High-density recording solutions
Radiens provides high-density recording solutions for Neuropixels, SiNAPS, and NeuroNexus high-channel-count probes.
HIGH-DENSITY RECORDING
Working definition
High-density recording places electrode sites close enough, and in enough number, that a single neuron lands on several sites at once. That redundancy is what lets a sorter tell one unit from its neighbours, and it is what turns a recording into a map of a circuit rather than a handful of channels. It sets up the two practical problems this page is about: capturing clean signal across every site, and handling the data and the sorting once the channel count is high.
Radiens supports the leading high-density silicon probes
System and workflow considerations
These are the key requirements for high-density recording in Radiens.
What every high-density workflow needs, whatever the probe:
- ✓Local-first custody — dense recordings stay on your own storage; nothing is uploaded.
- ✓A native probe model for your electrode — NeuroNexus, SiNAPS, and Neuropixels each ship with one; MEAs and any other geometry via the custom model builder (free).
- ✓Signal quality visible on every site, so a failing shank or connector shows before you trust the session.
- ✓Open formats in and out — reads the major acquisition systems natively; NWB read and write.
- ✓Provenance on every result, so a high-density analysis re-derives itself.
Getting clean signal off a dense probe, at the rig:
- ✓Native NeuroNexus DAQ acquisition, free — XDAQ at 1,024+ channels, and SmartBox.
- ✓Native Neuropixels and SiNAPS acquisition at Pro.
- ✓Live signal-quality and spectral monitoring across every channel, with Vantage™ at the rig.
- ✓Impedance on the bench and a live probe heat map, so hardware trouble is caught before the recording, not after.
Turning a dense recording into something ready to analyze:
- ✓Read most vendor formats with no conversion step; curate, resample, filter, and remap.
- ✓Quality metrics across every channel to triage a dense recording before you rely on it.
- ✓Bad-site detection from the redundancy pattern across neighbouring sites.
- ✓Export to NWB and open formats.
Making results from a high-density dataset:
- ✓Automated spike sorting, free; probe-specific models, manual curation, and cluster-quality metrics at Pro.
- ✓Field and LFP characterization — spectra, events, phase — over the same data.
- ✓On-probe and target-centered spatial views that keep every unit on its site.
- ✓Cross-session unit tracking, and full scripting through RadiensPy and the Radiens CLI.
- ✓On-demand, cluster-scale sorting is on the roadmap (Radiens Compute).
SiNAPS probe
Site pitch decides what you can resolve: whether one neuron lands on several sites at once, because that redundancy is what lets a sorter separate it from its neighbour. Channel count decides how much of the circuit you cover at that pitch — and what the session costs you in data, storage and sorting time. A 1,024-site probe at coarse pitch is a wide survey. A 256-site probe at fine pitch is a careful look at one place. Neither is denser in the sense that matters until you have said which of the two problems you have.
What a channel count commits you to
Data rate is channels times sample rate times sample width, and it does not care how interesting the experiment is. Per eight-hour session, at 16-bit stored width:
64 ch at 30 kHz
3.8 MB/s — about 111 GB per session.
SiNAPS 256 at 20 kHz
10.2 MB/s — about 295 GB per session.
384 ch at 30 kHz
23.0 MB/s — about 664 GB per session.
SiNAPS 1,024 at 20 kHz
41.0 MB/s — about 1.18 TB per session.
NEURONEXUS SOLUTIONS
Budget the disks with the probes
Ten animals at 1,024 channels for two hours each is roughly 3 TB before anything is analysed. A dense cohort is a storage project before it is an analysis project, and sizing follows throughput — channels times sample rate — rather than channel count alone.
Check three things before you trust a session
A dense probe fails gradually. A handful of poor sites is normal; a shank or a contiguous block behaving differently from its neighbours is not — and because the sites are redundant by design, the pattern across neighbours tells you which you have.
1 — Impedance, on the bench
Against each site's own baseline rather than a nominal figure. Residue from cleaning is the common cause, and it is reversible.
2 — The probe heat map, live
Near-uniform is what you want. Structure that matches the probe's own geometry — a shank, a column, a connector's worth of channels — points at hardware, not tissue.
3 — Raw traces, briefly
On a few sites spread across the array. Thirty seconds here saves a session you would otherwise finish and distrust.
NEUROPIXELS SOLUTIONS
Native support, from acquisition to analysis.
Record directly from Neuropixels and SiNAPS, drive 1,024+ channels through XDAQ, and read high-density data from any major system — then see and sort it at scale.
Native acquisition — Neuropixels & SiNAPS
Native Neuropixels and SiNAPS acquisition (Pro). NeuroNexus DAQ acquisition — XDAQ at 1,024+ channels, and SmartBox — is free at every tier.
Any electrode, natively modeled
Native probe models for NeuroNexus probes and both CMOS active-probe families; MEAs and any other electrode via the custom model builder (free).
See and sort at scale
Live signal-quality and spectral monitoring across every channel; probe-specific sorting and cluster-quality metrics (Pro).
What you get, by tier
Read and model high density for free; add native acquisition and scale at Pro.
- ✓Standard (free) — native NeuroNexus DAQ acquisition (XDAQ at 1,024+ channels, SmartBox), native read of any high-density recording, cross-vendor support, the model builder, and live monitoring.
- ✓Pro — native Neuropixels and SiNAPS acquisition, probe-specific sorting, and the any-DAQ live monitor.
- ✓On the roadmap — on-demand cluster-scale sorting and compute (Radiens Compute).
Built for scale
Local-first at any size
Dense recordings stay on your own storage — nothing uploaded, whatever the channel count.
Hardware-agnostic
Monitor any data stream in real time; read most vendor formats with no conversion step.
Quality you can see
Signal-quality metrics across every channel, so you triage a dense recording before you trust it.
Provenance at scale
Every result carries an ID and provenance, so a high-density analysis re-derives itself.
Frequently asked questions
How much density do I actually need?
Work back from what you are trying to tell apart. If two neurons must be separated, you need sites closer together than the distance over which their waveforms differ — that is a pitch question, and roughly 30 µm is where single-unit isolation becomes routine. If you need to see more of a circuit rather than resolve more finely within it, that is a count-and-shank question and pitch can stay where it is. Buying count when you needed pitch is the common and expensive mistake.
Can I keep the acquisition system I already own?
Usually, yes. Radiens reads most vendor formats with no conversion step and tails a file live while it is being written, so the analysis software and the acquisition hardware are separable decisions. A lab that has already bought a working acquisition system rarely needs to replace it to change what it can get out of the recordings.
What actually breaks first at high channel counts?
Not acquisition — storage and interpretation. A 1,024-channel probe at 20 kHz writes about 147 GB an hour, so a cohort becomes a storage project quickly, and sorting parameters become part of the method rather than a setting, because at that scale two threshold choices differ by thousands of units. Both are planning problems, which is why they are worth naming before the first surgery.
Does Radiens support Neuropixels?
Yes — native Neuropixels acquisition (Pro) and native read of Neuropixels/SpikeGLX data (free), with a native probe model for spatially-registered views and probe-specific spike sorting.
Does Radiens support SiNAPS probes?
Yes — native SiNAPS control for acquisition (Pro), plus the native probe model for on-probe visualization and probe-specific sorting.
How many channels can Radiens record?
1,024+ channels through XDAQ, with live monitoring and signal-quality readout across every channel.
Can Radiens read high-density data recorded on other systems?
Yes — cross-vendor probe and DAQ support reads high-density files from most major systems (Intan, SpikeGLX, Open Ephys, TDT, Plexon, NWB) with no conversion step.
What about a probe Radiens does not model natively?
Build it: the custom electrode model builder (free) creates a native model for MEAs and any other electrode, so on-probe views and coordinates work for your geometry.
Does high-density recording require the cloud?
No — Radiens is local-first. Dense recordings run on your own workstation against your own storage; nothing is uploaded.