Tutorials

• Overview of DYNAP-CNN and SPECK devlopment kit
  • DYNAP-CNN and SPECK
• Role of sinabs-dynapcnn Library
• Chip Resources
  • Key Features
    • Event Pre-processing Layer
    • DYNAP-CNN Layer
      • Feature
      • Internal Execution Order
      • Async event-driven feature
      • Memory Constraints and Network Sizing
      • Leak operation
      • Congestion banlancer
      • Decimator
    • Readout Layer
    • Internal Slow clock
• Known dev-kit Bugs
• FAQs
  • 1. What network structure can I define?
  • 2. What kinds of neural network operation I can use in Speck/DYNAP-CNN?
  • 3. What execution order should I notice when I am implementing a sequential structure?
    • Ex1. Bad Case: Cascaded convlution
    • Ex2. Bad Case: None sequential
    • Ex3. Bad Case: Use unsupport operation
    • Ex3. Good Case: Use unsupport operation
  • 4. What is the limination for network sizing?
• From a CNN model to a DYNAP-CNN DevKit
  • Import libraries
  • CNN definition
  • Model evaluation
  • Convert to Spiking CNN
  • DYNAP-CNN compatible network
  • Porting model to DYNAP-CNN DevKit
  • Inspecting memory
  • Sending a receiving spikes
• Visualize DVS Input
  • 1. Open devkit
  • 2. Build samna graph
    • 2.1 Get the source node of the devkit
    • 2.2 Init samna graph and VizEventStreamer node
  • 3. Create a sub-process for GUI visualization
    • 3.1 init samna node and get endpoints related with the visualizer
    • 3.2 create a sub-process which runs the GUI visualizer
    • 3.3 setup the visualizer
  • 4. Launch the devkit and start visualization
    • 4.1 start samna graph
    • 4.2 create devkit config and apply to devkit
  • 5. Stop the devkit and visualization
• Power Monitoring
  • 1. Power consumption of the devkit
  • 2. Monitor the idle power
  • 3. Monitor the dynamic power
    • 3.1 Create a simple SNN
    • 3.2 Create a devkit config based on the SNN
    • 3.3 Construct a samna graph for power monitoring & visualization
    • 3.4 Setup the GUI visualizer of samna
    • 3.5 Apply devkit config and launch the devit
    • 3.6 Start to monitor dynamic power
• Using Readout Layer
  • What is readout layer
  • 1. Create a 1-layer CNN which can be deployed to the devkit
  • 2. Two different types of output events from readout layer
  • 3. Prepare for deployment
    • Notice
    • 3.1 Create devkit configuration
    • 3.2 Construct samna graph
    • 3.3 Set up visualizer
    • 3.4 Set up readout layer’s driven slow-clock
  • 4. Start readout from devkit
    • 4.1 Create fake input for devkit
    • 4.2 Read the ReadoutPinValue
    • 4.3 Read the ReadoutValue
    • 4.4 Check the output from DynapCNN Layer
• Spike Count Visualization
  • 1. Create a 1-layer CNN which can be deployed to the devkit
  • 2. Prepare for deployment
    • Notice
    • 2.1 Create devkit configuration
    • 2.2 Construct samna graph
    • 2.3 Set up visualizer
  • 3. Start Visualization
    • 3.1 Create fake input for devkit
    • 3.2 Write into the fake input and visualize the spike count of output
    • 3.3 Check the output from DynapCNN Layer