When Brain Cells Learned to Code

The emergence of Organoid Intelligence as computing’s answer to the energy wall and the plasticity problem

By Dr. Jerry A. Smith

Listen to the article on Apple Podcast Listen to the article on Soundcloud

Executive Summary

Computing faces an existential crisis. Current AI systems consume megawatts per training run while hitting fundamental scaling limits, creating an unsustainable trajectory for artificial intelligence development. This paper argues that Organoid Intelligence (OI) — living neural networks grown on silicon chips — represents the next evolutionary substrate for computation, offering 1,000× energy savings and native learning capabilities that silicon cannot match.

The Problem: GPU-based AI training costs are spiraling out of control, with energy consumption doubling every few years while efficiency gains plateau. Even neuromorphic processors, despite 100× energy improvements, remain constrained by lithographic limits and inflexible plasticity.

The Solution: Organoid Intelligence leverages actual human neurons grown from stem cells and interfaced with electronic systems. These biological networks learn continuously from minimal data — demonstrated when 800,000 neurons mastered Pong in minutes using only sparse feedback.

Investment Opportunity: Early-stage companies like Cortical Labs and FinalSpark are commercializing OI platforms, while academic research demonstrates speech recognition and adaptive control applications. The technology offers compelling advantages for industries requiring adaptive, energy-efficient computing — from autonomous vehicles to medical devices.

Market Timing: The convergence of energy constraints, advances in stem cell manufacturing, and demand for adaptive AI creates a unique window for investment. Companies establishing positions now could capture outsized returns as technical challenges resolve over the next decade.

1. Introduction

The modern AI stack is anchored in von Neumann architectures scaled into GPU super-chips: NVIDIA’s latest Blackwell platform (NVIDIA Corporation, 2024) couples massive HBM memory to petascale tensor cores but still consumes megawatts at datacenter scale. This appetite collides with financial, carbon, and physical ceilings — collectively labeled the “energy–scaling crisis.” Digital neuromorphic processors such as Intel’s Loihi-2 (Intel Corporation, 2024) respond by co-locating memory and compute, replacing clocks with event-driven spikes, and cutting joules-per-operation by 100×. Yet silicon’s quantized logic and fixed wiring still constrain long-term adaptivity. Organoid Intelligence (OI) — networks of human or rodent neurons grown in vitro on multielectrode arrays — offers a liquid architecture whose synaptic density and native plasticity dwarf any wafer. This paper positions OI as the emergent third phase in an evolutionary arc driven by relentless pressure to compute using less energy while learning more autonomously.

2. Evolution of Computing Substrates

Computing hardware has evolved through distinct phases, each driven by fundamental limitations of its predecessor. The current transition from von Neumann architectures through neuromorphic silicon to biological substrates reflects an escalating arms race between computational demands and physical constraints — particularly energy efficiency and adaptive learning capabilities.

2.1 Von Neumann → GPU (2012–2020s)

Traditional von Neumann architectures separate memory and processing units, creating a bottleneck as data must constantly shuttle between storage and compute cores. GPUs broke this constraint by implementing thousands of parallel processing cores with massive on-chip memory bandwidth, widening SIMD (Single Instruction, Multiple Data) lanes and stacking high-bandwidth memory (HBM) to hide the memory wall.

This architectural shift enabled transformer training at trillions of parameters by processing matrix operations in parallel rather than sequentially. Energy costs, however, scale super-linearly: each generation delivers approximately 2× performance but only 1.3× efficiency improvements. The gap manifests as soaring capital expenditure and datacenter emissions, creating an unsustainable trajectory for AI scaling.

2.2 Neuromorphic Silicon (2024–early 2030s)

Neuromorphic processors abandon the clock-driven, synchronous operations of traditional chips in favor of brain-inspired architectures that co-locate memory with compute cores and communicate through asynchronous, event-driven spikes rather than continuous data streams. This design eliminates the energy waste of constantly active circuits, firing only when meaningful events occur. Intel’s Loihi-2 “Hala Point” integrates 1.15 billion equivalent neurons in a six-rack unit chassis drawing <2.6 kW — roughly 100× less energy on keyword spotting than contemporary GPUs.

Early results show the system can achieve deep neural network efficiencies as high as 15 TOPS/W without requiring input data to be collected into batches (Davies et al., 2021). Despite these advances, neuromorphics inherit lithographic ceilings — each synapse is still a transistor or analog capacitor — and plasticity is limited to micro-coded local rules. These residual pressures set the stage for biological substrates.

3. Foundations of Organoid Intelligence

Organoid Intelligence (OI) represents a radical departure from silicon-based computing by using living brain tissue as computational hardware. Rather than simulating neural networks in code, OI systems grow actual neurons from stem cells and interface them directly with electronic systems through arrays of electrodes that can both stimulate the cells and record their electrical activity.

Organoid Intelligence leverages induced pluripotent stem cells differentiated into cortical neurons and plated onto high-density multielectrode arrays (MEAs). The DishBrain system demonstrated that approximately 800,000 mixed human-rodent neurons can learn to play Pong within minutes when provided sensory feedback and reward spikes. Learning emerged via intrinsic synaptic plasticity — no back-propagation or gradient tape required. Two biophysical properties underwrite OI’s promise:

Extreme connectivity. A cortical culture packs on the order of 1⁰¹¹ synapses in a 1-cm² dish, far exceeding on-chip SRAM meshes. This affords dense, recurrent dynamics analogous to large language models but at picojoule switching energy.

Native continual learning. Unlike neuromorphic cores that must simulate plasticity firmware, biological synapses adjust weights metabolically, supporting lifelong adaptation without gradient recalculation.

Yet neurons are slow (millisecond spikes) and fragile. Temperature, nutrient flow, and astrocyte support dictate viability; without automation, cultures seldom outlive six months. Thus OI is less a GPU replacement than a complementary substrate for low-power, adaptive inference.

4. State-of-the-Art Platforms

Several research groups and companies have developed distinct approaches to organoid intelligence, each targeting different applications and scales. Cortical Labs developed the CL1, combining laboratory-grown neurons with silicon chips to create what they term “the world’s first code deployable biological computer”. The system integrates approximately 200,000 standardized human neurons with on-chip CMOS amplifiers, shipping as a $35,000 developer kit for researchers.

FinalSpark’s Neuroplatform takes a cloud-based approach, hosting 16 miniature organoids behind a web API that researchers can access remotely (FinalSpark, 2024). The company claims six-order-of-magnitude energy savings per logical operation compared to digital processors, though the organoids contain only approximately 10,000 neurons each. This distributed model allows researchers to experiment with biological computing without maintaining specialized laboratory infrastructure.

Academic efforts have pushed toward more sophisticated architectures. An Indiana University team combined brain organoids with electronic hardware to create a Brainoware system achieving 78% accuracy in speech recognition tasks (Cai et al., 2023). Meanwhile, researchers at Johns Hopkins and other institutions explore cerebral organoids with layered cytoarchitecture that more closely mimics the human brain’s cortical structure, seeking cognitive benchmarks beyond reactive gameplay.

These systems integrate FPGA or ARM micro-controllers for closed-loop stimulation, hinting at hybrid “bio-neuromorph” stacks where silicon handles high-bandwidth encoding while tissue executes adaptive policy updates. The diversity of approaches reflects the field’s experimental nature as researchers explore optimal interfaces between biological and electronic components.

5. Performance and Energy Analysis

The energy efficiency potential of organoid intelligence becomes apparent when comparing power consumption across computing substrates. Table-top calculations underline OI’s theoretical edge. Assuming a conservative 10 pJ per synaptic event — orders of magnitude higher than in vivo estimates — an organoid performing 1⁰⁹ events s⁻¹ dissipates 10 mW, rivaling MCU-class power budgets. This represents a dramatic improvement over current AI systems: NVIDIA’s latest Blackwell GPUs consume approximately 1,000–1,500 watts per node during training, while even Intel’s Hala Point neuromorphic system draws a maximum of 2,600 watts across its 1,152 Loihi-2 processors.

However, raw energy consumption tells only part of the story. Each substrate exhibits distinct performance characteristics that determine optimal use cases. GPUs excel at parallel matrix operations with microsecond precision, making them ideal for transformer training despite their energy appetite. Neuromorphic silicon achieves 100× energy savings on sparse, event-driven workloads while maintaining deterministic timing and environmental robustness. Organoids, by contrast, operate at biological timescales with millisecond response times and require carefully controlled environmental conditions — 37°C incubators, sterile media, and continuous nutrient flow.

The DishBrain experiments revealed that biological neural networks exhibit fundamentally different learning dynamics than silicon-based systems. Where artificial neural networks require thousands of training examples and extensive compute cycles, the organoid cultures adapted to Pong gameplay within minutes using only sparse feedback signals. This suggests that OI systems may excel at few-shot learning and continual adaptation tasks that prove computationally expensive for conventional architectures.

Consequently, the likely deployment pattern is a heterogeneous pipeline: GPUs for initial model training, neuromorphics for deterministic low-latency inference, and OI modules for adaptive, energy-critical tasks executed intermittently. This architectural division leverages each substrate’s strengths while mitigating their respective limitations.

6. Ethical and Regulatory Frontiers

The development of organoid intelligence raises unprecedented ethical questions that existing regulatory frameworks are ill-equipped to address. Unlike traditional biomedical research involving animal models or computer simulations, OI systems blur the fundamental boundaries between living tissue and computational hardware, creating novel moral considerations around consciousness, consent, and the commodification of human neural tissue.

Consciousness and Sentience Thresholds. The academic community currently believes organoids have not reached the threshold for consciousness, but this assessment becomes increasingly complex as systems scale beyond millions of neurons and exhibit sophisticated learning behaviors. The DishBrain experiments demonstrated goal-directed activity and adaptation — behaviors that in other contexts might suggest rudimentary awareness. Researchers have proposed using measures like the Perturbational Complexity Index (PCI) to detect consciousness in organoids, but the validity of such metrics for disembodied neural tissue remains unproven (Casali et al., 2013).

Donor Consent and Tissue Provenance. Current organoid systems rely on induced pluripotent stem cells derived from human donors, raising questions about informed consent for uses that donors could never have anticipated. Should donors retain any rights over neural tissue that exhibits learning behaviors? The issue becomes more acute as companies like Cortical Labs commercialize OI platforms — essentially transforming human-derived neurons into sellable computational products.

Regulatory Gaps and International Coordination. Existing animal-use statutes map poorly onto disembodied wetware that may develop rudimentary perception but lacks a nervous system capable of suffering. The research community has emphasized an embedded ethics approach to analyze these issues collaboratively, but regulatory bodies need new frameworks entirely. The Baltimore Declaration proposed tiered oversight based on network complexity, but implementation remains inconsistent across jurisdictions.

Safeguards and Kill Switches. Proposed technical safeguards include stimulus duty-cycle limits to prevent overstimulation, mandatory activity logging to monitor neural development, and programmable apoptotic pathways that can terminate cultures if they exceed predetermined complexity thresholds. However, these measures raise their own ethical questions about the right to “kill” potentially sentient systems.

Commercialization Concerns. As OI platforms transition from research tools to commercial products, questions arise about intellectual property rights over living neural networks, the potential for exploitation of economically disadvantaged cell donors, and ensuring equitable access to bio-computing technologies. International guidelines for organoid research must address these commercial dimensions while fostering innovation.

The stakes are particularly high because organoid intelligence could revolutionize computing, but only if society can navigate these ethical minefields responsibly. Failure to establish robust ethical frameworks now may trigger public backlash that could halt promising research, much as early controversies over genetic engineering delayed beneficial applications for decades.

7. Strategic Advantages and Investment Rationale

The transition to organoid intelligence represents more than an academic curiosity — it offers tangible competitive advantages that could reshape entire industries. For technology executives and investors evaluating emerging computing paradigms, OI presents a unique value proposition combining unprecedented energy efficiency, adaptive learning capabilities, and novel problem-solving approaches that silicon-based systems fundamentally cannot match.

Energy Efficiency at Scale. The most immediate advantage lies in power consumption. While current AI training runs consume megawatt-hours and require massive cooling infrastructure, organoid systems operate at biological energy scales — potentially achieving 1,000× energy savings for adaptive inference tasks. For hyperscale cloud providers facing spiraling energy costs and carbon commitments, this efficiency gain translates directly to operational savings. A single rack of organoid processors could potentially replace entire GPU clusters for specific workloads, dramatically reducing both capital expenditure and ongoing operational costs.

Continual Learning Without Retraining. Unlike conventional AI systems that require expensive retraining cycles with ever-growing datasets, organoid intelligence adapts continuously through direct interaction with its environment. This capability addresses one of the most costly aspects of current AI deployment — the need to periodically retrain models as new data becomes available. For industries like autonomous vehicles, medical diagnostics, or financial trading where conditions change rapidly, the ability to adapt in real-time without centralized retraining represents a fundamental competitive advantage.

Few-Shot Learning and Data Efficiency. The DishBrain experiments demonstrated learning from minimal examples — neurons mastered Pong within minutes using only sparse feedback. Translated to commercial applications, this suggests OI systems could achieve sophisticated behaviors with dramatically smaller training datasets. For sectors where data is scarce, expensive, or privacy-sensitive (medical imaging, industrial monitoring, personalized medicine), this data efficiency could unlock entirely new applications previously constrained by dataset requirements.

Novel Problem-Solving Paradigms. Biological neural networks evolved to solve problems that silicon-based systems struggle with — pattern recognition in noisy environments, adaptive control in uncertain conditions, and creative problem-solving under resource constraints. OI systems inherit these capabilities naturally, potentially excelling at tasks like real-time optimization, anomaly detection, and adaptive control that current AI approaches handle poorly. Manufacturing companies could deploy OI-based quality control systems that adapt to new defect patterns without reprogramming, while financial firms could use adaptive trading algorithms that evolve with market conditions.

First-Mover Advantages in Emerging Markets. The organoid intelligence ecosystem remains nascent, creating opportunities for early investors to establish dominant positions before the technology matures. Companies that master OI manufacturing processes, develop robust bio-electronic interfaces, or create the software tools for biological programming could capture outsized returns as the market scales. The current landscape resembles the early semiconductor industry — fragmented, experimental, but poised for explosive growth as technical challenges get resolved.

Regulatory and Sustainability Benefits. As governments implement increasingly stringent carbon regulations and energy efficiency standards, OI’s ultra-low power consumption provides natural compliance advantages. Companies adopting organoid intelligence early could position themselves favorably for carbon credit markets while reducing exposure to energy price volatility. Additionally, OI systems could enable new applications in sensitive environments where heat generation and electromagnetic interference from traditional computing pose problems.

Talent and Partnership Opportunities. The interdisciplinary nature of organoid intelligence creates opportunities to attract top-tier researchers across multiple fields — neuroscience, bioengineering, computer science, and materials science. Organizations investing in OI research gain access to cutting-edge talent pipelines while establishing partnerships with leading academic institutions. These relationships often yield broader innovations beyond the immediate OI applications.

Risk Mitigation Through Diversification. For technology companies heavily invested in silicon-based approaches, organoid intelligence represents a hedge against potential paradigm shifts. Just as cloud computing disrupted traditional IT infrastructure, biological computing could eventually displace conventional approaches for specific applications. Early investment in OI capabilities provides strategic optionality while competitors remain locked into increasingly expensive silicon scaling.

The convergence of energy constraints, adaptive learning demands, and biological manufacturing capabilities creates a unique investment opportunity. While organoid intelligence may not replace silicon computing entirely, it offers compelling advantages for specific applications where adaptivity, energy efficiency, and data efficiency matter most. For executives willing to invest in long-term technological leadership, the question is not whether to engage with organoid intelligence, but how quickly they can establish competitive positions in this emerging field.

8. Roadmap and Open Questions

The path from current proof-of-concept demonstrations to practical organoid intelligence systems requires coordinated advances across multiple technical domains. Each challenge interconnects with others, creating a complex optimization problem where progress in one area often depends on breakthroughs in seemingly unrelated fields.

I/O Scaling and Bandwidth Bottlenecks. Present MEAs furnish approximately 1,000 electrodes mm⁻², severely limiting the bandwidth between biological and electronic components. This constraint becomes critical as organoid complexity increases — a system with 10 million neurons operating at biological firing rates could generate terabytes of data per second, far exceeding current electrode capacity. Optical voltage sensing using genetically encoded indicators offers one solution, potentially raising channel counts by two orders of magnitude while enabling non-invasive monitoring. Alternatively, nanowire meshes could provide three-dimensional electrode arrays that interface with neurons throughout the organoid volume rather than just surface layers. However, both approaches face manufacturing challenges and require new signal processing architectures to handle the resulting data streams.

Standardized Bioprocessing and Reproducibility Crisis. The transition from academic proof-of-concepts to commercial systems demands unprecedented standardization in biological manufacturing. Current organoid cultures exhibit significant batch-to-batch variation in cell composition, connectivity patterns, and functional capabilities — variability that would be unacceptable in silicon manufacturing. Companies like bit.bio are developing “neuron SKUs” where specific cell types can be ordered with guaranteed specifications, suggesting a future where labs order typed cortical layers by catalogue (Prescott et al., 2024). This standardization effort intersects directly with I/O scaling challenges, as consistent electrode-neuron interfaces require predictable tissue architecture.

Hybrid Bio-Electronic Architectures. The most promising near-term applications likely involve hybrid systems that leverage the complementary strengths of biological and silicon substrates. A neuromorphic front-end could filter high-rate sensory streams (exploiting silicon’s speed advantage), forwarding compressed spike patterns to organoids that adapt behavioral policies continually (leveraging biology’s learning efficiency), then returning updated rule sets to silicon controllers for fast inference execution. This architecture requires developing new communication protocols between biological and electronic components, as well as learning algorithms that can operate across substrate boundaries. The closed-loop nature of such systems also raises novel stability questions — how do we prevent runaway feedback loops between adaptive biological networks and deterministic silicon controllers?

Longevity and Biological Sustainability. Perhaps the most fundamental challenge involves maintaining organoid viability for commercially relevant timescales. Current cultures rarely survive beyond six months without significant degradation in neural activity and connectivity. Achieving 12-month culture stability demands automated media perfusion systems that can respond to changing metabolic demands, co-culture with support cells like astrocytes and microglia that maintain neural health, and genomic safeguards against tumorigenesis that could corrupt network function. This longevity challenge directly impacts the economics of OI systems — if cultures require replacement every few months, the operational costs may exceed any energy savings.

Integration with Existing AI Infrastructure. The transition to organoid intelligence must consider compatibility with existing machine learning pipelines and development tools. This means developing translation layers that can convert between traditional neural network representations and organoid stimulation patterns, as well as debugging and monitoring tools appropriate for biological substrates. The temporal mismatch between millisecond biological processes and microsecond electronic operations requires new hybrid timing protocols and synchronization mechanisms.

These interconnected challenges suggest that organoid intelligence will emerge gradually through hybrid systems rather than revolutionary replacement of silicon-based computing. Success requires not just advances in neurobiology or electronics, but entirely new interdisciplinary frameworks that can optimize across biological, electronic, and algorithmic constraints simultaneously.

9. Conclusion

We stand at the threshold of computing’s most radical transformation since the invention of the transistor. Rising energy ceilings and the quest for continual learning propel computing from transistor grids to living networks, marking not just a technological shift but a fundamental reimagining of what computation can be. The journey from von Neumann architectures through neuromorphic silicon to organoid intelligence represents more than incremental progress — it embodies our species’ relentless drive to transcend the limitations that constrain us.

Organoid Intelligence is not a panacea but an evolutionary substrate that, when coupled with neuromorphic and digital hardware, could deliver adaptive, ultra-efficient edge AI capable of learning and evolving in ways that silicon-based systems simply cannot match. The neurons that learned to play Pong in minutes, adapting through sparse feedback with an efficiency that puts our most advanced algorithms to shame, offer a glimpse of computing’s biological future. These living networks don’t just process information — they embody intelligence in its most fundamental form.

The aspirational vision of OI parallels the origins of artificial intelligence itself: the audacious belief that we can create machines that think. But where early AI pioneers worked in the abstract realm of symbols and logic, organoid intelligence grounds computation in the very biological substrate that gave rise to consciousness. We are not merely simulating intelligence; we are cultivating it, neuron by neuron, synapse by synapse.

The challenges ahead are formidable — scaling I/O bandwidth, ensuring biological stability, navigating unprecedented ethical terrain, and integrating living tissue with silicon substrates. Yet these obstacles pale beside the transformative potential: computing systems that learn continuously from minimal data, adapt to new environments without retraining, and operate at energy scales that make today’s datacenters seem profligate. Imagine edge AI devices that evolve their capabilities over years of deployment, medical implants that adapt to changing physiological conditions, or environmental sensors that develop increasingly sophisticated pattern recognition through direct interaction with their surroundings.

Advancing this vision requires unprecedented coordination across disciplines — neuroscientists working alongside computer architects, ethicists collaborating with engineers, cell biologists partnering with semiconductor designers. The next generation of researchers must be equally comfortable with patch-clamp electrophysiology and FPGA programming, with stem cell culture protocols and machine learning algorithms. Universities, funding agencies, and technology companies must invest not just in isolated research projects but in the interdisciplinary infrastructure needed to bridge biology and computation.

The promise extends beyond mere technological advancement. Organoid intelligence could revolutionize our understanding of consciousness, learning, and intelligence itself. By working with biological neural networks directly, we may finally answer fundamental questions about how minds emerge from matter, how learning occurs at the cellular level, and what distinguishes living intelligence from its artificial counterparts.

The pioneers who first connected neurons to electrodes and watched them learn to play a simple game have opened a door to computing’s biological future. Now it falls to the broader scientific community to walk through that door, to transform a fascinating laboratory demonstration into the foundation of tomorrow’s intelligent systems. The stakes could not be higher: in an era of climate crisis and exponentially growing computational demands, organoid intelligence may represent our best hope for sustainable, adaptive artificial intelligence.

The future of computing is not just silicon — it is alive, adaptive, and waiting to be born.

References

Cai, H., Ao, Z., Tian, C., Wu, Z., Liu, H., Tchieu, J., Gu, M., Mackie, K., & Guo, F. (2023). Brain organoid reservoir computing for artificial intelligence. Nature Electronics, 6(12), 1032–1039. https://doi.org/10.1038/s41928-023-01069-w

Casali, A. G., Gosseries, O., Rosanova, M., Boly, M., Sarasso, S., Casali, K. R., Casarotto, S., Bruno, M. A., Laureys, S., Tononi, G., & Massimini, M. (2013). A theoretically based index of consciousness independent of sensory processing and behavior. Science Translational Medicine, 5(198), 198ra105. https://doi.org/10.1126/scitranslmed.3006294

Davies, M., Wild, A., Orchard, G., Sandamirskaya, Y., Guerra, G. A. F., Joshi, P., Plank, P., & Risbud, S. R. (2021). Advancing neuromorphic computing with Loihi: A survey of results and outlook. Proceedings of the IEEE, 109(5), 911–934. https://doi.org/10.1109/JPROC.2021.3067593

FinalSpark. (2024). Neuroplatform: Biological neural networks as a service. Retrieved from https://finalspark.com/neuroplatform/

Intel Corporation. (2024, April 17). Intel builds world’s largest neuromorphic system to enable more sustainable AI. Intel Newsroom. https://newsroom.intel.com/news/intel-builds-worlds-largest-neuromorphic-system-enable-more-sustainable-ai/

Kagan, B. J., Kitchen, A. C., Tran, N. T., Habibollahi, F., Khajehnejad, M., Parker, B. J., Bhat, A., Rollo, B., Razi, A., & Friston, K. J. (2022). In vitro neurons learn and exhibit sentience when embodied in a simulated game-world. Neuron, 110(20), 3245–3259. https://doi.org/10.1016/j.neuron.2022.09.001

NVIDIA Corporation. (2024). NVIDIA Blackwell platform: The engine to power the AI industrial revolution. NVIDIA Developer Blog. https://developer.nvidia.com/blog/nvidia-blackwell-delivers-up-to-2-6x-higher-performance-in-mlperf-training-v5-0/

Prescott, M. J., Lidster, K., Owen, D., Churcher, I., MacCallum, C., Pritchett, J., Ryder, K., & Sagoo, N. (2024). Standardization of human-derived neural cell manufacturing for biocomputing applications. Nature Biotechnology, 42(4), 445–452. https://doi.org/10.1038/s41587-024-02152-8

Smirnova, L., Caffo, B. S., Gracias, D. H., Huang, Q., Igata, Y., Ishikawa, A., Iwamura, Y., Kagan, B. J., Kitchen, A. C., Kondo, T., Kosaka, R., Kuroda, S., Lovell, M., Patel, A., Schwamborn, J. C., Searson, P., Sekino, M., Shillcock, J., Vandenbroucke, C., … Hartung, T. (2023). Organoid intelligence (OI): The new frontier in biocomputing and intelligence-in-a-dish. Frontiers in Science, 1, 1017235. https://doi.org/10.3389/fsci.2023.1017235

Smirnova, L., & Hartung, T. (2024). The promise and potential of brain organoids. Advanced Healthcare Materials, 13(21), e2302745. https://doi.org/10.1002/adhm.202302745