentiation and depression plus a gate-tunable, sign-reversible second-harmonic readout from pure carbon band-structure physics, with no ions and no charge traps. It is a cryogenic single Hall-bar device, not a synapse, but it weakens the argument that only living cells deliver efficient tunable plasticity.

8 min read · Read → August 1, 2026

SpikeTimer and the spike-timing access credential

SpikeTimer trains a spiking neural network to classify correctly only when an authorization token sits in the correct temporal slice of the input, collapsing to near random guessing otherwise. It is a simulation-only copyright scheme, but the spike-timing channel it exploits is exactly where a future living-tissue computer might carry, or hide, an access credential.

8 min read · Read → August 1, 2026

SNN features as better fMRI regression targets

Using one fixed linear decoder and varying only the target feature space, this study finds spiking-network features are far better regression targets for human visual-cortex fMRI than rate-based ANN features. It is a human-fMRI alignment result, not mind-reading and not organoid work, but better brain-aligned readouts carry both an opportunity for neural substrates and a genuine downstream privacy concern.

8 min read · Read → July 31, 2026

When variability becomes an identity, not a defect

A 65-nm chip harvests transistor process variation as physically unclonable entropy for hyperdimensional computing, measured across 100 dies. Read as an analogy, it reframes organoid non-reproducibility as an attestation asset, and exposes an un-auditable governance edge.

9 min read · Read → July 31, 2026

The genotype hidden inside a living computer

An active organoid programme shows single autism-risk mutations retime the inhibitory neuron lineage relative to the excitatory neurons it must wire with. For organoid intelligence, donor genotype is a substrate excitability variable that coarse quality control cannot see.

9 min read · Read → July 31, 2026

Silicon takes a task wetware was pitched for

A spike-driven transformer classifies raw radio waveforms end to end on a neuromorphic chip, posting a measured dynamic-power and throughput win on a temporal low-SNR task. It narrows the case for tissue reservoirs, though a mixed total-power picture keeps the obsolescence claim in check.

9 min read · Read → July 30, 2026

Event-driven dataflow, and where a substrate's presumed sparsity edge leaks

A neuromorphic accelerator recovers up to an order of magnitude in efficiency only by keeping every layer event-driven, showing that spiking sparsity is worthless if any stage re-densifies it. The same trap sits at an organoid's electrode boundary, where a conventional acquisition chain samples the tissue at full rate regardless of whether it spiked.

9 min read · Read → July 30, 2026

A cheap salience gate, and what it decides for a closed-loop organoid

A lightweight spiking network learns to flag salient audio so a heavy classifier runs only when it matters, cutting projected compute by an order of magnitude. Ported to organoid intelligence, that gating pattern lowers the dose a living substrate absorbs but quietly bounds what the system can ever observe.

8 min read · Read → July 30, 2026

Superposition coding for spikes, read as a lesson for organoid readout

A wireless protocol lets many neuromorphic sensors share one channel by algebraically bundling their spike streams and unbinding them at a shared decoder. The mechanism is a blueprint for reading many channels off a living substrate at once, and a caution about why tissue will not hand you the keys.

9 min read · Read → July 29, 2026

One-shot forecasting of noise-driven regime shifts

A reservoir trained at one noise level reconstructs how a system will bifurcate at other noise levels, and cancels dynamic noise, but only when the noise is uncorrelated in time. That caveat is exactly where living reservoirs differ.

9 min read · Read → July 29, 2026

Self-caused credit and durable spiking learning

A spiking agent retains a learned choice after its episodic buffer is pulled, but only when a self-credit channel gated on agency does slow work. That dissociation is a pointed test for anyone who claims a cultured network has learned.

9 min read · Read → July 29, 2026

A culture substrate that amplifies glial calcium

Zirconia culture surfaces raise the amplitude and speed the kinetics of glial calcium signals without changing their frequency, the chemistry doing most of it and nanostructure adding more. For biohybrid computing that turns the substrate into part of the circuit.

8 min read · Read → July 28, 2026

Sensor mechanics as a pre-neural spike encoder

Meng, Jayaram and Mongeau link cockroach antenna biomechanics to a calibrated spike encoder and an SNN that reads contact location and speed above 95 percent within 170 milliseconds. We read what it means that the body encodes part of the code before any neuron fires, for embodied organoid intelligence.

9 min read · Read → July 28, 2026

The astrocyte front end and its supremacy claim

Tsybina and colleagues bolt a simulated spiking neuron and astrocyte network onto a CNN or vision transformer and report accurate classification from one example per class and under heavy noise. We weigh the loaded word against the mechanism and ask what slow glial modulation offers living neural tissue.

9 min read · Read → July 28, 2026

Dendritic plateaus as a hold and integrate memory

Burger and colleagues model NMDA dendritic plateau potentials as a leaky integrate and hold element that rescues reliable spiking when input arrival times jitter. We read what a computation held in slow dendritic states, invisible to a spike-only electrode, means for reading and trusting organoid activity.

9 min read · Read → July 27, 2026

Learning that lives in the dendrite, not the synapse

A single-layer spiking model puts in-context learning in the subthreshold dynamics of one dendritic compartment, with every synapse frozen at inference. For organoid intelligence it questions whether the interesting computation is even in the spikes a microelectrode array records.

9 min read · Read → July 27, 2026

When inflammation pre-writes a living computer

A transient IL-6 exposure in interneuron-enriched forebrain organoids leaves a lasting inflammatory and maturation imprint on the inhibitory population one month after withdrawal. For organoid intelligence, the developmental immune history of a substrate is a hidden variable that averaged characterisation never records.

9 min read · Read → July 27, 2026

When silicon claims the embodied geometry job

A robust perspective-n-point pose solver is recast as a distributed algorithm that runs on Intel Loihi 2 at roughly 1 percent of an embedded CPU's power. It marks out the embodied perception territory neuromorphic silicon is taking, and the part it still cannot deliver.

9 min read · Read →