chip_factory#

class ChipFactory(device_str: str)[source]#

Factory class to access config builder and other device specific methods

Parameters

device_str

device_id: int#
device_name: str#
events_to_raster(events: List, dt: float = 0.001, shape: Optional[Tuple] = None) torch.Tensor[source]#

Convert events from DynapcnnNetworks to spike raster

Parameters
  • events (List[Spike]) – A list of events that will be streamed to the device

  • dt (float) – Length of each time step for rasterization

  • shape (Optional[Tuple]) – Shape of the raster to be produced, excluding the time dimension. (Channel, Height, Width) If this is not specified, the shape is inferred based on the max values found in the events.

Returns

raster (torch.Tensor) – A 4 dimensional tensor of spike events with the dimensions [Time, Channel, Height, Width]

get_config_builder() sinabs.backend.dynapcnn.config_builder.ConfigBuilder[source]#
raster_to_events(raster: torch.Tensor, layer, dt=0.001, truncate: bool = False, delay_factor: float = 0) List[source]#

Convert spike raster to events for DynapcnnNetworks

Parameters
  • raster (torch.Tensor) – A 4 dimensional tensor of spike events with the dimensions [Time, Channel, Height, Width]

  • layer (int) – The index of the layer to route the events to

  • dt (float) – Length of time step of the raster in seconds

  • truncate (bool) – (default = False) Limit time-bins with more than one spikes to one spike.

  • delay_factor (float) – (default = 0) Start simulation from this time. (in seconds)

Returns

events (List[Spike]) – A list of events that will be streamed to the device

supported_devices = {'dynapcnndevkit': <class 'sinabs.backend.dynapcnn.chips.dynapcnn.DynapcnnConfigBuilder'>, 'speck2b': <class 'sinabs.backend.dynapcnn.chips.speck2b.Speck2BConfigBuilder'>, 'speck2btiny': <class 'sinabs.backend.dynapcnn.chips.speck2b.Speck2BConfigBuilder'>, 'speck2cmini': <class 'sinabs.backend.dynapcnn.chips.speck2cmini.Speck2CMiniConfigBuilder'>, 'speck2dmini': <class 'sinabs.backend.dynapcnn.chips.speck2dmini.Speck2DMiniConfigBuilder'>, 'speck2e': <class 'sinabs.backend.dynapcnn.chips.speck2e.Speck2EConfigBuilder'>, 'speck2edevkit': <class 'sinabs.backend.dynapcnn.chips.speck2e.Speck2EConfigBuilder'>}#
xytp_to_events(xytp: numpy.ndarray, layer, reset_timestamps, delay_factor: float = 0) List[source]#

Convert series of spikes in a structured array (eg. from aermanager) to events for DynaapcnnDevKit

Parameters
  • xytp (torch.Tensor) – A numpy structured array with columns x, y, timestamp, polarity

  • layer (int) – The index of the layer to route the events to

  • reset_timestamps (Boolean) – If set to True, timestamps will be aligned to start from 0

  • delay_factor (float) – (default = 0) Start simulation from this time. (in seconds)

Returns

events (List[Spike]) – A list of events that will be streamed to the device