Creating ‘images’ of neural activity with the Neuropixels digital probe

By Carolina Mora Lopez, R&D Team Leader Circuits for Neural Interfaces, imec

The neural signatures for movement, touch, and thought processes are hidden in the activity of large neural populations. Each action and process is coordinated by a complex interplay between different cells from different regions — deep and superficial. The most common technologies to measure the neural activity — and, hence, the neural signatures — are electrical and optical signals.

Neural electrophysiology probes are used to map complex brain activity but have faced challenges in the areas of spatial resolution, temporal resolution, and size.

Typical extracellular probes record single neuron activity with high temporal resolution, with sub-millisecond precision, but only reach a few tens of neurons. Calcium imaging, on the other hand, covers more brain area but does not reach the same sub-millisecond resolution of recording because it is an indirect measure of neural activity.

The ideal tool for probing neural populations combines the sub-millisecond temporal resolution with the spatial resolution of imaging while, at the same time, maintaining a small footprint. The small footprint is critical for both the implantable part, where brain damage is to be avoided, but also for the external part, as extensive cabling impedes recording in freely moving animals.

The Neuropixels probe , designed and developed at imec ,¹ a research and innovation hub in nanoelectronics and digital technologies, meets these demands by offering a high-density integrated digital silicon CMOS neural probe packing 960 recording sites on a 70 × 24-µm shank, with minimal cabling.

Fig. 1: The Neuropixels chip architecture features a shank with electrodes and switches and a base with separate programmable gain amplifiers in the AP an LFP band for 384 channels.

One of the main challenges in probe design is to place as many electrodes as possible on a cross-sectional area that is as small as possible. Such a large and dense electrode array provides neuroscientists with a tool to interrogate large neural populations across brain regions while, at the same time, causing minimal damage to the brain tissue.

The narrow (70 µm) and thin (24 µm) implantable shank of the Neuropixels probes is tiled with 960 electrodes in a two-row checkerboard pattern. Electrodes are 144 µm² in size with a pitch of 20 µm covering a probe shank length of 10 mm. The site count on a single shank is an order of magnitude larger than that of existing single-shank silicon probes.

Fig. 2: The probe shank packs 960 electrodes, of which 384 can be addressed individually and simultaneously.

Dense electrode arrays enable scientists to distinguish and isolate signals from closely spaced neurons and detect “drift” due to shifts of brain tissue. These small dimensions are enabled with a custom 130-nm CMOS fabrication process. The electrodes are fabricated from a porous, low-impedance TiN material that is compatible with CMOS processing steps. The use of high-throughput silicon fabrication processes ensures the manufacturing of Neuropixels probes at low cost and in large volumes.

Fig. 3: Details of the probe fabrication. (a) Single neural probe device after post-CMOS processing. (b) Details of the probe neck and base. (c) Details of the shank tip and electrode arrangement. (d) Details of the TiN electrodes and via grids. (e) Diced probes on wafer.

The fundamental limit to the number of electrodes that can be simultaneously recorded is the number of wires that can be placed on the shank. For Neuropixels, a maximum of 384 interconnect lines are routed between the electrodes and the amplifiers in the base. This means that 384 electrodes selected from the total number of recording sites can be addressed at the same time. A local switch matrix provides the user with the flexibility to select the preferred sites so that the brain area of choice can be recorded from. After selection of an electrode, the memory bit placed underneath is enabled or disabled.

The shank is fabricated on the same silicon substrate as the 9 × 5-mm base, yielding a lightweight (250 mg) probe that is suited for long-term monitoring of neural activity in moving animals. The largest part of the base area is allocated to the 384 recording channels. These channels allow recording in the two electrophysiological relevant frequency bands for neural signals: local field potentials (LFP) (0.5 Hz –1 kHz) and action potentials (AP) (0.3–10 kHz).

Because APs typically have amplitudes in the order of a few µV, while LFPs can go up to several mV, it is imperative to provide low-noise amplification and high dynamic range to accurately record the neural signals in both frequency bands. In the Neuropixels base, the continuous data stream is, therefore, split into two bands that are separately amplified.

Additionally, the user can select the gain per recording channel to ensure optimal amplification for different signal amplitudes, as seen, for example, in different brain regions. The system with the low-impedance recording sites included achieves noise levels of ~5 µVrms in the AP band and ~20 µVrms in the LFP band, which is smaller than noise of biological origin.

The analog output from all channels is multiplexed and digitized on the base and subsequently transmitted over the probe flex cable to an interface board, or headstage. Because the voltage signals from the neurons are filtered, amplified, multiplexed, and digitized on the probe base itself instead of using external equipment — a so-called active digital design — no signal loss occurs when the signal leaves the probe. At the same time, the system with all electronics consumes very low power (

Fig. 4: The packaged Neuropixels system connects the probe with a flex PCB to the head stage. The data stream is then transmitted through an ultra-thin, lightweight cable to the PXIe acquisition module that communicates with the host computer.

The combination of a high channel count and a small cross-sectional area allows for simultaneous recording from hundreds of neurons in different brain regions, minimizing tissue damage. A Neuropixels probe displays neural activity in the brain as images, with the data from each recording site represented as a “pixel.”

The integrated circuitry ensures a small physical footprint of the probe, necessary to study behavior in awake animals. It highlights the potential of scalable semiconductor technology and high-performance electrode technology for tools to study brain-wide neural activity in behaving animals.

Note
[1] The Neuropixels project was supported by a grant from HHMI, the Allen Institute of Brain Science, the Wellcome Trust, and the Gatsby Foundation. Probes were designed, developed, and fabricated at imec in collaboration with HHMI Janelia Research Campus, Allen Institute for Brain Science, and University College London. For more information, go to https://www.neuropixels.org/.