Electrode arrays in numbers
From the papers in the sources below.
From 1972 to 26,400 electrodes
| Year | Array | Electrodes | What it showed |
|---|---|---|---|
| 1972 | Thomas and colleagues, Experimental Cell Research | A miniature array | You could follow the electrical activity of cultured cells from the dish they grew in |
| 1980 | Jerome Pine, Journal of Neuroscience Methods | 32, each about 8 by 10 micrometres | He recorded single neurons up to about 40 micrometres away and stimulated them through the same pads |
| 2022 | MaxOne chip used in DishBrain | 26,400, with up to 1,024 read at once | The cells got better at Pong while they played it |
| 2024 | FinalSpark Neuroplatform | 32 per array, 8 for each of 4 organoids | You can run experiments on living organoids from a Python notebook, sampled at 30 kHz |
| 2025 | Cortical Labs CL1 | 64 | It's the first one I know of that you can buy as a product |
I find the table useful because it shows how little the basic idea has changed since the 1970s. You grow the cells on a glass or silicon floor that has metal pads built into it, and each pad picks up the small voltage change when a neuron near it fires. I think what's really changed is how many pads there are and how fast you're able to read them.
How the electrodes hear a neuron
When a neuron fires, current flows across its membrane and the voltage in the fluid around it shifts, and an electrode that's close enough will pick that up. I'd look at Pine's 1980 paper if you want to know how close is close enough, because he got clean recordings from cells within about 40 micrometres of an electrode centre. That's why I think high-density chips matter so much. If your pads are spread out, most of the neurons will sit between them and you'll never hear them.
Pine also showed you could stimulate through the same pads, passing current to set off action potentials. I'd say that's the step that turns a recording tool into a computer, since you're now able to write as well as read. Every living-neuron system on this site, from DishBrain to the CL1, depends on that loop (the biocomputing guide covers the DNA side, which doesn't).
What the numbers mean in practice
I don't think more electrodes help much if you can't read them all. The DishBrain chip had 26,400 pads but its software read up to 1,024 at a time, at 20,000 samples a second, so the team had to pick which ones to listen to. FinalSpark went the other way, with only 32 electrodes per array but 30 kHz sampling and four organoids on each, and it swaps the organoids out as they die (I've covered that in the guide on how long brain organoids live).
If you're comparing platforms, I'd ask how many electrodes you can read at once and how fast you can read them before you look at the headline count. I'd also check whether each one can stimulate as well as record, since as far as I can tell that's what decides which experiments you'll actually be able to run.
Questions
What is a multielectrode array?
I'd describe it as a dish or chip with lots of small electrodes in its floor. You grow the neurons on top, and then you're able to record them firing and stimulate them through the same pads.
How many electrodes does a multielectrode array have?
It varies a lot, and I'd not read too much into the count. Pine's 1980 dish had 32 and so do FinalSpark's arrays, the CL1 has 64, and the chip used in DishBrain had 26,400 (though you could only read up to 1,024 at once).
How close does a neuron need to be to an electrode?
Pine got clean recordings from cells within about 40 micrometres of an electrode centre in his 1980 paper. I'd take that as a rough guide only, because I'd expect newer chips to differ.
Sources
- Thomas, Springer, Loeb, Berwald-Netter and Okun, A miniature microelectrode array to monitor the bioelectric activity of cultured cells, Experimental Cell Research, 1972
- Pine, Recording action potentials from cultured neurons with extracellular microcircuit electrodes, Journal of Neuroscience Methods, 1980
- Kagan et al., full text with methods, PubMed Central
- Jordan et al., Open and remotely accessible Neuroplatform for research in wetware computing, Frontiers in Artificial Intelligence, 2024
- Schottlender et al., Evaluating Encoding Strategies for Closed-Loop Classification in Biological Neural Networks, arXiv, 2026