Neuropixels Probe

A CMOS chip thinner than a human hair that slides silently into living brain tissue and simultaneously listens to hundreds of individual neurons — reading the electrical whispers of thought at single-cell, single-spike resolution.

The Neuropixels probe records the electrical activity of individual neurons — action potentials ('spikes') lasting less than 1 ms — from hundreds of cells simultaneously, across multiple layers or regions of the brain. The probe is inserted into living brain tissue; its 960 titanium-nitride (TiN) electrode sites detect the extracellular voltage fluctuations caused by nearby neurons firing. Any 384 of those sites can be selected and digitised simultaneously in the action-potential (AP) band at 30 kHz with approximately 5–7 μV RMS input-referred noise. The same channels also record the local field potential (LFP) at 2.5 kHz. On-chip analogue amplification, analogue-to-digital conversion, and digital multiplexing mean that all 384 channels of data exit the brain on a single thin cable, rather than 384 individual wires. This allows neuroscientists to follow dozens to hundreds of identified single neurons across multiple brain structures in freely moving mice, rats, and non-human primates during complex behaviours — a scale impossible with any prior electrode technology.

The probe shank is a silicon needle 10 mm long and 70 × 20 μm in cross-section — slender enough to displace minimal brain tissue and cause only transient mechanical injury. Along the shank, 960 low-impedance TiN electrode sites (each ~12 × 12 μm, impedance ~150 kΩ at 1 kHz) are tiled at 20 μm vertical pitch in a staggered two-column pattern. The probe is fabricated in a 130 nm CMOS process, with all analogue front-end circuitry (amplifiers, filters, ADCs, digital multiplexers) integrated on the probe base — a square die roughly 9 mm × 9 mm glued to the top of the shank. Each recording channel passes the extracellular signal through a high-pass filter (corner at 300 Hz for AP, 0.5 Hz for LFP), an amplifier with gain ~500, and a 10-bit ADC running at 30 kHz (AP) or 2.5 kHz (LFP). A digital serialiser combines all channel data into a 163.8 Mb/s data stream sent over a thin coaxial cable to a PXIe data acquisition module. Total power dissipation is approximately 15 mW — critically low enough that shank self-heating remains below 1°C, preserving the thermal environment of the recorded neurons. User-programmable switches under each electrode site allow any 384 of the 960 sites to be selected without moving the probe, enabling systematic mapping of the neuron population along the full shank depth.

In plain terms

Imagine you want to listen to individual musicians in an orchestra playing in a dark room. An old microphone could only hear the whole audience at once. The Neuropixels probe is like threading a strand of hair through the room, with 960 tiny microphones along its length — each one so sensitive it can pick out a single violin, even while the whole orchestra is playing. That's how it hears individual neurons firing inside a living brain.

Also called: Neuropixels 1.0, Neuropixels 2.0, high-density silicon probe, CMOS neural probe, NPx probe.

Why it matters

Where it lives

Key numbers

Surprising facts

Open questions

A short history

Before Neuropixels, the practical limit for simultaneous single-unit recording in an awake animal was roughly 100–150 channels per session, achieved by manually assembled tetrode bundles or the Utah array. The idea that CMOS technology used in consumer imaging sensors could be adapted to neural recording was explored in academic prototypes through the 2000s (e.g. Michigan probe designs at the University of Michigan), but integration density and noise performance remained inadequate. In 2011, imec (the Belgian microelectronics research centre) initiated a collaboration with HHMI Janelia Research Campus, UCL, and the Allen Institute to design a ground-up neural recording ASIC using imec's 130 nm CMOS process. The result — the Neuropixels 1.0 probe — was described by Jun, Steinmetz, Bhatt, Bhatt, Bhatt and 58 co-authors in Nature in November 2017. Within two years, more than 400 laboratories worldwide had adopted the probe. Neuropixels 2.0, reported by Steinmetz, Aydin, Bhatt and colleagues in Science in April 2021, extended the design to 5,120 electrode sites across four shanks, adding drift-correction algorithms that enabled stable single-unit tracking across months of chronic implantation. Neuropixels probes have since enabled the largest single-session neural population recordings in rodent history, fundamentally changing the questions neuroscientists can ask about population coding, learning, and brain-wide coordination.

Sources

Related