High-Density Recording: A Planning Guide
How to choose a recording density, what it costs in data and computation, and how to keep a dense recording interpretable.
7 min read
Updated November 13, 2026
NeuroNexus · neuronexus.com · Resources
High-density recording is spatial oversampling. The sites are close enough together that one neuron is recorded on several of them at once, and a spike sorter separates neurons by how differently they appear across neighboring sites. A site spacing under about 50 µm is generally needed to record a neuron with adequate signal-to-noise on several sites (Drake et al. (1988) ↗). The planning question is therefore whether the sites are closer together than the sources to be separated.
Research use only. NeuroNexus products and the procedures described here are for non-clinical research use, and must not be used in human or veterinary medical procedures.
This guide covers the choice of density, its cost in data and computation, the quality checks to make before a session, and the practices that keep a dense study reproducible.
What sets density#
Two parameters determine density, and they answer different questions.
Site pitch is the distance between neighboring electrode sites. It determines whether a neuron is recorded on more than one site, and therefore how well units can be isolated. Single units are generally distinguishable within about 50 to 100 µm of a site (Quian Quiroga & Panzeri (2009) ↗), so a pitch well under that distance places several sites within range of each neuron.
Channel count is how much tissue is covered at that pitch. It determines how much of the circuit is sampled, and it sets the data rate, the storage requirement, and the sorting time.
A design with 1,024 sites at a coarse pitch surveys a wide region. A design with 256 sites at a fine pitch samples one location in detail. Which is the better choice depends on whether the experiment needs coverage or isolation.
| Family | Sites | Pitch | Active length | Shanks |
|---|---|---|---|---|
| SiNAPS 1S-256 | 256 | 30 µm | 3,865 µm | 1 |
| SiNAPS 4S-1024 | 1,024 | 30 µm | 3,865 µm per shank | 4 at 560 µm |
| SiNAPS 8S-1024 | 1,024 | 30 µm | 1,925 µm per shank | 8 at 300 µm |
| Passive silicon | 4–256 | 20–500 µm | by design | 1–16 |
The SiNAPS pitch and active length follow Corticale's published specifications. The passive range is the span of the catalog, and the value for a specific design is on its datasheet.
What density costs#
The data rate is the channel count multiplied by the sample rate and the stored sample width. The resulting data volume is best planned before the first session.
| Configuration | Rate | Per hour | Per 8-hour session |
|---|---|---|---|
| 64 ch at 30 kHz | 3.8 MB/s | 14 GB | 111 GB |
| SiNAPS 256 at 20 kHz | 10.2 MB/s | 37 GB | 295 GB |
| 384 ch at 30 kHz | 23.0 MB/s | 83 GB | 664 GB |
| SiNAPS 1,024 at 20 kHz | 41.0 MB/s | 147 GB | 1.18 TB |
The table assumes a 16-bit stored width. Three consequences are worth planning for.
- A dense cohort needs a storage plan. Ten animals at 1,024 channels for two hours each is about 3 TB of raw data. Storage should be budgeted with the probes.
- Computer sizing follows throughput. The requirement is set by the product of channel count and sample rate, as described in Radiens System Requirements.
- Streaming from disk removes the memory limit. Radiens reads a recording from disk and does not load it into memory, so session length is limited by storage and not by RAM.
Read the quality before you trust the session#
A dense probe generally degrades gradually. A few poor sites are normal. A shank or a contiguous block of sites that behaves differently from its neighbors usually indicates a hardware fault and not a quiet region of tissue. Because neighboring sites record overlapping signals, the pattern across neighbors helps to tell the two apart.
Three checks are recommended before the experiment begins, in this order:
- Impedance, on the bench, compared with the baseline of each site and not with a nominal value. See Impedance ↗.
- The probe heat map, live. A near-uniform map is expected. Structure in the map that matches the geometry of the probe, such as a shank, a column, or the channels of one connector, generally points to the hardware and not to the tissue. See Signal Metrics ↗.
- The raw traces, briefly, on a few sites spread across the array. See Monitor ↗.
For an active-pixel probe, the calibration step comes first and is required. It is described in the SiNAPS Manual.
Sorting at scale#
Oversampling is what makes sorting work. It also makes sorting computationally expensive, because every added site adds to the data the sorter must process. Two practical points follow.
A sort is a derived dataset. In Radiens, a sort is written as a separate, linked dataset beside the raw recording, and the recording is never modified. A sort can therefore be discarded and repeated without affecting the session.
Sort parameters are part of the method. Spike sorting involves judgment, and the detection threshold and other settings can change the number of units reported, particularly at 1,024 or more channels. The preparation can be saved as a protocol and applied across the cohort so that the animals are comparable, using Curate ↗.
Keeping a dense study reproducible#
A dense study has the same reproducibility requirements as any recording study, at a larger scale. Three practices help.
- Identical preparation across the cohort. One saved protocol is applied to every animal, in place of a filter chain rebuilt for each session.
- Derived results stay linked to their source. Sorts, metrics, and curated datasets each refer to the recording they came from, so a figure can be traced to a session.
- Exclusions are recorded with the reason. Dropped channels are part of the methods. A dense recording with undocumented exclusions is difficult to defend and to repeat.
Choosing an acquisition path#
The probe and the software are separate decisions.
| If your lab… | Recommended path |
|---|---|
| Runs SiNAPS only | Smartbox Pro with Radiens, or the acquisition system of the probe manufacturer. Both record the probe. |
| Runs SiNAPS alongside passive arrays, grids, or Neuropixels | Radiens, so that every recording is in one workspace with one probe-geometry model |
| Needs stimulation as well as recording | XDAQ with an SR-series headstage. Active-pixel probes record only. |
| Already owns another acquisition system | Radiens reads its data, so the software choice does not depend on the acquisition hardware |
A lab that already owns an acquisition system does not need to replace it to use Radiens with its recordings.
Related documentation#
- SiNAPS: Active High-Density Microelectrodes
- SiNAPS Manual
- Neuropixels, SiNAPS and Silicon Probes: How to Choose
- Silicon Probes and Thin-Film Grids: An Introduction
- Radiens System Requirements
- Noise Reduction and Troubleshooting
- Glossary: Site pitch · Site density · Spike sorting · Active-pixel probe · SiNAPS probe · Data volume
Citations#
- Drake, K. L., Wise, K. D., Farraye, J., Anderson, D. J., and BeMent, S. L. (1988). Performance of planar multisite microprobes in recording extracellular single-unit intracortical activity. IEEE Transactions on Biomedical Engineering 35(9), 719–732. 10.1109/10.7273 ↗
- Quian Quiroga, R., and Panzeri, S. (2009). Extracting information from neuronal populations: information theory and decoding approaches. Nature Reviews Neuroscience 10(3), 173–185. 10.1038/nrn2578 ↗
