Physics / mechanism
Biocomputing uses living neuronal cells as the computational substrate rather than silicon or brain-inspired electronics. Neurons (rodent or human iPSC-derived) are cultured (2D layers or 3D organoids) on microelectrode arrays (MEAs); the cells form their own synaptic connections and perform the computation, while electronics handle stimulation, read-out, and modulation. This is one step beyond Neuromorphic Computing, which implements spiking neural networks in CMOS/memristive hardware — here the network is biological, with plasticity at molecular, cellular, and network scales that proponents argue cannot be captured by a software model.
Competitive landscape — the compute-substrate layer
Evidence base
The bear case
Named pioneers are actively warning against the hype (2025 Statnews Biocomputing Backlash): Zador (“scientific dead-end”), Pasca (“confuse the public and policymakers”), Lancaster (“science fiction… the science just isn’t there yet”). Even the bull (Hartung) concedes useful biocompute is a next-decade prospect, and that training “might take several years.” No killer app; demos are Pong/DOOM. Ethics/consciousness regulation on human neural tissue is a real tail risk.
Companies using
Connected ideas
Sources
Frontier (open questions)
- Vascularisation / scale: can organoids exceed ~500 µm and ~100k cells with a viable interior?
- Reproducible, durable, transferable training (beyond classifying which electrode fired)?
- A killer app where wetware beats silicon on a metric a customer pays for.
- Does value sit in the wetware or the picks-and-shovels (HD-MEA, organoid supply)?
- Promotion gate: this is a theme, not a thesis. The why-now (AI energy, 2025 Iea Datacentre Electricity) is real but does not bridge to a near-term biocompute product. Full gate verdict: Biocomputing Substrate.