See the protobuf definitions in the synapse-api repo for the most complete reference.

Electrical Broadband

Outputs broadband electrophysiology data from electrodes as a collection of channels with the same data rate. Electrical Broadband nodes are often paired with a Spectral Filter node followed by a Spike Detect node to downsample the signal to a more manageable size for streaming purposes.

Electrical Stimulation

Accepts input data to stimulate through electrodes.

Optical Broadband

Outputs broadband optical data from photodiodes as a collection of channels with the same data rate.

Optical Stimulation

Accepts input data to stimulate through LEDs.

Spectral Filter

Accepts an input and applies a basic frequency-domain filter.

Spike Detect

Accepts broadband neural data and detects action potentials (neural spikes) using a spike thresholding algorithm. Outputs binned spike data, where bin size (in milliseconds) is set using bin_size_ms.

Synapse Application

The application node is used to implement user-defined algorithms or models and accepts input from other nodes in the signal chain. Data is output using Taps API and is application-specific. See the Synapse App SDK for more information about developing your own Synapse App.

Disk Writer

The Disk Writer node saves broadband frame data to on-device storage. Data is saved in an HDF5 file with the .h5 extension. The Disk Writer node can ingest data from any node outputting data using the broadband frame format. Data attributes are preserved as they were output. For example, a Disk Writer node connected to the output of Electrical Broadband saves unfiltered electrical broadband data, whereas one connected to a Spectral Filter saves filtered data. A Disk Writer should be placed at the end of a signal chain or on a branch from an intermediate node.

Multiple Disk Writer nodes can operate simultaneously to save output from more than one node. Each Disk Writer creates a separate data file and should have a unique name. The HDF5 output facilitates interoperability with the Neurodata Without Borders format.

HDF5 Format

The Disk Writer creates a recording directory under data/disk_writer on the selected Storage Device:

/data/disk_writer/<recording-name>/
├── config.json
└── <recording-name>.h5

config.json contains the complete signal-chain configuration. If the requested directory or file already exists, the device adds a numeric suffix. Stopping and restarting the same configured Disk Writer creates a new .h5 file rather than appending to the previous file.

The HDF5 layout is inspired by NWB organization, but it is a Synapse recording format rather than a complete NWB file.

Dataset layout

All datasets are one-dimensional, chunked, and extendable.

Frame and channel order

ElectricalSeries is intentionally stored as a flat one-dimensional array, so the file itself has no row-major or column-major setting. The writer appends complete frames in order:

frame 0: channel 0, channel 1, ... channel C-1
frame 1: channel 0, channel 1, ... channel C-1
...

Conceptually, the data is frame-major: time advances by row and channel position advances by column. For frames with a constant width, reshape the flat array to (number_of_frames, values_per_frame) using C/row-major order:

import h5py
import numpy as np

with h5py.File("recording.h5", "r") as file:
    counts = file["/acquisition/ElectricalSeries"][:]
    timestamps_ns = file["/acquisition/timestamp_ns"][:]
    sequence_numbers = file["/acquisition/sequence_number"][:]

    frame_count = len(timestamps_ns)
    if frame_count == 0 or counts.size % frame_count != 0:
        raise ValueError("Recording does not contain constant-width frames")

    values_per_frame = counts.size // frame_count
    frames = counts.reshape((frame_count, values_per_frame), order="C")
    microvolts = frames.astype(np.float32) * file.attrs["lsb_uv"]

frames[i, j] is value j from frame i. A frame normally holds one sample per enabled channel, all electrodes. A source may also place non-electrode values, such as GPIO samples, in the same frame; the channel-range datasets identify which columns are which. channel_range_type holds the numeric ChannelType (0 electrode, 1 GPIO), channel_range_count its width, and channel_range_start its absolute offset into ElectricalSeries. Every frame contributes its own entries, so read the first frame's to learn a constant-width layout.

Scale only the electrode columns by lsb_uv; GPIO values are digital states. When the channel-range datasets are empty, every value is an electrode sample.

File attributes

To convert an ADC count to microvolts:

voltage_uv = adc_count * lsb_uv

The timestamps, sequence numbers, and data rows have the same frame order. Compare adjacent sequence numbers to detect dropped frames, and use timestamp_ns rather than array position when exact timing matters.