High-density microelectrode array (HD-MEA): definition and how it works

A high-density microelectrode array (HD-MEA) is a microelectrode array built on a CMOS silicon chip, with thousands to tens of thousands of electrodes spaced micrometers apart and amplifiers integrated beneath them. HD-MEAs record and stimulate neurons and brain organoids at subcellular resolution, so a single neuron’s spike can be seen across several neighboring electrodes.

Key takeaways

• HD-MEAs extend the classic microelectrode array by integrating electrodes with on-chip electronics.
• Berdondini et al. (2009) described an active pixel sensor array with 4,096 electrodes; Ballini et al. (2014) described a chip with 26,400 electrodes and 1,024 readout channels.
• Dense electrodes let each neuron be detected on several sites, which improves spike sorting and allows tracking signals along axons.
• HD-MEAs are widely used to record brain organoids and to run closed-loop stimulation experiments.
• Trade-offs include large data volumes, readout channel limits, opaque substrates and cost.

What is a high-density microelectrode array? #

A conventional passive MEA has tens of metal electrodes on glass, each wired to an external amplifier. That wiring limits how many electrodes fit on the surface. HD-MEAs solve the problem by building the electrodes directly on top of a CMOS chip, the same kind of technology used in camera sensors. Amplifiers, filters, and switching or multiplexing circuits sit right under the electrodes, so the chip can host thousands of sites in an area of a few square millimeters.

The result is a recording surface dense enough that the spacing between electrodes approaches or falls below the size of a neuron’s cell body. Müller et al. reported in Lab on a Chip (2015) that such a platform could study neurons “at subcellular, cellular, and network levels”, from axonal signals up to whole-culture dynamics.

How does an HD-MEA work? #

Integrated electronics #

Each electrode connects to circuitry on the chip that amplifies the tiny extracellular signal before it is affected by noise from long wires. The signals are digitized and streamed to a computer.

Two common architectures #

1. Active pixel sensor (APS) arrays. Every electrode is a “pixel” with its own front-end, and all are read out by fast multiplexing, much like a camera sensor. Berdondini et al. described a 4,096-electrode APS array in Lab on a Chip (2009).
2. Switch-matrix arrays. The chip has more electrodes than readout channels, and a reconfigurable switch matrix routes a selected subset of electrodes to the amplifiers. Ballini et al. described a design in IEEE Journal of Solid-State Circuits (2014) in which 1,024 channels can be connected to a chosen subset of 26,400 electrodes, with stimulation also supported.

A typical recording workflow #

1. Place a culture, tissue slice or organoid on the chip and let it attach.
2. Run a fast “activity scan” across the whole array to find where neurons are active.
3. Select electrode configurations that cover active regions densely.
4. Record, detect spikes, and run spike sorting that uses each neuron’s spatial footprint.
5. Optionally stimulate chosen electrodes and record responses, in open or closed loop.

HD-MEA vs standard MEA #

Why do HD-MEAs matter for biocomputing? #

Organoid intelligence needs a high-bandwidth channel between software and living tissue. HD-MEAs provide more points of contact for both input and output:

• Richer readout. Recording many identified neurons gives a more detailed picture of the network’s state, which helps neural encoding and decoding.
• Targeted input. Fine-grained stimulation lets experimenters deliver spatially structured patterns.
• Better tracking over time. Spatial footprints help follow the same neurons across sessions as networks change through synaptic plasticity.
• Organoid characterization. Sharf et al. reported in Nature Communications (2022) that HD-MEA recordings revealed functional circuitry and oscillatory dynamics in human brain organoids.

Closed-loop learning experiments, such as those described in how neurons learn in a dish, and reservoir approaches, such as those in reservoir computing with brain organoids, both benefit from denser interfaces.

How much data does an HD-MEA produce? #

A rough calculation shows why data handling matters. Assume, for illustration, 1,024 channels sampled at 20,000 samples per second with 2 bytes per sample. That is about 41 megabytes per second, or roughly 147 gigabytes per hour of raw recording. Real figures depend on the system’s channel count, sampling rate and compression, but the order of magnitude explains why many pipelines detect spikes on the fly and store only snippets for some experiments.

HD-MEA vs high-density implanted probes #

High-density recording is not limited to culture dishes. Jun et al. reported in Nature (2017) the Neuropixels probe, a CMOS silicon shank with 960 recording sites, of which 384 can be recorded at once, designed for implantation in animal brains. Neuropixels and in vitro HD-MEAs share the idea of integrating electronics with dense electrodes, but probes are long and thin to reach deep brain regions, while HD-MEAs are flat surfaces for cultures, slices and organoids.

What are the limitations of HD-MEAs? #

• Readout bottleneck. Switch-matrix chips cannot record every electrode simultaneously, so experimenters must choose configurations.
• Data volume. Thousands of channels sampled at high rates generate large datasets that require efficient pipelines and open formats like Neurodata Without Borders.
• Opaque substrate. Silicon chips block transmitted light, complicating some microscopy.
• Flat geometry. Like all planar MEAs, they contact only the bottom surface of a 3D organoid.
• Stimulation artifacts and heat. Dense electronics and stimulation must be managed to avoid masking signals or disturbing cells.
• Cost and expertise. HD-MEA systems and their analysis software require more investment than passive arrays.

How do HD-MEAs fit into the wider toolkit? #

HD-MEAs are part of the broader practice of electrophysiology. Spira and Hai’s review in Nature Nanotechnology (2013) places them among many MEA technologies with different trade-offs. For an accessible overview of the hardware, read microelectrode arrays explained.

Turning HD-MEA data streams into reusable protocols, closed-loop experiments and shareable datasets is the software problem we are working on at Organoid Programming.

Frequently asked questions

What makes a microelectrode array high-density?

There is no strict cutoff, but HD-MEAs usually have thousands of electrodes packed closely enough, with center-to-center spacing in the range of tens of micrometers or less, that a single neuron is detected on several electrodes. This density requires integrating amplifiers and multiplexing circuits on the chip itself, which is why HD-MEAs are typically built with CMOS technology.

Can an HD-MEA record from all electrodes at once?

It depends on the architecture. Some designs, such as active pixel sensor arrays, read every electrode by rapidly multiplexing. Others, such as the switch-matrix design described by Ballini et al. (2014), have more electrodes than readout channels; in that chip 1,024 channels can be connected to a chosen subset of 26,400 electrodes, and the selection can be changed by software.

Why do HD-MEAs help spike sorting?

Because each neuron’s spike appears on several neighboring electrodes with a characteristic spatial pattern, or footprint. Spike sorting algorithms can use that footprint, not just the waveform on one electrode, to tell nearby neurons apart and to track the same neuron over time. Sparse arrays offer only one view of each neuron, which makes separation harder.

Are HD-MEAs used for brain organoids?

Yes. Researchers place brain organoids, or slices of them, on HD-MEAs to record activity from many individual neurons and to map network dynamics. Sharf et al. reported in Nature Communications (2022) that HD-MEA recordings of human brain organoids revealed functional neuronal circuitry and oscillatory dynamics.

Sources

1. Berdondini et al. (2009). Active pixel sensor array for high spatio-temporal resolution electrophysiological recordings from single cell to large scale neuronal networks. Lab on a Chip.
2. Ballini et al. (2014). A 1024-channel CMOS microelectrode array with 26,400 electrodes for recording and stimulation of electrogenic cells in vitro. IEEE Journal of Solid-State Circuits.
3. Müller et al. (2015). High-resolution CMOS MEA platform to study neurons at subcellular, cellular, and network levels. Lab on a Chip.
4. Sharf et al. (2022). Functional neuronal circuitry and oscillatory dynamics in human brain organoids. Nature Communications.
5. Jun et al. (2017). Fully integrated silicon probes for high-density recording of neural activity. Nature.
6. Spira & Hai (2013). Multi-electrode array technologies for neuroscience and cardiology. Nature Nanotechnology.