Electrode and instrumentation reporting · Specifications over marketing · Cited to primary sources
OrganoidArray
Hardware · Spec sheet

Organoid-on-a-chip and MEA hardware

The hardware that lets a computer talk to living neurons is a stack of hard problems: an electrode-electrolyte interface that must not poison the cells, low-noise electronics that can hear microvolts, microfluidics that keep tissue alive, and a feedback loop fast enough to teach it.

This is the layer the rest of biocomputing takes for granted. Get the electrode chemistry or the latency wrong and nothing above it works. What follows is the spec sheet, with the physics that sets each number.

Extreme macro of a high-density CMOS microelectrode array chip with a grid of tiny gold electrode pads, lit with warm amber edge light against a black background.
A high-density microelectrode array: thousands of pads, each one a place where electronics meets ions. Imaging is illustrative.

The bio-electronic boundary

Where a metal electrode meets the salty fluid around a cell, it forms an electrical double layer that behaves like a capacitor and resists passing current. To hear a 10 to 100 microvolt action potential through it, the interface impedance has to be low, which is why electrodes are coated in platinum black or titanium nitride: their fractal micro-texture multiplies the effective surface area and drops impedance at the key 1 kHz neural frequency from megaohms to under 50 kilohms.10 Stimulation runs as charge-balanced biphasic pulses so no net current accumulates, because a DC offset would electrolyze water, shift pH, and kill the tissue.

The electrode-to-tissue stack A vertical stack from the silicon CMOS substrate up through the coated electrode, the electrical double layer, the culture medium, and the neural tissue above. tissue -> silicon Neural tissue organoid, microvolt signals Culture medium ionic electrolyte Electrical double layer capacitive interface Coated electrode (Pt-black/TiN) fractal, low impedance CMOS substrate amplifiers, multiplexing
Every signal crosses this stack twice. The double layer is where electronic charge becomes ionic charge; the coating is what makes that crossing efficient enough to hear single cells.

Recording and stimulation electronics

A representative high-density platform routes on the order of 1,024 recording channels and 64 stimulation channels through a low-noise analog front end. Each channel has a low-noise amplifier with a programmable bandpass, roughly 100 Hz to 3 kHz for spikes, 1 to 100 Hz for local field potentials. Signals are digitized at the chip edge by 16-bit converters at about 25 kHz per channel, which adds up to roughly 400 megabits per second per chip, streamed to an FPGA for spike detection and feedback.

~1,024recording channels
25 kHzper-channel ADC, 16-bit
<50 kOhmelectrode impedance @ 1 kHz
1.2 msclosed-loop round trip

Reference specifications

Multi-electrode array layout (representative)
ParameterValue
Active electrodes1,024 recording channels
Stimulation ports64 dedicated channels
Electrode pitch~17.5 micrometers
Electrode diameter~5 micrometers
CoatingPlatinum black / TiN
Spike capture band100 Hz to 10 kHz
Impedanceunder 50 kOhm at 1 kHz

Keeping the tissue alive on the chip

Without perfusion an organoid suffocates in hours. Commercial chips favor cyclic olefin copolymer over the more common PDMS, because PDMS soaks up small-molecule drugs and would corrupt a screen, while COC is chemically inert; a thin gas-permeable window over the chamber restores the oxygen exchange that the inert plastic blocks. Indium tin oxide micro-heaters, transparent so imaging still works, hold the chamber at 37 degrees within a tenth of a degree under PID control.23

Why latency is the real spec

To teach tissue, feedback has to land inside the synaptic plasticity window, under about ten milliseconds. That rules out a normal operating system in the loop. The fast path detects a spike, sorts it by template match in around 100 microseconds, and routes the stimulus straight from the FPGA back to the analog ports, closing the loop in roughly 1.2 milliseconds. That number, not the channel count, is what determines whether a culture can learn in real time.

Signal acquisition and feedback pipeline A left-to-right chain of processing stages from the electrode array through amplification, digitization, spike sorting and decoding, then back to the stimulator. Spike threshold detect Sort template match FPGA decide Stimulate analog ports ~1.2 ms round trip
The real-time hardware loop. A spike is detected and template-matched on the FPGA, decoded, and routed straight back to the stimulator, bypassing the host operating system to close the loop in about 1.2 milliseconds.

How that loop is used to train living tissue is the subject of the biocomputing primer.

Frequently asked questions

What is an organoid-on-a-chip?

A microfluidic device that holds a living organoid on a microelectrode array, keeping it alive with perfusion and environmental control while electronics record and stimulate it.

Why coat electrodes with platinum black or titanium nitride?

Their fractal texture multiplies effective surface area, dropping interface impedance at 1 kHz from megaohms to under 50 kilohms, which is what makes microvolt neural signals audible.

Why does closed-loop latency matter so much?

Because feedback must reach the tissue within the roughly 10-millisecond plasticity window to shape learning. Specialized hardware closes the loop in about 1.2 milliseconds by bypassing the operating system.

Why use COC instead of PDMS for the chip?

PDMS absorbs small-molecule drugs, which corrupts screening results. COC is chemically inert, so a gas-permeable window is added to restore the oxygen exchange the inert plastic blocks.

References

  1. Mueller J, et al. High-resolution CMOS MEA platform to study neurons at subcellular, cellular, and network levels. Lab on a Chip. 2015;15(13):2767-2780. doi:10.1039/C5LC00133A. Accessed 2026-06-12.
  2. Bhatia SN, Ingber DE. Microfluidic organs-on-chips. Nature Biotechnology. 2014;32(8):760-772. doi:10.1038/nbt.2989. Accessed 2026-06-12.