Biological computing: from DNA molecules to living neurons

Biological computing is the use of biological materials, such as DNA molecules, engineered living cells and networks of neurons, to store and process information. Instead of moving electrons through transistors, a biological computer represents data as molecular sequences, gene expression or electrical spikes, and computes through chemistry, genetic regulation or neural learning.

Key takeaways

• Biological computing (often shortened to biocomputing) works at three levels: molecular (DNA and other molecules), cellular (programmed bacteria and yeast) and neural (living neurons and brain organoids).
• The field started with Leonard Adleman’s 1994 DNA computing experiment in Science and now spans DNA data storage, synthetic gene circuits and neurons trained in closed loops.
• Interest is rising because silicon scaling has slowed: Dennard scaling ended in the mid-2000s, the memory wall persists, and data centres used about 415 TWh of electricity in 2024, according to the IEA.
• Each level has a different strength: molecules offer parallelism and density, cells offer computing inside living systems, and neurons offer adaptive, energy-efficient learning.
• No biological computer outperforms a conventional computer on general workloads today. The realistic near-term role is complementary, specialized and hybrid.
• Organoid intelligence is the neural branch of biological computing, and the one most focused on learning rather than fixed logic.

What is biological computing? #

At its core, any computer needs three things: a way to represent information, a way to transform it according to rules, and a way to read out the result. Silicon does this with voltages on transistors. Biocomputing does it with the physical and chemical machinery of life.

That machinery is surprisingly well suited to the job. Cells already process information constantly. A bacterium senses nutrients and decides which genes to switch on. DNA stores the instructions for building an organism in a four-letter code. A cortical neuron integrates thousands of inputs and decides whether to fire an action potential. Biological computing tries to repurpose these processes deliberately, so that we choose the inputs, the rules and the outputs.

The term covers several quite different technologies, which is a common source of confusion. A useful way to organize them is by the scale of the computing medium:

1. Molecular computing: information is encoded in molecules, usually DNA or RNA, and processed by hybridization, enzymes or strand displacement reactions in a test tube.
2. Cellular computing: living cells, usually E. coli or yeast, are engineered with synthetic gene circuits that implement logic, memory and timing.
3. Neural computing: networks of living neurons, from flat cultures to three-dimensional brain organoids, process signals and adapt through learning. This is often called wetware computing.

These are different answers to one question: which biological substrate fits which job? Together they make up what researchers sometimes call the biological computing paradigm, an approach that treats living matter as computing hardware rather than only as something to be simulated.

Is biocomputing the same as bioinformatics? #

No. Bioinformatics uses conventional computers to analyze biological data, such as genome sequences. Biological computing flips the direction: it uses biology to do the computing. The two meet in practice, because every biological computer relies on silicon for design, control and readout.

Why look beyond silicon computers? #

Conventional computing has improved at an extraordinary pace for six decades. Understanding why that pace has changed helps explain why researchers are exploring biological computers at all.

What happened to Moore’s law? #

In 1965, Gordon Moore observed in Electronics that the number of components on an integrated circuit had been doubling roughly every year, and projected that the trend would continue for at least a decade. In a 1975 paper at the IEEE International Electron Devices Meeting, he revised the expected pace to roughly a doubling every two years. That revised version became the industry’s planning rhythm.

Transistor counts alone never delivered speed, though. A second rule did much of that work.

Why did Dennard scaling break down? #

In 1974, Robert Dennard and colleagues at IBM described how MOSFET transistors could be scaled down. If dimensions and voltages shrank together, each new transistor would be smaller, faster and use less power, so power per unit of chip area stayed roughly constant. This is Dennard scaling, and it is why clock speeds climbed for years without chips melting.

By the mid-2000s, the voltage side of that bargain stopped working. Below a certain threshold, leakage currents rise sharply, so supply voltages could not keep falling. Power density started to climb, clock speeds plateaued, and the industry turned to multicore designs. Esmaeilzadeh and colleagues showed in 2011 that power limits would leave growing fractions of a chip unpowered at any moment, a problem they called “dark silicon.” Hennessy and Patterson, in their 2019 Turing lecture paper, describe the end of Dennard scaling and the slowing of Moore’s law as the defining constraints of modern computer architecture.

What is the von Neumann bottleneck and the memory wall? #

Most computers follow the von Neumann architecture: a processor and a separate memory, connected by a channel that carries both instructions and data. John Backus named the limiting effect of this channel the “von Neumann bottleneck” in his 1978 Turing Award lecture. In 1995, Wulf and McKee warned of a “memory wall”: processor speed was improving much faster than memory speed, so processors would spend more and more time waiting for data.

Modern AI workloads make this worse, because training and running large neural networks means moving enormous numbers of weights between memory and compute units. A large share of the energy goes into moving data rather than calculating with it. Brains do not have this split. In neural tissue, memory (synaptic strength) and computation (integration and firing) live in the same place.

How much energy does computing use? #

The International Energy Agency’s Energy and AI report (2025) estimates that data centres used around 415 TWh of electricity in 2024, about 1.5% of global electricity consumption, and projects that demand could roughly double to around 945 TWh by 2030. Compare this with the human brain, which runs on roughly 20 watts, as Balasubramanian summarizes in PNAS (2021), while coordinating around 86 billion neurons (Azevedo et al., 2009).

There is also a physical floor to consider. Landauer showed in 1961 that erasing one bit of information must dissipate at least a tiny, temperature-dependent amount of heat. Today’s chips operate far above that limit, which means there is, in principle, a great deal of room for more efficient computation. We explore the energy argument in depth in biocomputing vs silicon: why living neurons could fix AI’s energy problem.

Silicon is not finished. But the easy gains are gone, which makes alternative substrates worth serious study.

What are the types of biological computing? #

The three levels below share a goal but differ in almost everything else: speed, how they are programmed, and what they are good for.

Molecular computing: how does DNA computing work? #

Adleman’s experiment (1994). The field of DNA computing began with a single paper. Leonard Adleman (the “A” in the RSA cryptosystem) published “Molecular computation of solutions to combinatorial problems” in Science in 1994. He solved a small instance of the Hamiltonian path problem: find a route through a graph with seven vertices that visits each vertex exactly once.

He encoded each vertex and each edge as short synthetic DNA strands, designed so that edge strands could bind (hybridize) to the vertex strands they connected. Mixing them in a tube generated huge numbers of candidate paths at once. Standard molecular biology techniques, such as PCR and gel electrophoresis, then filtered out the wrong answers. The insight was parallelism: trillions of molecules explore possibilities simultaneously.

The limitation was just as clear. The lab work took days, and the amount of DNA needed grows exponentially with problem size for this kind of brute-force search. In 2002, Braich and colleagues pushed the approach to a 20-variable 3-SAT problem, a striking demonstration, but one that also showed DNA would not out-compete electronic computers on hard combinatorial problems.

Autonomous molecular computers (2001). Benenson and colleagues, in Ehud Shapiro’s group, reported in Nature (2001) a programmable, autonomous computing machine made of biomolecules: a two-state finite automaton built from DNA and enzymes, where the input, the transition rules (“software”) and the enzymes (“hardware”) were all molecules. Once mixed, it ran without human intervention. A 2004 follow-up in Nature used a similar molecular automaton to sense specific mRNA markers in a test tube and, in response, release a molecule designed to affect gene expression. This pointed to a new idea: molecular computers are interesting not because they are fast, but because they can operate where electronics cannot, such as inside biochemical environments.

DNA strand displacement circuits. A later wave of DNA computing dropped enzymes altogether. In strand displacement, a single DNA strand binds to a partly double-stranded complex and displaces one of its strands, which can then trigger the next reaction. Seelig and colleagues built enzyme-free nucleic acid logic circuits in Science (2006). Lulu Qian and Erik Winfree scaled the approach in Science (2011) using “seesaw” gates, building a circuit that computed the square root of a four-bit number from 130 DNA strands. The same year, Qian, Winfree and Bruck reported in Nature a small DNA-based neural network that performed associative memory, recalling stored patterns from partial input. In 2018, Cherry and Qian described a DNA winner-take-all network in Nature that classified handwritten digit patterns.

These circuits run in hours, not nanoseconds. Their value lies in programmability: their behavior follows from sequence design, which is predictable enough to automate. Related DNA nanotechnology, such as DNA origami (Rothemund, Nature, 2006), uses the same sequence-level control to build nanoscale shapes.

DNA data storage. Storage is the most practical molecular application today. In 2012, Church, Gao and Kosuri reported in Science that they had encoded a book draft of about 5.27 megabits, including images and a JavaScript program, into DNA and read it back. In 2013, Goldman and colleagues at the European Bioinformatics Institute stored 739 kilobytes, including all 154 of Shakespeare’s sonnets and an audio clip, using an encoding designed to reduce errors. Erlich and Zielinski’s “DNA Fountain” (Science, 2017) improved the efficiency further and estimated a theoretical density of about 215 petabytes per gram of DNA.

DNA’s advantages for archiving are density and durability: it can last a very long time if kept cool and dry, and the means of reading it (sequencing) will not become obsolete while biology depends on it. The obstacles, as Ceze, Nivala and Strauss review in Nature Reviews Genetics (2019), are the cost and speed of writing (synthesis) and reading (sequencing).

Cellular computing: how do synthetic gene circuits compute? #

Cellular computing moves from the test tube into living cells. Here the components are genes, promoters and the proteins that regulate them. A repressor protein that turns off a gene works much like a NOT gate. Combine enough of these and you can build logic, memory and clocks inside a bacterium.

The toggle switch and the repressilator (2000). Synthetic biology as a design discipline is often dated to two papers published together in Nature in January 2000. Gardner, Cantor and Collins built a genetic toggle switch in E. coli: two repressors that each shut off the other’s gene. The circuit settles into one of two stable states, and a chemical or heat pulse flips it, giving the cell a one-bit memory. Elowitz and Leibler built the repressilator: three repressors arranged in a ring, each inhibiting the next. The result was an oscillator, a biological clock that made cells blink with a fluorescent reporter over a period of hours.

Logic across many cells. Single cells are noisy, so researchers began distributing computation across populations. Tamsir, Tabor and Voigt (Nature, 2011) built NOR gates in separate E. coli strains, connected by diffusible signaling molecules acting as “wires,” and combined them to implement every two-input logic function. Regot and colleagues (Nature, 2011) built multicellular circuits in yeast, including a one-bit adder. Siuti, Yazbek and Lu (Nature Biotechnology, 2013) used recombinase enzymes that physically flip DNA segments, so that logic results were stored in the genome and inherited.

Genetic circuit design automation (2016). Writing gene circuits by hand is slow. Nielsen and colleagues introduced Cello in Science (2016), a tool that takes a circuit description written in Verilog, a hardware description language used for electronic chips, and compiles it into a DNA sequence. Of 60 circuits they built, 45 worked correctly in every output state, and about 92% of all output states across the circuits behaved as specified. Biology, it showed, can be engineered with an electronic-style design workflow.

Recording in living cells. Cells can also act as recorders. Shipman and colleagues (Nature, 2017) used the CRISPR-Cas adaptation system to write images and a short animated sequence into the genomes of a population of bacteria, then read them back by sequencing.

The natural home of cellular computing is not general-purpose processing. It is sensing and decision-making inside living systems: biosensors that report on a gut environment, engineered therapeutic cells that act only when several conditions hold, or production strains that adjust their own metabolism.

Neural computing: how do living neurons compute? #

Neurons are biology’s purpose-built information processors. Unlike DNA circuits and gene networks, which are mostly programmed by design, neural networks are trained. Their connections change with activity through synaptic plasticity, which means they can learn from feedback.

Neurons on electrodes. The key interface is the microelectrode array (MEA): a dish with embedded electrodes that record spikes from nearby neurons and deliver electrical stimulation. Modern high-density microelectrode arrays pack thousands of electrodes into a small area. Our guide to how microelectrode arrays work covers the hardware in detail.

The animat (2001). DeMarse and colleagues, in Steve Potter’s lab, reported in Autonomous Robots (2001) a “neurally controlled animat”: cultured rat cortical neurons on an MEA whose activity moved a simulated animal through a virtual environment, with sensory feedback from that world delivered back to the culture as stimulation. It was an early demonstration of closed-loop stimulation, where the neurons’ output changes what they receive next.

DishBrain (2022). Kagan and colleagues at Cortical Labs reported in Neuron (2022) that cultured human and mouse neurons on a high-density MEA could improve at a simulated game of Pong. Stimulation encoded the ball’s position, recorded activity moved the paddle, and the feedback differed depending on whether the paddle hit or missed. Performance improved within minutes of play. The authors framed the result in terms of the free energy principle. We unpack the training methods in how neurons learn in a dish.

Brain organoids. Flat cultures lack the three-dimensional structure of real tissue. Brain organoids are 3D clusters of neural cells grown from stem cells, often induced pluripotent stem cells (iPSCs), first described by Takahashi and Yamanaka (Cell, 2006). Lancaster and colleagues reported cerebral organoids in Nature (2013). For the full process, see how brain organoids are grown. In 2023, Smirnova and colleagues proposed organoid intelligence as a field in Frontiers in Science.

Brainoware (2023). Cai and colleagues at Indiana University placed a brain organoid on a high-density MEA and used it as the core of a reservoir computing system, reported in Nature Electronics (2023). Inputs were converted into stimulation patterns, and a conventional readout layer was trained on the organoid’s responses. The system was used for speech recognition (identifying speakers from recordings of Japanese vowel sounds) and for predicting a chaotic time series, with performance improving through training. Our post on reservoir computing with brain organoids explains why this approach suits living tissue so well.

Remote access (2024). Jordan and colleagues at FinalSpark described an “open and remotely accessible Neuroplatform” in Frontiers in Artificial Intelligence (2024), where researchers could run experiments on brain organoids over the internet. The paper reported organoid lifetimes of more than 100 days on the system. It marked a shift from one-off demonstrations toward shared infrastructure.

How do the three levels of biological computing compare? #

The table below summarizes the trade-offs. Speeds are orders of magnitude, not precise benchmarks.

A pattern stands out. Molecular and cellular approaches excel when the computation must happen in a chemical or biological environment. Neural approaches are the only ones that natively learn, which makes them the closest biological counterpart to machine learning.

How does biological computing compare with silicon, quantum and neuromorphic computing? #

Biological computing is one of several alternatives to classical digital electronics. It helps to see them side by side.

Neuromorphic computing, a term Carver Mead set out in Proceedings of the IEEE (1990), builds silicon that imitates neural principles, often running spiking neural networks. Biological neural computing uses the real thing. The two inform each other: neuromorphic hardware borrows ideas from biology, and biological systems rely on silicon for every input and output.

That dependency is why the most realistic future is hybrid. A biological component handles what it does best, while chips handle control, encoding, decoding and storage. We explore this architecture in bio-silicon hybrid computing.

Where does organoid intelligence fit in biological computing? #

Organoid intelligence sits at the neural end of the spectrum. Its distinctive claim is not speed or density but learning. A brain organoid is a three-dimensional, self-organizing network of human neurons and support cells that changes its connectivity in response to experience, much as developing brain tissue does.

That shifts what “programming” means. A DNA circuit is programmed by choosing sequences. A gene circuit is programmed by choosing genetic parts. A neural culture is programmed by choosing what it experiences: how inputs are encoded as stimulation, what feedback it receives, and how its activity is decoded. This is the domain of neural encoding and decoding, closed-loop control and careful experimental design. We cover the field in full in what is organoid intelligence?.

Because the experiment itself becomes the program, reproducibility matters enormously. Protocols need to be versioned, cultures tracked, and data shared in open formats such as Neurodata Without Borders. This practice of writing, running and reproducing experiments on living neural tissue as code is what we call organoid programming.

What are the honest limits of biological computing? #

Biological computing attracts bold claims, so it is worth being direct about the gaps.

• Speed. Every biological level is slower than silicon at the level of single operations. Molecular reactions take minutes to hours, gene circuits respond over hours, and neurons fire on millisecond timescales while transistors switch billions of times per second.
• Reliability. Molecular reactions produce side products and errors. Genetic circuits vary between cells and can break as cells mutate. Neural cultures vary between batches and change over days and weeks.
• Scale. The largest gene circuits contain at most tens of regulators. Organoids contain a tiny fraction of the neurons in a human brain and, without blood vessels, can develop dead cores (see organoid vascularization).
• Interfaces. Reading results requires sequencing, fluorescence microscopy or electrodes, each of which adds cost, latency and a bottleneck of its own. An MEA reaches only the neurons near its surface.
• Life support and logistics. Living systems need nutrients, sterility and temperature control. Microfluidics helps, but keeping biological hardware stable over months remains hard.
• Energy accounting. Neurons are efficient per operation, but incubators, pumps and electronics also use power. Fair comparisons must count the whole system.
• Ethics. Using human-derived neural tissue raises real questions about donor consent and, in the long run, moral status. These deserve attention now, as discussed in the ethics of organoid intelligence.

These limits do not rule biological computing out. They define where it can contribute first.

What is the outlook for biological computers? #

Each level is on its own trajectory.

Molecular computing is likely to reach practical use first through storage. DNA archives for rarely accessed data are a plausible near-term product as synthesis costs fall. DNA logic is more likely to find a home in diagnostics and smart therapeutics than in data centres.

Cellular computing is advancing through design automation, larger part libraries and better models of how circuits behave in hosts. Its impact will mostly show up in medicine, agriculture and biomanufacturing, as cells that sense and decide.

Neural computing is the least mature but the most directly connected to AI. The near-term value is scientific: understanding how real neurons learn, and testing how drugs affect human neural network function. The longer-term question is whether living neurons can deliver learning that is more sample-efficient and energy-efficient than artificial networks. That depends on progress in organoid maturation, vascularization, higher-bandwidth interfaces and, above all, reproducible experimental infrastructure. Our future of biocomputing roadmap goes deeper into these open problems.

Across all three, the direction is the same: biology as a specialized co-processor working alongside silicon, not a replacement for it. That reproducibility and access problem at the neural level is what we’re working on at Organoid Programming, with protocols as code and open, provenance-tracked data.

The bottom line on biological computing #

Biological computing is not one technology but a family: DNA computing and storage at the molecular level, synthetic gene circuits at the cellular level, and living neurons and brain organoids at the neural level. Each has a clear, verified history, from Adleman’s 1994 experiment to the toggle switch, Cello, DishBrain and Brainoware. Each also has clear limits in speed, reliability and scale. The case for biocomputing rests not on beating silicon at its own game, but on doing things silicon cannot: computing inside living systems, storing data at molecular density, and learning the way brains do.

Frequently asked questions

What is biological computing in simple terms?

Biological computing is the use of living or biologically derived materials, such as DNA molecules, engineered cells or cultured neurons, to store and process information. Instead of switching electrons through transistors, a biological computer uses chemical reactions, gene expression or electrical activity in cells to represent data and carry out a computation, with conventional electronics usually reading the result.

What are the main types of biological computing?

There are three main types, grouped by the scale of the medium. Molecular computing uses DNA or other molecules, including DNA computing, DNA strand displacement circuits and DNA data storage. Cellular computing programs living cells, usually bacteria or yeast, with synthetic gene circuits. Neural computing, which includes organoid intelligence, uses networks of living neurons on electrode arrays that can learn from feedback.

Who invented DNA computing?

Leonard Adleman is credited with the first DNA computation. In a 1994 paper in Science, he encoded a small instance of the Hamiltonian path problem (a graph with seven vertices) in synthetic DNA strands and used standard laboratory techniques to find a valid path. The experiment showed that molecules could perform a computation, though the lab steps took days.

Are biological computers faster than normal computers?

No. Individual biological operations are far slower than transistor switching: chemical reactions and gene expression take minutes to hours, and neurons fire on millisecond timescales while transistors switch in fractions of a nanosecond. Biological systems can be massively parallel and energy efficient, but no biological computer today beats a conventional computer on any practical general workload.

Is DNA data storage the same as DNA computing?

Not quite. DNA data storage uses DNA as an archive: digital files are encoded into sequences, synthesized, stored and later read back by sequencing. DNA computing uses DNA molecules to actually process information, for example by running logic gates or searching for solutions through chemical reactions. Both belong to molecular biological computing, but storage is much closer to commercial use.

How is organoid intelligence related to biological computing?

Organoid intelligence (OI) is one branch of biological computing, at the neural level. It uses three-dimensional brain organoids grown from human stem cells, placed on microelectrode arrays that stimulate and record them. Where DNA and gene circuits are mostly programmed by design, neural systems are trained: they adapt their connections in response to feedback, which makes them attractive for learning tasks.

Is biological computing better than quantum computing?

They solve different problems. Quantum computers use superposition and entanglement to speed up specific tasks such as simulating quantum systems. Biological computing aims at other strengths: molecular parallelism, computation inside living organisms, dense storage, and energy-efficient adaptive learning in neurons. Both are early and both are likely to complement classical silicon rather than replace it.

What are the biggest limitations of biological computing?

The main limits are speed, error rates, and scale. Molecular reactions are slow and imperfect, gene circuits behave differently from cell to cell and burden their hosts, and living neurons need life support and drift over time. Interfacing with digital systems is also hard, because results must be read out by sequencing, fluorescence or electrodes, which adds cost and latency.

Sources

1. Moore (1965). Cramming more components onto integrated circuits. Electronics 38(8); reprinted in Proceedings of the IEEE (1998).
2. Moore (1975). Progress in digital integrated electronics. IEEE International Electron Devices Meeting, Technical Digest, pp. 11-13; reprinted in IEEE Solid-State Circuits Society Newsletter (2006).
3. Dennard et al. (1974). Design of ion-implanted MOSFET's with very small physical dimensions. IEEE Journal of Solid-State Circuits.
4. Esmaeilzadeh et al. (2011). Dark silicon and the end of multicore scaling. Proceedings of the 38th International Symposium on Computer Architecture.
5. Hennessy & Patterson (2019). A new golden age for computer architecture. Communications of the ACM.
6. Backus (1978). Can programming be liberated from the von Neumann style? Communications of the ACM.
7. Wulf & McKee (1995). Hitting the memory wall: implications of the obvious. ACM SIGARCH Computer Architecture News.
8. Landauer (1961). Irreversibility and heat generation in the computing process. IBM Journal of Research and Development.
9. IEA (2025). Energy and AI.
10. Balasubramanian (2021). Brain power. PNAS.
11. Azevedo et al. (2009). Equal numbers of neuronal and nonneuronal cells make the human brain an isometrically scaled-up primate brain. Journal of Comparative Neurology.
12. Adleman (1994). Molecular computation of solutions to combinatorial problems. Science.
13. Braich et al. (2002). Solution of a 20-variable 3-SAT problem on a DNA computer. Science.
14. Benenson et al. (2001). Programmable and autonomous computing machine made of biomolecules. Nature.
15. Benenson et al. (2004). An autonomous molecular computer for logical control of gene expression. Nature.
16. Seelig et al. (2006). Enzyme-free nucleic acid logic circuits. Science.
17. Rothemund (2006). Folding DNA to create nanoscale shapes and patterns. Nature.
18. Qian & Winfree (2011). Scaling up digital circuit computation with DNA strand displacement cascades. Science.
19. Qian, Winfree & Bruck (2011). Neural network computation with DNA strand displacement cascades. Nature.
20. Cherry & Qian (2018). Scaling up molecular pattern recognition with DNA-based winner-take-all neural networks. Nature.
21. Church, Gao & Kosuri (2012). Next-generation digital information storage in DNA. Science.
22. Goldman et al. (2013). Towards practical, high-capacity, low-maintenance information storage in synthesized DNA. Nature.
23. Erlich & Zielinski (2017). DNA Fountain enables a robust and efficient storage architecture. Science.
24. Ceze, Nivala & Strauss (2019). Molecular digital data storage using DNA. Nature Reviews Genetics.
25. Gardner, Cantor & Collins (2000). Construction of a genetic toggle switch in Escherichia coli. Nature.
26. Elowitz & Leibler (2000). A synthetic oscillatory network of transcriptional regulators. Nature.
27. Tamsir, Tabor & Voigt (2011). Robust multicellular computing using genetically encoded NOR gates and chemical 'wires'. Nature.
28. Regot et al. (2011). Distributed biological computation with multicellular engineered networks. Nature.
29. Siuti, Yazbek & Lu (2013). Synthetic circuits integrating logic and memory in living cells. Nature Biotechnology.
30. Nielsen et al. (2016). Genetic circuit design automation. Science.
31. Shipman et al. (2017). CRISPR-Cas encoding of a digital movie into the genomes of a population of living bacteria. Nature.
32. DeMarse et al. (2001). The neurally controlled animat: biological brains acting with simulated bodies. Autonomous Robots.
33. Kagan et al. (2022). In vitro neurons learn and exhibit sentience when embodied in a simulated game-world. Neuron.
34. Lancaster et al. (2013). Cerebral organoids model human brain development and microcephaly. Nature.
35. Takahashi & Yamanaka (2006). Induction of pluripotent stem cells from mouse embryonic and adult fibroblast cultures by defined factors. Cell.
36. Smirnova et al. (2023). Organoid intelligence (OI): the new frontier in biocomputing and intelligence-in-a-dish. Frontiers in Science.
37. Cai et al. (2023). Brain organoid reservoir computing for artificial intelligence. Nature Electronics.
38. Jordan et al. (2024). Open and remotely accessible Neuroplatform for research in wetware computing. Frontiers in Artificial Intelligence.
39. Mead (1990). Neuromorphic electronic systems. Proceedings of the IEEE.
40. Merolla et al. (2014). A million spiking-neuron integrated circuit with a scalable communication network and interface. Science.
41. Davies et al. (2018). Loihi: a neuromorphic manycore processor with on-chip learning. IEEE Micro.
42. Arute et al. (2019). Quantum supremacy using a programmable superconducting processor. Nature.